IREN IREN NASDAQ

IREN's financing record moves through three phases. The first is a conventional growth-equity start: IPO net proceeds of about $215.3M (claim #14390) and a $71M NYDIG equipment facility (claim #14216). The second, through 2023, relies on equity sold in small increments. The B. Riley committed equity facility (claim #13623) recurs throughout the year at $25M (claims #13325, #12442, #12203), and the $300M at-market program follows it (claim #12188). The third phase, which began in December 2024, swaps incremental equity for successively larger convertibles: $440M (claim #5269), then $550M after an upsize from $450M (claims #28843, #28839), then $875M due 2031 (claim #21761). Each step is larger and closer to the last. The equipment-financing layer from the first phase did not survive intact. It went through receivership (claim #12447), litigation over collateral (claim #12270) and a settlement in August 2025 (claim #4429).

The convertibles share a consistent structure. Capped calls and prepaid forward repurchases accompany each deal (claims #5273, #5274, #28841, #28842), which suggests the company is deliberately managing conversion dilution rather than issuing opportunistically. The ATM layer, however, keeps growing. It was replaced by a $1B program (claim #5159), with $111M drawn by March 2025 (claim #31453). Over the same period, management repeatedly described project financing, GPU financing and debt workstreams as forthcoming (claims #31521, #31446, #29812). By mid-2025, the convertible notes were described as the only non-equity financing (claim #28352), and a $130M GPU purchase was funded from existing cash (claim #28341). The non-dilutive debt that was signaled has therefore been delayed, and the record does not show it executed.

The stated use of proceeds also requires explanation. In May 2025, management put the net funding requirement for the year at up to $250M (claim #29811). The two convertibles that followed raised well beyond that figure. The October deal's proceeds are allocated only to capped calls and general corporate purposes (claim #21764), while the disclosed Childress build is estimated at $1.4B (claim #31155). The record does not show how the raised capital maps to specific projects.

The record contains no insider transactions, and the claims supplied stop in October 2025, so later activity cannot be assessed here. A careful reader should watch three things. First, whether any GPU-backed or project-level debt actually closes. Second, the pace of drawdowns under the $1B ATM. Third, how share price compares with conversion terms such as the roughly $13.64 strike (claim #28857), since that determines whether the capped calls are enough to contain dilution.

IREN's financing record shows three distinct phases, each larger than the last. The first ran from the IPO, which produced $215.3M in net proceeds per claim #14390, through the NYDIG equipment facility of $71M per claim #14216. The second followed those equipment facilities into lender enforcement and a PwC receivership (claims #12447 and #12462). In that phase the company leaned on small committed equity lines with B. Riley and then a $300M ATM per claim #12185. The third, beginning in late 2024, is built on convertible notes: $440M per claim #5269, $550M closed per claim #28839, and $875M plus a $125M option per claims #21761 and #21762. Alongside these sits a $1B ATM per claim #5159. Each convertible tranche roughly doubled the prior one inside a ten-month span, so potential dilution is accelerating. The rise in initial conversion price from about $13.64 (claim #28857) to about $85.63 (claim #21758) means the latest tranche converts into proportionally fewer shares.

The mix reads as more deliberate than in 2023. Each convertible carried capped calls, and the first two also funded prepaid forward repurchases. That treatment absorbed about $118M of $425M net in December 2024 (claims #5270, #5273, #5274) and about $146M of $534.9M in June 2025 (claims #28840 to #28842). The October tranche dropped the prepaid forward and kept only a $49.6M capped call per claim #21756. That suggests the company now worries less about hedge-fund shorting pressure.

Stated plans and actual deployment diverge most on non-dilutive capital. Debt, project and GPU financing workstreams appear repeatedly as imminent (claims #31521, #31446, #29812, #31056). Yet as of claim #28352, convertibles were the only non-equity financing in place, and a $130M GPU purchase was funded from existing cash per claim #28341. These workstreams appear delayed. Meanwhile disclosed capital needs keep growing. They include an estimated $1.4B for Childress per claim #31155, Bitmain balances including $119.3M due October 2025 per claim #29846, and a 2025 net funding requirement of up to $250M per claim #29811. That last figure looks modest beside the $1.4B of convertible issuance that followed.

The record contains no insider transactions, so no read on insider behavior is possible. The October 2025 entries are truncated. A careful reader should watch for three things: whether any asset-level debt actually closes; how fast the $1B ATM is drawn against the $111M raised by claim #31453; and how the 2023 equipment-financing defaults, settled per claim #4429, affect lender appetite for future secured facilities.

IREN's financing record moves through three phases, and each one funds a different business. After the November 2021 IPO, which brought in net proceeds of $215.3M (claim #14390), the company relied first on asset-level debt. The $71M NYDIG equipment facility (claim #14216) later appears under lender enforcement with a PwC receiver appointed (claims #12447, #12462). It was then the subject of court rulings on collateral scope (claim #12270) and was settled only in August 2025 (claim #4429). That episode seems to have pushed the company away from secured equipment debt and toward equity-linked capital.

The second phase, from 2022 to 2024, relied on incremental equity. The B. Riley committed facility (claim #13623) is referenced repeatedly through 2023 at the $25M level (claims #13325, #12442, #12203). It gave way to a $300M ATM (claim #12188) and then a $1B ATM in January 2025 (claim #5159), of which $111M had been drawn by March 2025 (claim #31453). This looks more like continuous replenishment than a planned capital structure.

The third phase is a sequence of convertibles that keep getting larger. The first was $440M in December 2024 (claim #5269). Next came the June 2025 notes, launched at $450M (claim #28843), priced at $500M (claim #28849) and closed at $550M (claim #28839). An $875M tranche due 2031 followed in October 2025 (claim #21761). Each deal carried capped calls and, in the first two cases, prepaid forward repurchases. For example, the June notes allocated $53.8M to capped calls and $92.5M to the prepaid forward (claims #28841, #28842). The repeated use of these structures suggests a consistent effort to manage visible dilution, but each one also reduces the cash that is actually deployable.

The stated use of proceeds has shifted with the business. Large capex figures, such as the $1.4B Childress estimate (claim #31155) and Horizon 1 at $300–350M (claim #31447), sit alongside repeated references to GPU and project-debt workstreams (claims #31446, #29812, #31521). In June 2025, however, convertibles were still described as the only non-equity financing (claim #28352), and a $130M GPU purchase was funded from existing cash (claim #28341). The debt workstreams have been announced but have not yet been delivered.

A careful reader should watch three things. The first is whether non-dilutive project or GPU financing actually closes. The second is how heavily the $1B ATM is used between convertible deals. The third is how the stacked 2029 and 2031 maturities interact with the June notes' roughly $13.64 conversion price and $20.98 cap (claims #28857, #28858). The record contains no insider transactions, and the October 2025 use-of-proceeds detail is incomplete in the claims provided.

IREN's financing record shows two distinct phases. The first runs from the November 2021 IPO, which netted $215.3M per claim #14390, through 2023. In that period the company relied on small committed equity facilities with B. Riley, $25M per claim #13325, and on non-recourse equipment financing, $71M from NYDIG per claim #14216. That equipment debt ended in a receivership and lender enforcement per claims #12447 and #12462, followed by a court ruling and an August 2025 settlement per claims #12270 and #4429. After that episode, conventional asset-level debt largely leaves the record. Funding moves to at-the-market equity, first a $300M program per claim #12185 and then a $1B program per claim #5159, plus a sequence of convertible notes. Each convertible was upsized and each was larger than the last: $440M in December 2024 per claim #5269, $550M closed in June 2025 per claim #28839 against a $450M launch per claim #28843, and $875M in October 2025 per claim #21761. Taken together, this reads as a deliberate, repeatable template rather than ad hoc patching. A meaningful share of each raise went to dilution management rather than operations. In June 2025, capped calls took $53.8M and a prepaid forward took $92.5M out of $534.9M of net proceeds, per claims #28841, #28842 and #28840. Disclosed funding needs and the amounts actually raised require explanation side by side. In May 2025 the company cited up to $250M of net funding required for 2025 (claim #29811, not yet verifiable), yet it raised roughly $1.4B in convertible notes over the following five months. Separately, it described debt-financing workstreams as expected to execute in coming months (claim #29812). Weeks later, it stated that convertibles were its only non-equity financing (claim #28352), and the record shows no project or GPU facility closing afterward. The capital being mapped against the raises is large: $1.4B for Childress per claim #31155 and $300–350M for Horizon 1 per claim #31447. The record does not tie specific raises to specific projects, so deployment cannot be reconciled from this data. The record contains no insider transactions, and the October 2025 entries are truncated, so the terms of the latest notes are incomplete here. A careful reader should watch four things. First, whether the long-discussed project or GPU debt actually closes. Second, how much of the $1B ATM has been drawn. Third, whether the Bitmain balances due through October 2025, $119.3M per claim #29846 among others, were paid as scheduled. Fourth, how share count behaves relative to the June 2025 conversion price of roughly $13.64 and capped-call ceiling of $20.98, per claims #28857 and #28858.

The record shows a company that has changed funding regimes three times rather than drifting between them. After the IPO, which netted roughly $215.3M per claim #14390, IREN relied on non-recourse equipment debt. The $71M NYDIG facility, per claim #14216, ended in lender enforcement and a PwC receivership (claims #12447 and #12462). The company then leaned on a B. Riley committed equity line, the $100M facility per claim #13623 that is repeatedly referenced at $25M through 2023 (claims #13325 and #12203). It moved to a $300M ATM (claim #12185), later a $1B ATM (claim #5159). From December 2024 onward it has relied on convertible notes. That arc reads less as opportunistic patching than as a migration away from asset-level debt after it failed, though the 2023 equity-line phase looks like bridging.

The convertible program is the dominant dilution vector, and each deal grew between launch and close. December 2024 went from $300M to $440M (claims #5300 and #5269). June 2025 went from $450M to $550M (claims #28843 and #28839), alongside a withdrawn $600M entry (claim #4782) that requires explanation. October 2025 was $875M due 2031 (claim #21761). The first two deals paired capped calls with prepaid forwards to absorb hedge-fund shorting, $74M and $92.5M respectively (claims #5274 and #28842). The October deal drops the forward and directs the remainder to general corporate purposes (claim #21756), which loosens the link between capital raised and named projects. The jump in conversion price from about $13.64 (claim #28857) to about $85.63 (claim #21758) within four months means the June 2029 notes sit well beyond their $20.98 cap (claim #28858). The capped call no longer shields holders above that level.

Use of proceeds tracks only partially. In May 2025 management framed a net funding need of up to $250M for the year (claim #29811) and described debt workstreams expected "in coming months" (claim #29812). A month later it stated that convertibles were its only non-equity financing (claim #28352), and it then raised roughly $1.4B in notes. The Childress capex estimates of $592.5M and $605M for Phases 2 and 3 (claims #31362 and #31333) give context for that scale, but the record does not map deployment to those phases. Disclosed ATM usage is $111M against the $1B program (claim #31453).

The record contains no insider transactions and ends in October 2025. A sophisticated reader should watch four things. The first is whether the promised GPU and data-center debt financing (claims #31446 and #31056) materializes or remains discussed only as workstreams. The second is how much of the 2029 notes converts. The third is ATM drawdowns. The fourth is whether general-corporate proceeds are later attributed to specific builds.

IREN's capital history falls into three phases, each larger and more structured than the last. After the IPO, which delivered $215.3M in net proceeds (claim #14390), the company's first leverage experiment was the $71M NYDIG equipment facility (claim #14216). That facility later moved to lender enforcement with a PwC receiver appointed (claims #12447, #12462). A court affirmed that bitcoin was not collateral (claim #12270), and the matter reached a settlement only in August 2025 (claim #4429). That episode appears to have shaped what followed. Through 2023, financing relied on the B. Riley committed equity facility, drawn in repeated $25M tranches (claims #13325, #12442, #12369, #12203). It then moved to a $300M ATM (claim #12185) and a $1B ATM (claim #5159). This reads as opportunistic patching that gradually settled into a programmatic equity backstop.

The third phase is the convertible sequence: $440M in December 2024 (claim #5269), $550M in June 2025 (claim #28839), and $875M at 0.00% in October 2025 (claim #21752). Each was paired with capped calls, and the first two also carried prepaid forwards of roughly $74M and $92.5M (claims #5274, #28842). The dilution-management architecture is coherent and repeated. The June 2025 deal was upsized from the $450M initially described (claim #28843), and a $600M figure was withdrawn (claim #4782). The sharpest change is the conversion price, which moved from about $13.64 (claim #28857) to about $85.63 four months later (claim #21758). Equity revaluation, not lower issuance, is what has slowed dilution per dollar raised.

Use of proceeds tracks only partially. In April 2025 the company described a strategy of customer prepayments and debt, with the ATM as a backstop (claim #31056). It said in May that debt workstreams would execute "in coming months" (claim #29812). By late June it stated that convertibles were the only non-equity financing (claim #28352), and no executed debt facility appears in this record. Childress capex estimates of $1.4B overall (claim #31155) far exceed any single raise. A $130M GPU purchase was funded from existing cash (claim #28341), and the October notes direct most proceeds to general corporate purposes (claim #21756). The link between specific raises and specific projects is therefore loose.

There are three things to watch. The first is whether the debt workstreams ever close or are no longer discussed. The second is how the ATM capacity is actually drawn against the Bitmain balances due through October 2025 (claims #29845–#29847). The third is how conversion overhang compounds across three note series. The record contains no insider transactions, and the claim list is truncated after October 2025, so later events cannot be assessed.

IREN's financing record describes a company that moved through three distinct funding regimes rather than following a single plan. After the IPO, which produced $215.3M of net proceeds per claim #14390, the first non-equity instrument was the $71M NYDIG equipment facility (claim #14216). By mid-2023 that facility was under lender enforcement with a PwC receiver appointed (claims #12447, #12462). A court affirmed that Bitcoin was not collateral for those limited-recourse arrangements (claim #12270), and a settlement of the subsidiary defaults followed in August 2025 (claim #4429). That episode explains much of what came next. Through 2023 the company relied on small, repeatedly re-disclosed $25M committed equity facilities with B. Riley (claims #13325, #12203). It then stepped up to a $300M ATM (claim #12188) and a $1B ATM in January 2025 (claim #5159). The record does not show how much of either ATM was drawn, so the actual pace of equity dilution cannot be measured from this data.

The convertible phase is where the strategy becomes coherent. There were three offerings in roughly ten months: $440M in December 2024 (claim #5269), $550M in June 2025 (claim #28839) and $875M at 0.00% due 2031 (claim #21752). Each was upsized from its launch size, which together comes to about $1.87B of face value. The first two carried capped calls and prepaid forwards, the latter $92.5M in June per claim #28842. The October deal kept only a $49.6M capped call (claim #21756). The jump in initial conversion price from about $13.64 (claim #28857) to about $85.63 (claim #21758) means the 2029 notes sit far below the newer series, and their eventual settlement mechanics require attention.

Use of proceeds tracks only loosely. The May 2025 disclosures described debt financing workstreams expected to execute in coming months (claims #29812, #29825), alongside GPU financing for Prince George (claim #31055). By late June, convertibles were described as the only non-equity financing (claim #28352), and a $130M GPU purchase was funded from existing cash (claim #28341). Childress capital needs of $605M for Phase 3 and $1.4B overall (claims #31333, #31155) far exceed the stated up-to-$250M 2025 net funding requirement (claim #29811, not yet verifiable). The October proceeds were allocated largely to general corporate purposes.

The record contains no insider transactions, so no read on insider behavior is possible. A reader should next watch whether the promised debt facilities are ever executed or are no longer discussed. Other open items are the disclosed ATM utilization, settlement of the Bitmain balances due through October 2025 (claims #29844–#29847), and whether the 2029 notes settle in shares or cash.

IREN's financing record describes a company that moved through three distinct capital regimes rather than following a single plan. The first regime was the IPO, with net proceeds of $215.3M per claim 14390, followed by asset-level equipment debt such as the US$71M NYDIG facility per claim 14216. That second layer did not hold. By mid-2023 the equipment facilities were under lender enforcement with a PwC receiver appointed, per claims 12462 and 12447. A court then affirmed that Bitcoin was not collateral, per claim 12270, and a settlement of the subsidiary defaults is only recorded in August 2025, per claim 4429. The interim period relied on equity backstops: the B. Riley committed facility, per claims 13623 and 12175, a $300M ATM per claim 12188, and later a $1.0B ATM per claim 5159. That sequence reads as patching rather than design. The third regime, from December 2024, is coherent and escalating. Each convertible was upsized between launch and close. The first went from $300M to $440M, per claims 5300 and 5269. The second went from $450M to $550M, per claims 28843 and 28839, and a $600M figure was withdrawn, per claim 4782. The third went from $875M to $1.0B, per claims 21752 and 21221. Terms improved sharply over that run, from a 3.50% coupon and a ~$13.64 conversion price, per claims 28849 and 28857, to 0.00% and ~$85.63, per claims 21752 and 21758. Dilution was managed through capped calls and prepaid forwards, per claims 5273, 5274, 28841 and 28842, though the October 2025 deal dropped the prepaid forward. Use-of-proceeds language requires explanation in one respect. In May 2025 the company said debt financing workstreams were expected to execute in coming months, per claim 29812, against a net funding requirement of up to $250M, per claim 29811. The subsequent record shows convertibles instead, and claim 28352 describes them as the only non-equity financing. Disclosure has also become less specific over time: the October 2025 proceeds are framed as capped calls plus general corporate purposes, per claim 21756, while Childress improvements alone are estimated at $1.4B, per claim 31155. No insider transactions appear in the data provided, and the record is truncated after claim 21224, so later events are not assessed. A careful reader should watch for four things: whether the promised debt facilities materialize, how much of the $1.0B ATM has actually been drawn, whether the Bitmain balances flagged in claims 29844 through 29847 are confirmed as paid, and how the 2029 and 2031 conversion and redemption windows, per claim 21760, interact with the share price.

IREN's capital history falls into three phases, and the shift between them matters more than any single raise. After the November 2021 IPO, which delivered $215.3M of net proceeds per claim #14390, the company used asset-level debt: the $71M NYDIG equipment facility per claim #14216. By June 2023 those facilities were under lender enforcement with a PwC receiver appointed, per claims #12447 and #12462. A settlement of the subsidiary defaults was reached only in August 2025, per claim #4429. That episode appears to have ended secured equipment debt in the mix. The company spent 2022 and 2023 on small, repeated B. Riley committed equity facilities, with the $25M tranches restated across claims #13325, #12526 and #12203. It then moved to a $300M ATM per claim #12188 and a $1B ATM per claim #5159. This was incremental, needs-driven equity rather than a planned structure.

From December 2024 the strategy becomes coherent and repeatable. Three convertibles followed the same template: a $440M issue per claim #5269, a $550M issue per claim #28839, and a $1.0B zero-coupon issue due 2031 per claim #21221. Each was upsized from its launch size, as claims #5300, #4796 and #21761 show. Each was paired with capped calls, including $53.8M per claim #28841 and $49.6M per claim #21756. The first two also carried prepaid forward repurchases of $74M and $92.5M, per claims #5274 and #28842. Together these show deliberate management of headline dilution and of hedging-driven selling. The initial conversion price moved from about $13.64 per claim #28857 to about $85.63 per claim #21758 in four months. That move shows how much the equity's repricing lowered the dilution cost of each dollar raised.

Disclosed plans and actual funding do not fully match. In April and May 2025 the company described GPU financing and debt workstreams expected to execute in coming months, per claims #31055 and #29812. By late June, convertibles were described as the only non-equity financing, per claim #28352, and a $130M GPU purchase was funded from existing cash per claim #28341. The debt workstreams are delayed or no longer discussed, and that requires explanation. The October proceeds are largely allocated to general corporate purposes per claim #21756, which gives limited traceability.

Readers should next watch for three things. First, whether the 2025 funding requirement of up to $250M per claim #29811 and the Bitmain balances in claims #29844 through #29847 were met as scheduled. Second, how much of the $1B ATM has been drawn. Third, whether project-level debt materializes or convertibles remain the default instrument. This record contains no insider transactions, and it is truncated after claim #21223, so later events cannot be assessed here.

IREN's financing record shows two distinct phases. From its 2021 listing through 2023, the company relied on equity: $215.3M of IPO net proceeds (claim 14390), a B. Riley committed equity facility (claims 13623, 13325), and a $300M at-market program (claim 12188). The one non-equity instrument from that period, the $71M NYDIG equipment facility (claim 14216), ended in lender enforcement and a PwC receivership (claims 12447, 12462). It then went through litigation over collateral (claim 12270) and was settled only in August 2025 (claim 4429). From December 2024 the company shifted to convertibles, and the cadence quickened. Issues of $440M (claim 5269), $550M (claim 28839) and $1.0B (claim 21221) came within about ten months, alongside a $1B ATM program (claim 5159). Conversion prices rose from about $13.64 (claim 28857) to about $85.63 (claim 21758), so the most recent issue carries much less potential share count per dollar raised.

The mix looks more deliberate than improvised. Each convertible came with capped calls of $44M, $53.8M and $49.6M (claims 5273, 28841, 21756). The first two also included prepaid forwards of $74M and $92.5M (claims 5274, 28842). That is consistent with managing dilution while using the ATM as a backstop, as the April 2025 strategy described (claim 31056). Two points need explanation. First, the May 2025 debt financing workstreams, said to be expected "in coming months" (claim 29812), are not reflected in later claims. A June 2025 statement describes convertible notes as the only non-equity financing (claim 28352). Second, a $600M convertible figure appears as withdrawn (claim 4782), while the closed amount was $550M (claim 28839).

Whether announced uses tracked actual deployment can be judged only in part. The specific hedging allocations match across the priced and closed disclosures. Beyond those, the stated uses are general corporate purposes. Capital raised in 2025 also well exceeded the up-to-$250M net funding requirement flagged in May (claim 29811), and that gap is not reconciled in the record. A $130M GPU purchase was described as funded from existing cash (claim 28341). Several Bitmain balances, including $119.3M due October 2025 (claim 29846), remain not yet verifiable.

The record provided ends mid-entry at claim 21223 and contains no insider transactions, so this read does not cover insider selling. A careful reader should next watch for three things: whether the debt workstreams produce a facility, how much of the $1B ATM is drawn, and how the Bitmain obligations and conversion exposure are settled.

IREN's financing record falls into three phases. The first is a conventional equity raise, the November 2021 IPO with $215.3M of net proceeds (claim 14390), paired with asset-backed debt in the $71M NYDIG equipment facility (claim 14216). That facility later came under lender enforcement with a PwC receiver appointed (claims 12462, 12447), went through litigation over whether Bitcoin was collateral (claim 12270), and reached a settlement agreement in August 2025 (claim 4429). The episode explains much of what followed. After it, the company did not return to secured equipment debt in this record.

The second phase, from 2022 to 2023, relied on drip-feed equity. The B. Riley committed facility (claims 13623, 13325) was followed by a $300M ATM (claim 12188), which was later replaced by a $1B ATM (claim 5159). The record does not show how much of either ATM was actually drawn, so the true pace of dilution in this period cannot be determined from the data here.

The third phase, beginning December 2024, is a run of convertible notes that grew in size each time and were consistently hedged. The issues were $440M (claim 5269), then $550M (claim 28839), then $1B including the overallotment (claims 4212, 4213). Each carried capped calls, and the first two also carried prepaid forwards (claims 5274, 28842). Taken together, the structure looks deliberate: the company leans on convertibles while using derivatives to shift dilution further out. The move in initial conversion price from roughly $13.64 (claim 28857) to roughly $85.63 (claim 28758) shows that it priced each issue on the share price available at the time. The 0% coupon on the 2031 notes (claim 21752) reflects the same approach.

The use-of-proceeds language is generic, mostly hedge costs plus general corporate purposes (claims 21756, 28848), so tracking how the money was deployed is limited. The one specific allocation, a $130M GPU purchase, was attributed to existing cash rather than to any offering (claim 28341). The May 2025 disclosures pointed to debt financing workstreams expected within months (claims 29825, 29812) and a net funding need of up to $250M (claim 29811). In June 2025, convertibles were described as the only non-equity financing (claim 28352), and the subsequent record shows no completed debt facility. Those workstreams are delayed or no longer discussed. The Bitmain balances (claims 29844–29846) remain not yet verifiable, and an earlier $600M issuance figure was withdrawn (claim 4782).

The record contains no insider transactions, and it ends mid-entry in October 2025. Readers should watch for disclosure of ATM utilisation, for whether non-convertible debt actually materialises, and for how the company plans to handle the 2029–2031 maturities as they build up.

IREN's financing history moves through three distinct regimes. After an IPO that netted roughly $215.3M (claim 14390), the company first relied on asset-level leverage through the $71M NYDIG equipment facility (claim 14216). That route effectively closed when the lender enforced and a receiver was appointed (claims 12447, 12462). The litigation that followed (claims 12270, 12274) was only resolved by a settlement in August 2025 (claim 4429). Equity facilities filled the gap: a B. Riley committed facility (claims 13623, 13325), a $300M ATM (claim 12188), and then a $1B ATM (claim 5159). This pattern reads as reactive rather than planned. Dilution in this period was continuous and price-taking, although the record does not disclose actual ATM draw-downs, so the realized cadence cannot be measured here.

From December 2024 the mix shifts decisively to convertibles, and the trajectory is one of escalating size and improving terms. Each deal was upsized from its launch amount. The first went from $300M to $440M (claims 5300, 5269). The second went from $450M to $550M (claims 28843, 28839). The third went from $875M to $1.0B (claims 21752, 21221), with the greenshoe described as fully exercised (claim 21222). The conversion price rose from about $13.64 (claim 28857) to about $85.63 (claim 21758), with capped calls set at $20.98 and $120.18 respectively (claims 28858, 21225). That progression suggests a coherent strategy of deferring dilution to higher prices once market access allowed it. However, the earlier deals spent meaningful proceeds on prepaid forwards: $92.5M from the June 2025 deal alone (claim 28842).

Use of proceeds has become less specific over time. The October 2025 deal allocates $56.7M to capped calls and the remainder to general corporate purposes (claims 21224, 21764). In May 2025, the company described debt financing workstreams expected to execute in coming months (claim 29812). By June it characterized convertibles as its only non-equity financing (claim 28352), and the debt workstreams are no longer discussed in the record. Meanwhile, a $130M GPU purchase was funded from existing cash (claim 28341), while sizable Bitmain balances remain not yet verifiable (claims 29846, 29847).

A $600M convertible figure appears as withdrawn (claim 4782) and requires reconciliation against the $550M close. No insider transactions appear in the supplied record, and the claim list appears truncated. A careful reader should watch for disclosed ATM utilization, whether non-convertible debt materializes, confirmation that the Bitmain obligations were settled, and how the roughly $2B of notes due 2029 and 2031 is ultimately serviced or converted.

IREN's record traces a shift from emergency equity to deliberately structured convertible debt. After the IPO (net proceeds of $215.3M, per claim 14390), the company relied first on asset-level debt. The NYDIG equipment facility (claim 14216) later went into lender enforcement with PwC appointed as receiver (claims 12447, 12462). That matter ran through a court ruling (claim 12270) and was settled only in August 2025 (claim 4429). During that period the B. Riley committed equity facility is restated across at least six filings at $25,000,000 (claims 13325, 12442, 12203, among others). An ATM followed at $300M (claim 12188) and was later enlarged to $1.0B (claim 5159). Through 2023 the pattern looks like patching: small, repeated equity draws while the original debt structure was being unwound. From December 2024 the approach becomes coherent. Three convertibles grew in size: $440M (claim 5269), then $550M at 3.50% (claim 28839), then $1.0B at 0.00% due 2031 (claim 21221). Each carried capped calls ($44M, $53.8M and $56.7M, per claims 5273, 28841 and 21224). The first two also funded prepaid forward repurchases ($74M and $92.5M, per claims 5274 and 28842). The October 2025 deal drops the prepaid forward and directs the remainder to general corporate purposes (claim 21764), which leaves less of the proceeds hedged than before. The conversion price rose from roughly $13.64 (claim 28857) to roughly $85.63 (claim 21758), so each dollar raised now carries much less potential dilution. One upsized tranche of $600M is marked withdrawn (claim 4782) and requires explanation. On use of proceeds, the May 2025 disclosure described separate debt financing workstreams expected within months (claims 29812, 29825). By late June the company described convertible notes as its only non-equity financing (claim 28352), and the debt workstreams are no longer discussed in this record. The $130M GPU purchase was funded from existing cash (claim 28341). The record does not yet tie the Bitmain balances (claims 29844 to 29847, all not_yet_verifiable) to specific raises. Readers should watch for three things: how much of the $1.0B ATM has been drawn, which this record does not disclose; whether the share price rises past the $20.98 cap on the 2029 notes (claim 28858), beyond which the hedge no longer offsets dilution; and whether the promised debt workstreams resurface. No insider transactions appear in the supplied data, and the claim list appears truncated after October 2025.

IREN's financing record shows a clear sequence: equity, then partly failed asset-backed debt, then equity-linked capital in rapidly growing sizes. After the IPO, which delivered roughly $215M net per claim #14390, the company relied on a $71M NYDIG equipment facility (claim #14216). That facility ended in lender enforcement and a receivership (claims #12447, #12462), followed by litigation over collateral scope (claim #12270) and a settlement (claim #4429). Through 2022 and 2023 the funding was mostly incremental equity: a B. Riley committed equity facility repeatedly referenced at $25M (claims #13325, #12203), then a $300M ATM (claim #12188) and a $1B ATM program (claim #5159). That stretch looks more like patching than design.

From December 2024 the mix becomes more coherent. Three convertible offerings followed a consistent template: the $440M notes (claim #5269), the $550M 3.50% 2029 notes (claim #28839), and the $1.0B 0.00% 2031 notes (claim #21221). Each was paired with capped calls, and the first two also carried prepaid forward repurchases of $74M and $92.5M (claims #5274, #28842). Each deal was upsized from its launch size, which points to strong investor demand. The jump in conversion price from about $13.64 (claim #28857) to about $85.63 (claim #21758) within four months suggests the June 2025 notes now carry real share-count exposure, which the capped calls only partly offset.

The stated use of proceeds has become less specific over time. Earlier offerings itemized hedging costs. The October 2025 deal named capped calls and general corporate purposes only (claim #21764). In May 2025 the company described debt financing workstreams expected to execute in coming months (claims #29812, #29825) against a funding need of up to $250M (claim #29811). By June it described convertible notes as its only non-equity financing (claim #28352), so the debt workstreams are no longer discussed in this record. A $130M GPU purchase funded from cash (claim #28341) shows capital moving toward compute, but the record does not link specific proceeds to specific deployments.

Going forward, a reader should watch three things: whether the debt workstreams reappear, how much of the $1B ATM is drawn, and whether the Bitmain balances listed as not yet verifiable (claims #29844 to #29847) were paid as scheduled. One gap needs noting: this record contains no insider transactions, and the claim list appears truncated after October 2025, so conclusions about later activity cannot be drawn.

IREN's financing record traces an arc from an equity-funded miner to a recurring issuer of convertible debt, with each convertible round larger than the last. After the IPO (net proceeds of $215.3M, claim 14390), the company used small-ticket equity tools in 2022 and 2023. These were a B. Riley share purchase facility (claim 13623), repeated $25M committed equity facility filings (claims 13325, 12442, 12203) and a $300M at-market program (claim 12188). That period also included equipment financing placed under lender enforcement with a receiver appointed (claims 12447, 12462), which reached a settlement only in August 2025 (claim 4429). From late 2024 the cadence changes. Convertible offerings closed at roughly six-month intervals: $440M (claim 5269), $550M (claim 28839) and $1.0B (claim 21221). Each was upsized from its initial announcement (claims 5300, 28843, 21752).

The mix since 2024 looks deliberate. Every convertible carried capped calls (claims 5273, 28841, 21224), and the first two also funded prepaid forward repurchases of $75.9M and $92.5M (claims 5306, 28842). That structure is consistent with managing dilution rather than patching gaps. The rise in initial conversion price from about $13.64 (claim 28857) to about $85.63 (claim 21758) within four months shows that later issuance came on far stronger equity terms. The $1B ATM authorization (claim 5159) sits alongside the converts, but the record does not show how much of it has been drawn.

Use of proceeds is harder to trace. The October 2025 notes direct the remainder after capped calls to general corporate purposes (claim 21756). The record does not link those proceeds to specific deployment. Separately, a $130M GPU purchase was described as funded from existing cash (claim 28341). In May 2025 management stated a net funding requirement of up to $250M (claim 29811) and debt workstreams expected to execute in coming months (claim 29812). By late June, the convertible notes were described as the only non-equity financing (claim 28352), and the debt workstreams are no longer discussed in the record. Several Bitmain balance obligations remain not yet verifiable (claims 29844 to 29847). A $600M issuance figure is marked withdrawn (claim 4782) and requires explanation against the $550M close.

No insider transactions appear in the data provided, and the final claim is truncated. A careful reader should watch four things: whether the debt workstreams reappear, how much of the ATM is actually drawn, whether proceeds are linked to identified capital projects, and how the converts are settled as the 2029 maturity and the January 2029 redemption window for the 2031 notes (claim 21760) approach.

IREN's financing record shows two distinct regimes. From 2021 to 2023 the company relied on equity: IPO net proceeds of $215.3M (claim 14390), the NYDIG equipment facility (claim 14216), a B. Riley share purchase facility first described at $100M (claim 13623) and later restated several times at $25M (claims 12460, 12203), and a $300M ATM (claim 12185). That period ended with the equipment financing in lender enforcement and a PwC receiver appointed (claim 12447). The court affirmed that Bitcoin was not collateral (claim 12270), and a settlement of the subsidiary defaults followed only in August 2025 (claim 4429). From December 2024 the company moved to serial convertibles, each upsized at pricing: $440M (claim 5269), $550M against an initial $450M (claims 28839, 28843), and $1.0B of zero-coupon 2031 notes against an initial $875M (claims 21221, 21761). The initial conversion price rose from about $13.64 (claim 28857) to about $85.63 (claim 21758), so each dollar raised carries less potential dilution even as gross issuance grows. Alongside this sat a $1B ATM (claim 5159), whose actual usage the record does not show.

The convertible structure looks deliberate rather than patched together. Every deal carried capped calls ($44M, $53.8M and $56.7M, per claims 5273, 28841 and 21224), and the first two also funded prepaid forward repurchases of $74M and $92.5M (claims 5274, 28842). Together these hedges absorbed a meaningful share of net proceeds, which management described as limiting dilution (claim 21765). Use of proceeds is harder to trace. The October 2025 notes left most of the money for general corporate purposes (claim 21756). The one specific deployment on record, a $130M GPU purchase, was funded from existing cash rather than a named raise (claim 28341). The record does not link GPU-related capex of $5.8B (claim 4160) to identified funding sources. The GPU equipment financing figure (claim 20254) has no stated unit and needs clarification. No insider transactions appear in the data provided, and the record is cut off at claim 19124.

The November 2025 package needs the closest reading. It pairs 2032 notes (claim 4029) with a registered direct share offering used to repurchase earlier convertibles (claims 4028, 19124). That is the first instance of straight equity being issued to retire convertibles, and its terms are not visible here. Readers should watch for the size and pricing of that exchange, any disclosed ATM drawdowns, and how the gap between raised capital and stated GPU capex is closed. Given the 2023 enforcement history, they should also watch whether new equipment financing is structured with the same limited recourse.

IREN's financing record traces a company that moved from equity-and-equipment-debt funding to a repeated convertible-note program within roughly three years. After the IPO's $215.3M in net proceeds (claim 14390), the early capital stack relied on non-recourse equipment financing, including the $71M NYDIG facility (claim 14216), and on equity lines, notably the B. Riley share purchase facility (claim 13623). The equipment debt ended up in lender enforcement with a PwC receiver appointed (claims 12462, 12447), followed by a court ruling that bitcoin was not collateral (claim 12270) and a settlement agreement in August 2025 (claim 4429). That episode appears to mark the end of asset-level debt as a primary tool. Capital then came from ATM capacity, $300M in September 2023 (claim 12185) rising to $1B in January 2025 (claim 5159), and from convertibles.

The convertible cadence is the defining pattern, and each deal was upsized before closing. The December 2024 notes grew to $440M (claims 5300, 5269). The June 2025 notes moved from $450M to a $550M close (claims 4796, 28839). The October 2025 notes moved from $875M to $1.0B with the greenshoe fully exercised (claims 21761, 21222). Every deal paired the notes with capped calls and, in the first two, prepaid forward repurchases: $44M and $74M in December 2024 (claims 5273, 5274), and $53.8M and $92.5M in June 2025 (claims 28841, 28842). This reads as a deliberate structure for managing dilution, not ad hoc patching. A sophisticated reader should still note that hedging absorbed a meaningful share of gross proceeds. The step-up in conversion price, from roughly $13.64 (claim 28857) to roughly $85.63 (claim 21758), leaves the earlier tranches sitting at very different dilution thresholds from the later ones.

Use of proceeds is described mainly as capped calls plus general corporate purposes (claims 21756, 21764), so tracking actual deployment is hard. One GPU purchase was funded from existing cash, $130M per claim 28341. The later disclosure of $5.8B in GPU-related capital expenditure (claim 4160) is far larger than the convertible net proceeds on record, and how that gap is funded requires explanation. Claim 28352's June 2025 statement that convertibles were the only non-equity financing also predates the GPU equipment financing recorded in claim 20254. Claim 4782, a $600M issuance, is marked withdrawn.

The record provided is truncated after November 2025 and contains no insider transactions, so that dimension cannot be assessed here. The next items to watch are the 2032 notes (claim 4029) and the registered direct offering of ordinary shares to fund repurchases of existing notes (claims 4028, 19123). That refinancing directly swaps contingent dilution for issued shares. Disclosure of how much of the ATM capacity has been drawn would also help.

IREN's financing history falls into three phases. The first ran from the IPO, which netted $215.3M per claim #14390, through 2023, and relied on small committed equity facilities, typically $25M per claims #13325 and #12203, alongside asset-level equipment debt such as the $71M NYDIG facility per claim #14216. That equipment debt did not hold. A receiver was appointed per claim #12447, a court affirmed that Bitcoin was not collateral per claim #12270, and a settlement of subsidiary defaults followed in August 2025 per claim #4429. The second phase replaced these facilities with ATM capacity: $300M per claim #12185, then $1.0B per claim #5159. The third phase, from December 2024, is a run of convertibles of rising size. The $440M notes per claim #5269 were followed by $550M per claim #28839 and $1.0B per claim #21221, with a further 2032 tranche announced per claim #4029. Each was upsized from launch, which points to strong demand rather than distress. The record does not show how much of either ATM program was actually drawn, so the true pace of share dilution cannot be measured from these claims.

The convertible structure has been consistent. Capped calls ($53.8M per claim #28841, $56.7M per claim #21224) and prepaid forwards ($92.5M per claim #28859) were bundled to soften dilution. The initial conversion price rose from roughly $13.64 per claim #28857 to roughly $85.63 per claim #21758. This reads as deliberate rather than opportunistic, though it builds a layered stack of notes that will need managing. The November 2025 registered direct offering used to fund repurchases of existing notes, per claim #4028, is the first sign of that management. It swaps debt for equity, and its terms are cut off in the record supplied. A $600M issuance is marked withdrawn per claim #4782, which requires explanation.

Use of proceeds is described mainly as capped calls and general corporate purposes, per claim #21764, so the record cannot link raises to specific deployment. The one explicit tie is a $130M GPU purchase funded from existing cash, per claim #28341. Planned GPU capital expenditure of $5.8B, per claim #4160, linked to the Microsoft agreement per claim #4163, is well above the cumulative convertible proceeds. Claim #28352 describes convertibles as the only non-equity financing, yet GPU equipment financing reappears per claim #20254.

No insider transactions are on record. Readers should watch how the $5.8B program is funded, how far the ATM is drawn, the final terms of the 2025 note repurchase, and whether the new equipment financing differs structurally from the limited-recourse arrangements that failed in 2023.

IREN's record shows two distinct funding regimes. From the IPO through 2023, the company raised equity in small increments. The IPO netted $215.3M (claim 14390). After that came a $100M B. Riley share purchase facility (claim 13623), and a $25M committed equity facility that reappears in at least five filings across 2023 (claims 13325, 12526, 12442, 12369, 12203). A $300M ATM followed in September 2023 (claim 12188). Over the same stretch, the 2022-vintage equipment financing (claim 14216) went into lender enforcement with a PwC receiver appointed (claims 12447, 12462). It was litigated over collateral scope (claim 12270) and only reached a settlement agreement in August 2025 (claim 4429). That period reads as patching around a distressed debt layer rather than executing a planned structure.

From December 2024 the mix turns coherent and repetitive. Each convertible was upsized: $300M to $440M (claims 5300, 5269), $450M to $550M (claims 28843, 28839), and $875M to $1.0B with the greenshoe fully exercised (claims 21752, 21221). Each was paired with capped calls, and the first two also carried prepaid forward repurchases (claims 5273, 5274, 28841, 28842). The template is consistent, and the hedging allocations closely match what was pre-announced. The steep rise in conversion price, from roughly $13.64 (claim 28857) to $85.63 (claim 21758) four months later, means each later dollar raised converts into far fewer shares. A $1B ATM (claim 5159) sits alongside the converts, but the record contains no draw-down figures.

Use-of-proceeds disclosure has become less specific over time. Earlier deals itemized hedges. The October 2025 notes assign only $56.7M to capped calls (claim 21224) and the remainder of $979.0M net (claim 21223) to general corporate purposes (claim 21764). The record does not trace how that remainder was spent. It does show $5.8B of GPU-related capex tied to the Microsoft agreement (claim 4160), against which disclosed GPU equipment financing is $400M (claim 20254).

The record contains no insider transactions, so no conclusion can be drawn there. The disclosed history also ends mid-entry in late November 2025, so the analysis does not extend past that date. Readers should watch for four things. The first is final sizing and terms of the concurrent 2032 and 2033 notes (claim 19126). The second is the scale of the registered direct offering (claim 19123). The third is how much of the 2029 and 2030 notes is repurchased (claim 19124), since swapping lower-strike paper for new shares changes the dilution math. The fourth is how the gap between committed capex and identified financing is ultimately closed.

IREN's record divides into two distinct financing regimes, and the shift between them is the main story. From the November 2021 IPO (net proceeds of $215.3M, per claim #14390) through 2023, the company relied on equipment debt and small equity facilities. The NYDIG equipment facility (claim #14216) later fell into lender enforcement with a PwC receiver appointed (claims #12447, #12462). A court then confirmed that Bitcoin was not collateral under those limited-recourse arrangements (claim #12270), and a settlement was reached only in August 2025 (claim #4429). Over the same period, the B. Riley committed equity facility was cited at $25M in at least six filings during 2023 (e.g., claims #13325, #12203) before a $300M ATM was added (claim #12188). That sequence reads as incremental patching rather than a planned capital stack.

From December 2024 the pattern becomes repetitive and deliberate. Each convertible was upsized between launch and close: $300M grew to $440M (claims #5300, #5269), $450M grew to $550M (claims #4796, #28839), and $875M grew to $1.0B with the greenshoe fully exercised (claims #21761, #21221). Each also carried capped calls, and the first two carried prepaid forwards, with $92.5M allocated in June 2025 (claim #28842). The repeated use of these hedges indicates management is actively trying to manage dilution optics. The trajectory of the conversion price, from roughly $13.64 (claim #28857) to $85.63 (claim #21758), shows the company raising capital into strength. A $600M figure on the June 2025 notes was withdrawn (claim #4782) and requires explanation.

Use of proceeds is largely generic ("growth and data center," claim #13625; "general corporate purposes," claim #21764). The one specific deployment on record, a $130M GPU purchase, was funded from existing cash rather than tied to a raise (claim #28341). It is therefore difficult to trace proceeds to assets. The scale gap is notable: $5.8B of GPU-related capex is stated (claim #4160) against $400M of GPU equipment financing (claim #20254).

The late-2025 entries combine a concurrent $2.0B 2032/2033 convertible offering (claim #19126), a registered direct share offering (claim #19123), and repurchases of the 2029 and 2030 notes (claim #19124). This shifts the strategy from layering converts to refinancing them. The record is truncated at that point, and no insider transactions appear. A careful reader should watch the size and price of the registered direct, the terms of the note repurchases, actual ATM utilization under the $1B program (claim #5159), and how the remaining capex is funded.

IREN's financing record falls into three phases. The company listed in November 2021 with net proceeds of $215.3 million (claim 14390). It then relied on asset-level and committed-equity tools, which gave way to a run of convertible issuance from late 2024 that grew sharply in size. The $440 million notes of December 2024 (claim 5269) were followed by $550 million in June 2025 (claim 28839) and $1.0 billion in October 2025 (claim 21221). Each deal closed above its launch size. The June offering launched at $450 million (claim 28843), and a $600 million figure for it is recorded as withdrawn (claim 4782). The initial conversion price rose from about $13.64 (claim 28857) to about $85.63 (claim 21758) within four months. Each dollar raised therefore carried much less potential dilution, even as the dollar amounts rose.

The earlier period reads more as patching than strategy. The $71 million NYDIG equipment facility (claim 14216) ended in lender enforcement and a receivership (claims 12462, 12447). It was then litigated (claim 12270) and settled only in August 2025 (claim 4429). The B. Riley committed equity facility was restated at $25 million across several 2023 filings (claims 13325, 12369, 12203) and then superseded by a $300 million ATM (claim 12185). That ATM was later replaced by a $1 billion program (claim 5159). The convertible phase is more coherent. Every deal paired capped calls with prepaid forward repurchases, including the $92.5 million forward in June 2025 (claim 28842), which shows a consistent intent to manage dilution.

Use of proceeds is hard to test because most of it is stated as general corporate purposes (claim 21756). The one concrete deployment, $5.8 billion of GPU-related capital expenditure (claim 4160), is far larger than the cumulative net convertible proceeds. The record identifies only part of the bridge: GPU equipment financing (claim 20254) and the Microsoft agreement payment terms (claim 4163). How the full commitment is funded requires explanation.

The record contains no insider transactions, so no view on insider selling is possible. It also ends with the November 2025 transactions. These comprise concurrent 2032 and 2033 notes described at $2,000 (in millions, per claim 19126), a repurchase of the 2029 and 2030 notes (claim 19127), and a registered direct share offering to fund that repurchase (claim 4028). A careful reader should watch how much of that refinancing settles in ordinary shares, since that would turn contingent dilution into realized dilution. Drawdowns under the $1 billion ATM, which are not disclosed here, are also worth tracking.

IREN's record divides into two distinct financing regimes, and the transition between them is the most informative feature of the history. Between the November 2021 IPO (net proceeds of $215.3M, per claim #14390) and late 2023, the company relied on small, repeatedly re-disclosed equity lines. The $25M B. Riley committed equity facility appears in at least six filings (claims #13325 through #12203). In the same period, the $71M NYDIG equipment financing (claim #14216) moved into lender enforcement and receivership (claims #12462, #12447). The record shows that episode was still being resolved through a settlement in August 2025 (claim #4429). That phase reads as patching rather than design. The $300M ATM in September 2023 (claim #12185) and the $1B ATM in January 2025 (claim #5159) widened capacity considerably, but the record contains no utilization data for either, so the realized dilution from these programs cannot be assessed here.

From December 2024 onward the instrument mix becomes coherent: four convertible issues of escalating size, each paired with capped calls and, in the first two, prepaid forward repurchases. The $440M issue (claim #5269), the $550M issue (claim #28839), the $1.0B zero-coupon 2031 notes (claim #21221) and a further $2B across 2032 and 2033 maturities (claim #19126) show both scale and pricing power improving. The initial conversion price rose from roughly $13.64 (claim #28857) to roughly $85.63 (claim #21758). Stated uses track closed allocations closely. The prepaid forward held at $92.5M through upsizing (claims #28854, #28842), and the capped-call amounts scaled with greenshoe exercise ($48.9M to $53.8M, claims #28853 and #28841; $49.6M to $56.7M, claims #21756 and #21224). A $600M figure was withdrawn in favor of $550M (claims #4782, #28839).

The larger question concerns deployment against commitments. The record cites $5.8B of GPU-related capital expenditure (claim #4160) alongside a GPU equipment financing whose units are not stated (claim #20254). A June 2025 statement that convertibles were the only non-equity financing (claim #28352) sits awkwardly against the earlier equipment debt and requires explanation. The November 2025 plan to repurchase the 2029 and 2030 notes using a registered direct share offering (claims #19127, #4028) converts contingent dilution into immediate dilution. The record is truncated at this point, so its scale and pricing are not visible. No insider transactions appear on record.

A careful reader should watch for ATM utilization disclosures, the final share count from the registered direct offering, and whether funding sources disclosed under the Microsoft payment terms (claim #4163) close the gap to the stated capex.

IREN's financing record traces a company that moved from equity-funded growth, through a period of balance-sheet repair, into large-scale convertible issuance. The cadence has accelerated sharply. The December 2024 convertible closed at $440M (claim 5269). The June 2025 deal reached $550M (claim 28839), and the October 2025 zero-coupon 2031 notes closed at $1.0B (claim 21221). A further 2032 offering was announced in November 2025, listed at 1,000 (claim 4029). Each deal was upsized from its initial terms, for example $450M to $550M across claims 4796 and 28839, which suggests investor demand was not a constraint. The rise in conversion price, from roughly $13.64 (claim 28857) to roughly $85.63 (claim 21758), means each dollar raised later implies fewer potential shares.

The mix reads as more deliberate than opportunistic after 2023. The earlier period does require context. The $71M NYDIG equipment facility (claim 14216) ended in lender enforcement and a PwC receivership (claims 12447, 12462), and a settlement of subsidiary defaults followed in August 2025 (claim 4429). Equity programs then scaled from a $25M B. Riley facility (claim 13325) to a $300M ATM (claim 12188) and later a $1B ATM (claim 5159). Every convertible was paired with capped calls, and the earlier two also carried prepaid forwards, which absorbed roughly $145M of the June 2025 net proceeds (claims 28841, 28842). This is a consistent anti-dilution posture, though it reduces the cash actually available for deployment.

On use of proceeds, the record states intentions, mainly capped calls, prepaid forwards and general corporate purposes (claims 21756, 28848), rather than deployment. The disclosed GPU-related capex of $5.8B (claim 4160) is several times the cumulative convertible proceeds. One $130M GPU purchase is described as funded from existing cash (claim 28341), and the record does not reconcile the remaining gap. The November 2025 plan to repurchase 2029 and 2030 notes using a registered direct share offering (claims 19127, 4028) requires explanation. No 2030 issue appears on the record, and the swap converts debt back toward equity.

A reader should watch for the final size and pricing of the 2032 notes and the companion share offering. They should also watch actual drawdowns on the $1B ATM and the financing sources disclosed for the remaining GPU commitments. The record supplied is truncated after claim 19123 and contains no insider transactions, so no conclusions about insider behavior are possible.

IREN's financing record falls into three phases. The 2021 IPO raised net proceeds of about $215.3M (claim 14390). Through 2022 and 2023 the company relied mainly on small instruments: a $71M NYDIG equipment facility (claim 14216) and a B. Riley committed equity facility of $25M, restated across several filings (claims 13325, 12442, 12203). That period ended with the equipment facilities under lender enforcement and a PwC receiver appointed (claims 12462, 12447). Litigation followed (claim 12270), and the matter was resolved only by the August 2025 settlement (claim 4429). From late 2023 the company moved to programmatic equity, with a $300M ATM (claim 12185) expanded to $1B (claim 5159). It then turned to convertibles of accelerating size: $440M (claim 5269), $550M (claim 28322), $1.0B (claim 21221), and $2B across 2032 and 2033 maturities (claims 18752, 18753).

The convertible sequence looks deliberate rather than patched. Each tranche carried capped calls (claims 5273, 4784, 21224, 4017), and the first two also carried prepaid forwards (claims 5274, 4785), a consistent approach to managing dilution. The December 2025 transaction changes the picture. The 2029 and 2030 notes, $1,024.2M and $608.2M (claims 4019, 4018), were repurchased using a registered direct offering whose 39,699,102 figure (claim 18755) appears to be a share count. The June 2025 notes converted at roughly $14 (claim 4788), while the October notes converted at roughly $85.63 (claim 21758). Retiring debt with shares at that gap turns contingent dilution into realized dilution. The record does not show the per-share economics of that exchange, and they require explanation.

Stated uses of proceeds have narrowed to capped calls and general corporate purposes (claims 21764, 21756), so tracking actual deployment against them is difficult. The one concrete deployment item, a $130M GPU purchase, was funded from existing cash (claim 28341) rather than from any specific raise. The disclosed GPU capital expenditure of $5.8B (claim 4160) is larger than any single financing on record. Beyond a $400M GPU equipment financing (claim 20254), the record does not show how that gap is bridged.

A careful reader should watch four things. First, actual draws under the $1B ATM, which the record does not report. Second, whether further convertibles are refinanced through share issuance. Third, the funding bridge for the $5.8B commitment tied to the Microsoft agreement (claim 4163). Fourth, a $600M issuance figure that was withdrawn (claim 4782). No insider transactions appear in the record, so no read on insider positioning is possible.

IREN's financing record has two phases. From the 2021 IPO through 2023, the company relied on small, repeated equity facilities. The B. Riley committed equity facility appears in at least six filings at $25M each (claims #13325, #12442, #12203). It was later joined by a $300M at-the-market program (claim #12188). Over the same period, the NYDIG-linked equipment financing was in enforcement, with a PwC receiver appointed (claims #12462, #12447). A court ruling held that Bitcoin was not collateral for those limited-recourse arrangements (claim #12270), and a settlement of the subsidiary defaults followed only in August 2025 (claim #4429). This phase looks less like a set strategy and more like keeping capital available while the debt overhang was resolved. From December 2024 the mix moves decisively to convertible notes, and each deal is larger than the last. The sequence runs from $440M (claim #5269) to roughly $535M net in June 2025 (claim #4783), then $1.0B in October 2025 (claim #21221), then $2B across 2032 and 2033 maturities in December 2025 (claim #18762). Every issue carried capped calls, and the earlier two also carried prepaid forwards (claims #5273, #4785, #21224, #18757), which shows consistent attention to managing headline dilution. The initial conversion price rose from about $14 (claim #4788) to about $85.63 (claim #21758). That rise let later paper carry lower coupons. The June 2025 deal was revised several times on the record, at $450M, $500M, $550M and a withdrawn $600M (claims #4796, #4790, #28322, #4782). Readers should treat those figures as sequential revisions, not separate raises. The December transaction requires the closest reading. A registered direct offering of about 39.7M shares (claim #18755) funded the repurchase of $1,632M of 2029 and 2030 notes (claim #18756). This crystallized dilution that had previously been contingent. On use of proceeds, the disclosed uses are mostly hedging costs and general corporate purposes. The stated GPU capital requirement of $5.8B (claim #4160) and the Microsoft agreement (claim #4163) give the likely destination, but the record shows actual deployment only in a $130M GPU purchase funded from cash (claim #28341), so tracking cannot be confirmed from this data. The record contains no insider transactions, and the final entries are truncated. Going forward, readers should watch how the $1B ATM (claim #5159) is drawn against the $2B of 2032 and 2033 notes, whether further share-funded repurchases follow, and how the GPU equipment financing (claim #20254) is sized against the $5.8B commitment.

IREN's record shows a clear shift in its financing mix over time. From 2021 to early 2025 the company relied mainly on common equity. After the IPO (net proceeds of $215.3 million, per claim #14390), it layered committed equity facilities with B. Riley (claims #13623, #13325) and then at-the-market programs of $300 million (claim #12185) and $1.0 billion (claim #5159). The record does not show how much of either ATM was actually drawn, so the dilution those programs produced cannot be measured here. The one early experiment with secured debt, the $71 million NYDIG equipment facility (claim #14216), ended in lender enforcement and a PwC receivership (claim #12447), went through litigation (claim #12270), and was settled in 2025 (claim #4429). Since then, asset-level debt is largely absent until the $400 million GPU equipment financing (claim #20254). From December 2024 onward the company moved to convertible notes. The $440 million (claim #5269), the roughly $550 million (claim #28322) and the $1.0 billion zero-coupon 2031 notes (claim #21221) each came with prepaid forwards or capped calls meant to limit dilution (claims #5274, #4785, #21224). That repeated structure reads as a deliberate template rather than patching. The December 2025 transaction is different and needs closer reading. The company used a registered direct share offering to fund a $1.632 billion repurchase of its 2029 and 2030 notes (claims #18761, #4019, #4018) and issued $2.0 billion of new 2032 and 2033 notes alongside it (claim #18762). In effect, earlier convertible debt became equity, and the maturities were refinanced further out. Stated use of proceeds has also changed. It went from growth and data center buildout (claim #13625) to capped calls plus general corporate purposes (claim #21764). Meanwhile a $5.8 billion GPU-related capital commitment tied to the Microsoft agreement (claims #4160, #4163) far exceeds the convertible proceeds on record. Only $130 million of GPU spending is explicitly attributed to existing cash (claim #28341). The listed June 2025 issuance of $600 million is marked withdrawn (claim #4782) against a stated $550 million (claim #28322), which also requires reconciliation. A careful reader should watch for disclosure of how the GPU commitment will be funded, ATM drawdowns, the share count after the December exchange, and whether another refinancing follows. The record provided contains no insider transactions and appears truncated after December 2025.

IREN's financing record shows two distinct regimes. Through 2023 the company funded itself in small, reactive increments. After its IPO, which raised $215.3M net per claim #14390, it signed the B. Riley committed equity facility (claim #13623), referenced repeatedly at $25M through 2023 (claims #13325, #12442, #12203), and a $300M at-the-market program (claim #12185). Over the same period the $71M NYDIG equipment facility (claim #14216) went into lender enforcement with a receiver appointed (claims #12447, #12462). That episode was resolved in the company's favour on the limited-recourse question (claim #12270) and finally settled in August 2025 (claim #4429). Funding then was patching around a constrained balance sheet rather than executing a plan. From December 2024 the mix shifts decisively to convertibles, at rapidly escalating size: $440M (claim #5269), roughly $550M (claim #28322), $1.0B at a zero coupon (claim #21221), and a further $2B across 2032 and 2033 maturities (claim #18762). A larger $1B ATM sits alongside them (claim #5159). The structures are consistent. Each deal pairs the notes with capped calls (claims #5273, #4784, #21224, #4017) and, in the earlier two, prepaid forwards (claims #5274, #4785). This reads as a deliberate effort to manage headline dilution rather than opportunism. The trajectory of conversion terms is notable, moving from an initial price near $14 (claim #4788) to roughly $85.63 (claim #21758). The December 2025 transaction then partly crystallised the earlier dilution. A registered direct share sale of about $1.63B (claim #18760) funded repurchases of the 2029 and 2030 notes (claims #4019, #4018). Hedging allocations reconcile closely between announcement and pricing across deals, but the bulk of each raise is described only as general corporate purposes. The record therefore cannot confirm deployment. The disclosed GPU capital commitment of $5.8B (claim #4160) exceeds the cumulative raises plus the $400M GPU equipment financing (claim #20254). How that gap is funded requires explanation. The June 2025 statement that convertibles were the only non-equity financing (claim #28352) is no longer accurate on the later record. No insider transactions appear in the data, and the record provided is truncated after December 2025. A careful reader should watch for drawdowns on the $1B ATM, further induced exchanges of outstanding notes, the funding source for the GPU commitment, and the Microsoft payment terms (claim #4163).

IREN's financing record traces a company that moved from small, facility-based equity access to very large convertible issuance in about three years. After the IPO's stated net proceeds of $215.3M (claim #14390), the 2022–2023 period relied on a B. Riley committed equity facility. It was first described at $100M (claim #13623) and then repeatedly referenced at $25M through 2023 (claims #13325, #12442, #12203). A $300M ATM followed in September 2023 (claim #12185), and a $1B ATM in January 2025 (claim #5159). The record does not show how much of either ATM was actually drawn, so the realized dilution from these programs cannot be assessed here and requires explanation. The debt side has a distinct history. The NYDIG-style $71M equipment facility (claim #14216) moved into lender enforcement and receivership (claims #12462, #12447), was litigated over whether bitcoin served as collateral (claim #12270), and was resolved by a settlement only in August 2025 (claim #4429). After that episode the company relied almost entirely on convertibles, which a June 2025 claim describes as its only non-equity financing (claim #28352). The convertible cadence has accelerated sharply: $440M in December 2024 (claim #5269), $550M in mid-2025 (claim #28322), $1.0B in October 2025 (claim #21221), and $2B across 2032 and 2033 notes in December 2025 (claim #18762). The earlier deals paired notes with prepaid forwards and capped calls (claims #5274, #4785), which suggests deliberate dilution management rather than patching. The December 2025 transaction is more complex. A registered direct sale of roughly 39.7M shares (claim #4012) funded repurchases of $1,024.2M of 2029 notes and $608.2M of 2030 notes (claims #4019, #4018). In effect, contingent dilution was converted into realized dilution while maturities were extended. Stated use of proceeds shifted from 'growth and data center' (claim #13625) to capped calls and general corporate purposes (claim #21764). Meanwhile, the GPU capital expenditure of $5.8B tied to the Microsoft agreement (claims #4160, #4163) is far larger than the convertible proceeds on record. Only a $400M GPU equipment financing (claim #20254) is identified as a bridge, and the funding plan for the remainder is not discussed. A sophisticated reader should watch for ATM utilization disclosures, further refinancings of convertibles through equity, and how the company funds the Microsoft-related build-out. No insider transactions appear in the record, and the claims list appears to end in December 2025.

IREN's financing record shows a company that moved from small, defensive equity facilities to very large convertible issuance in about two years. After the IPO's $215.3M in net proceeds (claim #14390), the 2022–2023 period consists mainly of B. Riley committed equity facilities. These include the $100M facility (claim #13623) and the repeated $25M facility references through 2023 (claims #13325, #12442, #12203). Alongside them came a $300M ATM (claim #12185). Over the same period, the NYDIG-style equipment financing went into enforcement, with a receiver appointed (claim #12447), a court ruling on collateral (claim #12270), and a settlement only in August 2025 (claim #4429). That stretch reads as patching more than strategy. From December 2024 the approach becomes coherent and repeated: upsized convertibles paired with capped calls and, at first, prepaid forwards. The December 2024 deal grew from $300M to $440M (claims #5300, #5269). The October 2025 zero-coupon notes closed at $1.0B including a fully exercised greenshoe (claim #21221), with a conversion price near $85.63 and a $120.18 cap (claims #21758, #21759). In December 2025 the company issued $2B across 2032 and 2033 maturities (claims #18752, #18753). Part of that was used to repurchase $1,024.2M of 2029 notes and $608.2M of 2030 notes (claims #4019, #4018), funded alongside a registered direct placement of 39,699,102 shares (claim #18755). The share placement makes dilution that the earlier notes had left contingent immediate, and the refinancing pushes the remaining contingent dilution further out. Stated hedging uses track closely from announcement to close; the capped call allocations of $56.7M and $174.8M (claims #21224, #4017) are examples. The broader uses do not track as clearly. The $5.8B of GPU capital expenditure (claim #4160) is far larger than the convertible proceeds plus the $400M GPU equipment financing (claim #20254), and the record does not explain how the gap is funded. The June 2025 record also needs explanation, because a $600M issuance is marked withdrawn (claim #4782) while $535M in net proceeds is stated (claim #4783). There are no insider transactions on record, and the data stops in early December 2025. A reader should watch for drawdowns against the $1B ATM (claim #5159), how the remaining GPU commitment is financed, the Microsoft payment terms (claim #4163), and whether the prepaid-forward repurchase authorization (claim #4091) is used to offset share count growth.

IREN's financing record moves through three distinct phases, and the dilution trajectory changes with each. The first was an IPO that netted $215.3M (claim #14390), paired with a $71M NYDIG equipment facility (claim #14216). That debt ended poorly: by mid-2023 the facilities were under lender enforcement with a PwC receiver appointed (claims #12447, #12462), a court affirmed that bitcoin was not collateral (claim #12270), and a settlement of the subsidiary defaults was reached only in August 2025 (claim #4429). The second phase, through 2023, relied on incremental equity: a B. Riley share purchase facility (claim #13623), a $25M committed equity facility that recurs across at least six filings (for example, claims #12442 and #12203), and a $300M ATM (claim #12185), later expanded to $1B (claim #5159). This was steady, price-sensitive dilution, closer to patching than to a plan. The third phase is the convertible market, and it shows more deliberate structuring. The December 2024 $440M notes (claim #5269) and June 2025 notes (claim #4783, $535M net) each carried capped calls and prepaid forwards. The October 2025 $1.0B zero-coupon notes (claim #21221) carried a roughly $85.63 conversion price (claim #21758). Proceeds allocations broadly tracked announcements, with modest drift: the 2024 capped call was guided at $40.3M (claim #5305) and recorded at $44M (claim #5273). The June 2025 issue appears as a $600M issuance marked withdrawn (claim #4782) alongside $535M net proceeds, and this requires explanation. Beyond the hedges, use of proceeds is largely 'general corporate,' which the record links only indirectly to the $5.8B GPU capex figure (claim #4160) and the Microsoft agreement (claim #4163). The December 2025 transaction is the pivotal event. Two new $1B tranches due 2032 and 2033 (claims #18752, #18753) funded the repurchase of the 2029 and 2030 notes, $1,024.2M and $608.2M respectively (claims #4019, #4018). A registered direct placement of 39,699,102 shares accompanied the repurchase (claim #18755). Contingent dilution from low-strike notes, the 2030 series having converted near $14 (claim #4788), was crystallized into actual shares while debt maturities were pushed out. A sophisticated reader should watch whether the ~$175M of new capped calls (claim #4017) is sized to the higher share base, how the $400M GPU financing (claim #20254) ranks against the converts, and whether ATM usage resumes. The record supplied here is truncated after December 2025, so later events are not assessed.

IREN's record divides into two eras. From 2021 to 2023 the company relied on small, layered sources: IPO net proceeds of about $215.3M (claim #14390), a $71M NYDIG equipment facility (claim #14216), and a B. Riley committed equity line that appears repeatedly in $25M tranches (claims #12442, #12369, #12203). That period ended with the equipment financing in lender enforcement and a PwC receiver appointed (claims #12447, #12462). Court rulings followed (claims #12270, #12274), and a settlement was not reached until August 2025 (claim #4429). Early financing was opportunistic patching rather than a coherent plan, and one leg of it failed.

From late 2024 the strategy becomes coherent and escalates sharply. Convertible issuance grew from $440M (claim #5269) to a June 2025 deal whose $600M size is marked withdrawn (claim #4782) against $535M of net proceeds (claim #4783). It then grew to $1.0B in October 2025 with the greenshoe fully exercised (claims #21221, #21222), and to $2B across 2032 and 2033 notes in December (claims #18752, #18753). The initial conversion price rose from roughly $14 (claim #4788) to roughly $85.63 (claim #21758), which shows the market repricing the equity. Each deal paired capped calls or prepaid forwards to manage dilution (claims #5273, #4785, #4017). In December the company repurchased its 2029 and 2030 notes for $1,024.2M and $608.2M (claims #4019, #4018), partly through a registered direct placement of 39,699,102 shares (claim #18755). This turns convertible overhang into realized share issuance. The repurchase amounts exceed the original principal of those series, which requires explanation from the filings rather than inference.

Where announced uses can be checked, they track closely. The capped call allocations moved roughly in step with upsizing, for example $40.3M announced against $44M executed (claims #5305, #5273). Most remaining proceeds are described as general corporate purposes (claim #21764), so actual deployment cannot be verified from this record. The $5.8B of GPU capital expenditure (claim #4160) is far larger than the $400M of disclosed GPU equipment financing (claim #20254).

The record contains no insider transactions. Draw-down on the $1B ATM (claim #5159) is not disclosed. The listing is also truncated after December 2025, so later events are not assessed here. A careful reader should watch four things: how the GPU capital expenditure gap is funded, whether the 2031 notes are handled the way the 2029 and 2030 series were, actual ATM usage, and settlement confirmation for the December transactions still marked not yet verifiable (claim #18754).

IREN's financing record splits into three distinct periods. The first is an equity-funded build-out: the IPO raised $215.3M net (claim 14390), and NYDIG provided a $71M equipment facility (claim 14216). The second, from 2022 through 2023, looks more like patching. A run of B. Riley committed equity facilities was registered and re-registered (claims 13623, 13325, 12442, 12203), and a $300M ATM followed (claim 12185). Over the same months the equipment financing came under lender enforcement with a PwC receiver appointed (claims 12447, 12462). A court later held that Bitcoin was not collateral for those limited-recourse loans (claim 12270), and a settlement of the subsidiary defaults came only in August 2025 (claim 4429). The third period, from December 2024, is a convertible-led strategy that is coherent but escalating in size. The issues were $440M (claim 5269), roughly $600M in June 2025 (claim 4783 records $535M net, though the $600M issuance line, claim 4782, is marked withdrawn and requires explanation), $1.0B in October 2025 (claim 21221), and $2B across 2032 and 2033 maturities in December 2025 (claims 18752, 18753). A $1B ATM sat alongside these (claim 5159).

Use of proceeds tracks the announcements closely at the hedging layer. Capped call and prepaid forward allocations moved only modestly from estimate to close, and the moves are consistent with upsizing: $40.3M to $44M (claims 5305, 5273) and $49.6M to $56.7M (claims 21756, 21224). The December 2025 transaction changes the character of the program. About $1.632B went to repurchasing the 2029 and 2030 notes (claims 4019, 4018, 18756), funded in part by a registered direct placement (claim 18755, whose 39,699,102 figure appears to be a share count). Another $174.8M went to capped calls (claim 4017). Much of that raise therefore refinanced earlier debt into longer-dated paper and equity rather than funding new deployment. That sits against stated GPU capex of $5.8B (claim 4160) and GPU equipment financing of $400M (claim 20254). The record does not yet show how the gap between those figures is closed.

The record contains no insider transactions, so no view on insider behavior is possible. A sophisticated reader should watch four things: drawdowns under the $1B ATM, which are not disclosed here; whether the Microsoft agreement's payment terms (claim 4163) materially offset the capex requirement; how often near-dated converts are pre-emptively repurchased with equity; and the cumulative share overhang implied by conversion prices ranging from about $14 (claim 4788) to about $85.63 (claim 21758). The supplied record ends mid-entry after claim 4013, so later events may be missing.

IREN's financing record moves through three distinct phases, and the progression is more orderly than the volume of claims suggests. After the IPO, which delivered $215.3M in net proceeds (claim 14390), the company relied on equipment debt and equity lines: the $71M NYDIG facility (claim 14216) and a B. Riley committed equity facility, described as $100M (claim 13623) and repeatedly cited in $25M increments through 2023 (claims 13325, 12442, 12203). The equipment debt did not hold. It entered lender enforcement with a PwC receiver (claims 12447, 12462), was tested in court (claim 12270), and was settled only in August 2025 (claim 4429). The roughly two-year resolution is a fact worth keeping in view when reading the later, larger structures. The second phase replaced equity lines with at-the-market programs, a $300M ATM (claim 12185) followed by a $1B program (claim 5159). The record does not show how much of either was drawn, so actual dilution from the ATMs cannot be measured here. The third phase is convertible-led and escalates quickly: $440M in December 2024 (claim 5269), a June 2025 issue announced at $450M (claim 4796) with $535M in net proceeds (claim 4783), $1.0B closed in October 2025 (claim 21221), and $2B across 2032 and 2033 notes in December 2025 (claims 18752, 18753). The mix reads as deliberate rather than patched. Each convertible carried capped calls, and the first two also carried prepaid forwards (claims 5273, 5274, 4784, 4785). The December transaction used new notes and a registered direct placement (claim 18755) to retire the 2029 and 2030 notes, $1,024.2M and $608.2M respectively (claims 4019, 4018). That is liability management, not new funding. Announced uses track the stated hedge allocations closely. The October capped call came in at $56.7M against a $49.6M estimate (claims 21224, 21756), a revision tied to the upsizing. Beyond the hedges, proceeds are described only as general corporate purposes, so this record cannot connect them to deployment. The $5.8B of GPU capital expenditure (claim 4160) is far larger than the $400M GPU equipment financing on record (claim 20254), and that gap requires explanation in future disclosures. A careful reader should watch for confirmed settlement of the December 2025 transactions, several of which remain not yet verifiable (claims 18754, 18760). The other open items are ATM utilization and how the GPU commitment is funded. The supplied record is truncated, so later events may be missing.

IREN's financing record traces a company that moved from small, facility-based equity to large convertible issuance in about four years, with each step larger than the last. After the IPO's $215.3M net proceeds (claim #14390), 2022–2023 relied on B. Riley committed equity facilities. The $25M facility is restated across at least five filings (claims #13325, #12442, #12203). That period ended with a $300M ATM (claim #12185), later replaced by a $1B program (claim #5159). From December 2024, convertible notes became the main instrument. A $440M issue (claim #5269) was followed by a June 2025 deal with $535M net proceeds (claim #4783), a $1.0B zero-coupon 2031 tranche (claim #21221), and a combined $2B 2032/2033 offering (claim #19126). The trajectory is steep rather than steady.

The mix reads as deliberate more than opportunistic. Every convertible carried capped calls, and the earlier ones also carried prepaid forwards ($74M per claim #5274; $92M per claim #4785), which signals active management of headline dilution. The rise in initial conversion price from roughly $14 (claim #4788) to roughly $85.63 (claim #21758) shows the company raising into strength. The December 2025 structure, however, pairs a registered direct share offering with repurchase of the 2029 and 2030 notes (claims #19127, #18761). That exchanges contingent dilution for issued shares. Those items remain marked delayed (claims #18760–#18762) and require explanation once settled. The legacy NYDIG equipment financing (claim #14216) passed through receivership and litigation (claims #12447, #12270) before a settlement (claim #4431). It now appears to be resolved rather than ongoing.

Use-of-proceeds language is too general to test against deployment. Early disclosures cite growth and data centers (claim #13625), and later ones say general corporate purposes after hedging costs (claim #21756). The stated $5.8B of GPU-related capex (claim #4160) and $400M of GPU equipment financing (claim #20254) suggest where capital is going. The record does not reconcile raised capital against spend, and it contains no insider transaction data at all.

A careful reader should watch four things: how the delayed December 2025 share issuance and note repurchases settle, how much of the $1B ATM gets drawn, how the gap between the $5.8B capex figure and capital raised to date is closed, and whether the Microsoft agreement terms (claim #4163) are disclosed in enough detail to underwrite that spending. The record is truncated in early December 2025, so later developments are not assessed here.

The record shows a funding model that moved through three distinct regimes, and the transitions matter more than any single deal. From the IPO, which netted $215.3M per claim 14390, through 2023, IREN relied on small-ticket equity tools: successive B. Riley committed equity facilities (claims 13623, 13330, 12203) and then a $300M at-market program per claim 12185. Alongside these, the company used asset-level debt, the $71M NYDIG equipment facility per claim 14216. That facility ended in receivership and lender enforcement (claims 12447, 12462), then court proceedings (claim 12270), and was not settled until August 2025 per claim 4429. Asset-backed miner debt has not reappeared in the record since. That reads as a deliberate retreat from secured equipment leverage rather than a pause.

From December 2024 the instrument of choice became the hedged convertible, and the sizing escalated quickly: $440M per claim 5269, an upsized offering netting $535M per claim 4783, and $1.0B of 2031 notes closed per claim 21221. Each early deal paired capped calls with prepaid forwards ($74M and $92M per claims 5274 and 4785), which suggests consistent management of dilution optics. The $1B ATM per claim 5159 sat in reserve throughout. This sequence is more coherent than opportunistic, although the June 2025 tranche was preceded by a $600M figure now marked withdrawn (claim 4782), which requires explanation.

The late-2025 activity is the most consequential. A concurrent $2.0B offering of 2032 and 2033 notes (claim 19126) was paired with a registered direct share sale to repurchase the 2029 and 2030 notes (claims 19127, 4028). Some of that paper, the notes due 2029 per claim 4796, had been issued only about six months earlier. Converting that debt into ordinary shares crystallizes dilution the hedges were meant to defer. The jump in conversion price, from roughly $14 per claim 4788 to roughly $85.63 per claim 21758, explains why.

On use of proceeds, the stated purposes shifted from growth and data center build-out (claim 13625) to hedges plus general corporate purposes (claim 21756). The record then discloses $5.8B of GPU capex tied to the Microsoft agreement (claim 4160). It does not show how that commitment maps to the capital raised. No insider transactions appear in the data provided, and the record is truncated in December 2025.

A careful reader should watch for the share count after the registered direct, remaining ATM capacity, terms on the GPU equipment financing (claim 20254), and whether the gap between the GPU capex commitment and the capital raised to date is closed with equity or with debt.

IREN's financing record falls into three phases. The first two read as patching, and the third is more deliberate. After the IPO (net proceeds of $215.3M, per claim 14390), the company relied on asset-level debt, notably the $71M NYDIG equipment facility (claim 14216). By mid-2023 that structure was under lender enforcement with a receiver appointed (claims 12462, 12447). The matter then went through a court ruling on collateral scope (claims 12270, 12274) and was not settled until August 2025 (claims 4429, 4431). Through 2023, equity came in small increments: the $25M B. Riley committed facility is restated across at least six filings (claims 13325 through 12204). A $300M ATM followed (claim 12185), and in January 2025 a $1B ATM (claim 5159). The record does not show how much of either ATM was actually drawn, and that gap requires explanation.

From December 2024 the mix shifts decisively to convertibles, and each deal grew between announcement and close. The first was announced at $300M (claim 5300) and closed at $440M (claim 5269). The June 2025 deal moved from $450M (claim 4796) to a $600M issuance whose status is marked withdrawn (claim 4782) against $535M of stated net proceeds (claim 4783). The October 2025 deal went from $875M (claim 21761) to $1.0B closed (claim 21221). Over the same period the initial conversion price rose from roughly $14 (claim 4788) to about $85.63 (claim 21758). That rise means each dollar raised later carries less share dilution, but the absolute convertible overhang has grown with each deal.

On use of proceeds, the early language pointed to growth and data center build-out (claim 13625). Later disclosures name only hedging lines specifically, such as capped calls and prepaid forwards (claims 5273, 5274, 4784, 4785, 21224), and leave the rest as general corporate purposes (claim 21764). The record links no specific deployment back to any raise. The $5.8B GPU capex figure (claim 4160) and the Microsoft agreement (claim 4163) are far larger than capital raised to date, so it remains open how that gap is funded beyond the $400M GPU equipment financing (claim 20254, units as recorded).

The late-2025 transaction pairs new 2032 and 2033 notes (claims 4029, 4030) with a registered direct share offering to repurchase the 2029 and 2030 notes (claims 4028, 19124, 18756). That is a maturity extension funded partly with equity. The record is truncated at this point, and several December 2025 items are marked delayed (claims 18760 to 18762). A careful reader should watch for the final settled terms, actual ATM utilization, and how the GPU commitment is financed. No insider transactions appear in the data provided.

IREN's financing record shows three distinct phases, and the dilution trajectory steepens sharply across them. After a November 2021 IPO that netted $215.3M (claim 14390), the company relied through 2022–2023 on a small B. Riley committed equity facility, repeatedly restated at $25M (claims 13330, 12580, 12369). It then moved to a $300M at-market program in September 2023 (claim 12185) and a $1B ATM in January 2025 (claim 5159). The record does not show how much of either ATM was actually drawn, so realized share issuance from these programs cannot be confirmed from the data here.

The convertible sequence is where the capital structure changed character. The December 2024 deal grew from $300M (claim 5300) to $440M (claim 5269). The June 2025 deal was announced at $450M (claim 4796) and reported net proceeds of $535M (claim 4783), alongside a $600M issuance entry that is now marked withdrawn (claim 4782); that discrepancy requires explanation. October 2025 brought a $1.0B zero-coupon note after the greenshoe was fully exercised (claim 21221). The concurrent $2B 2032/2033 offering followed in late November (claim 19126). Initial conversion prices rose from roughly $14 (claim 4788) to about $85.63 (claim 21758), which reflects equity appreciation rather than structural restraint.

The mix reads as coherent rather than opportunistic. Most deals pair capped calls or prepaid forwards with the notes ($44M and $74M respectively, per claims 5273 and 5274; $56.7M per claim 21224; $175M per claim 18757). The December 2025 step is notable. A registered direct share offering (claim 4028) funds the repurchase of $1,632M of 2029/2030 notes (claim 18756), which converts deferred dilution into immediate dilution. The earlier NYDIG equipment facility (claim 14216) moved through receivership (claim 12447) to a settlement in August 2025 (claim 4431), so that legacy is largely resolved.

On use of proceeds, early miner commitments are corroborated by itemized pre-shipment and deferred payments (claims 12141, 12142). Later offerings default to capped calls plus general corporate purposes (claim 21764). That leaves the $5.8B of GPU capital expenditure tied to the Microsoft agreement (claims 4160, 4163) without a disclosed funding bridge beyond $400M of GPU equipment financing (claim 20254).

No insider transactions appear in this record, and the claim list ends mid-entry at claim 4019, so later events are not visible. A careful reader should watch four things: the final settlement of the December offerings, several of which are marked delayed (claims 18760–18762); the share count from the registered direct; the drawdown pace under the $1B ATM; and how the remaining GPU capex is financed.

IREN's financing history shows a company that has moved through three distinct funding regimes, with each step larger and more structurally complex than the last. The early period relied on the IPO, which delivered $215.3M in net proceeds per claim #14390, and on equipment debt such as the $71M NYDIG facility in claim #14216. That debt became the period's defining episode: a receiver was appointed (claim #12447), the court affirmed that Bitcoin was not collateral under the limited recourse arrangements (claim #12270), and a settlement of the subsidiary defaults was not recorded until August 2025 (claim #4429). During that stretch the company leaned on equity sources it could draw at its own discretion. The record repeatedly restates the B. Riley committed equity facility at $25M increments (claims #12580, #12460, #12203). It then adds a $300M ATM (claim #12185), followed by a $1B ATM (claim #5159). The record does not show how much of either ATM was actually drawn, and that requires explanation before dilution from this channel can be assessed.

From December 2024 the mix shifts decisively to convertibles, and their size escalates quickly: $440M (claim #5269), net proceeds of $535M in June 2025 after a $600M figure was withdrawn (claims #4783, #4782), $1.0B in October (claim #21221), and $2.0B in concurrent 2032/2033 notes (claim #19126). Each issue carried capped calls, and the earlier issues also carried prepaid forwards, which reads as a deliberate effort to manage dilution rather than ad hoc patching. The initial conversion price, however, rose from roughly $14 (claim #4788) to $85.63 (claim #21758). The repurchase of the 2029 and 2030 notes, recorded at $1,632 (claim #18756), is funded through a registered direct share offering (claim #4028). That structure converts contingent dilution into realized share issuance now.

Use-of-proceeds language is largely confined to hedging costs and general corporate purposes. Against that, the only stated deployment is $5.8B of GPU-related capital expenditure tied to the Microsoft agreement (claim #4160), alongside $400 of GPU equipment financing (claim #20254). The gap between those two figures is not explained within this record. No insider transactions appear in the data, so no read on insider behavior is possible.

The record is truncated at the December 2025 offering. Several settlement items there are marked delayed (claims #18760, #18762). A careful reader should watch for confirmation of the share count issued in the registered direct, actual ATM utilization, and how the remaining GPU capex is ultimately funded.

IREN's financing record falls into three distinct phases, and the shift between them is more informative than any single transaction. After the IPO, which the record shows netting $215.3M (claim #14390), the company spent 2022 and 2023 on small, incremental equity. The B. Riley committed equity facility appears repeatedly at the $25M level (claims #13330, #12580, #12203), and a $300M ATM followed (claim #12185). Over the same period, the NYDIG equipment financing (claim #14216) moved into lender enforcement and receivership (claims #12447, #12462). It then went through a court ruling that bitcoin was not collateral (claim #12270), and a settlement was recorded only in August 2025 (claim #4431). That stretch reads as patching rather than strategy: the company funded itself through small equity tranches while a secured facility unwound.

From December 2024 the mix becomes coherent and much larger. Convertibles closed at $440M (claim #5269), then $535M net (claim #4783), then $1.0B (claim #21221), and then a concurrent $2B 2032/2033 offering (claim #19126). Each was paired with capped calls, and the earlier two also carried prepaid forwards (claims #5274, #4785), so management has consistently tried to blunt headline dilution. The November 2025 transaction requires explanation. It used a registered direct share offering to fund a repurchase of the 2029 and 2030 notes (claims #4028, #19127), with the repurchase recorded at $1,632M (claim #18756). In effect, earlier convertible exposure was turned into immediate share issuance while new, longer-dated paper was added. A June 2025 issuance figure of $600M is marked withdrawn (claim #4782), and several closing and settlement dates carry a delayed status.

On use of proceeds, the stated purposes moved from growth and data centers (claim #13625) to general corporate purposes (claim #21764). The record contains no deployment data, so it cannot confirm whether stated uses tracked actual spending. The disclosed $5.8B GPU capex tied to the Microsoft agreement (claim #4160), set against a $400M GPU equipment financing (claim #20254), points to a funding gap the record does not resolve. No insider transactions are on record.

A careful reader should watch four things: how the GPU capex is funded beyond the convertibles; how much has been drawn on the $1B ATM (claim #5159); the share count after the registered direct offering; and whether the delayed items settle as described. The record provided is truncated after claim #4019, and many figures appear without units, so these readings should be held provisionally.

The financing record runs from the November 2021 IPO through early December 2025. It contains no insider transactions, so nothing can be said about insider selling, and it ends mid-entry after claim 4018, which limits any read of the most recent events. Within those bounds, the record shows two distinct financing regimes.

From 2022 through 2023 the company relied on small, overlapping instruments. These included the $71 million NYDIG equipment facility (claim 14216), a B. Riley committed equity facility repeatedly referenced at $25 million (claims 13330, 12580, 12203) and later at $100 million (claim 12175), and a $300 million at-the-market program (claim 12185). The NYDIG-linked equipment financing then moved into lender enforcement with a receiver appointed (claims 12447, 12462), was litigated over collateral (claim 12270), and was settled only in August 2025 (claim 4429). That period reads more as patching than as a designed capital structure.

From December 2024 the mix shifted to large convertibles, each paired with dilution hedges. The $440 million notes (claim 5269) came with $44 million of capped calls and a $74 million prepaid forward (claims 5273, 5274). The June 2025 notes netted $535 million (claim 4783). The $1.0 billion 2031 notes netted roughly $979 million (claims 21221, 21223). The initial conversion price rose from about $14 (claim 4788) to about $85.63 (claim 21758), which means each new dollar of convertible debt carries less potential dilution per dollar raised. The repeated use of capped calls and prepaid forwards suggests this hedging pattern was deliberate rather than incidental.

The late-2025 sequence requires explanation. The company launched $2 billion of 2032 and 2033 notes (claim 19126) alongside a registered direct share offering whose stated purpose was to fund repurchases of the 2029 and 2030 notes (claims 4028, 19127). The record includes $1,024.2 million of 3.50% 2029 notes repurchased (claim 4019). Retiring earlier converts with newly issued shares turns contingent dilution into realized dilution, and several of these items remain marked delayed (claims 18760, 18761, 18762).

On use of proceeds, the convertible disclosures name hedges and general corporate purposes (claims 21756, 21764). Against those sit $5.8 billion of GPU-related capex (claim 4160) and the Microsoft agreement (claim 4163). The record contains no deployment data that would let a reader reconcile the funds raised with that spend.

A careful reader should watch for three things: how much of the $1 billion ATM (claim 5159) has been drawn, how the registered direct share count settles, and how the remaining GPU capex is funded beyond the equipment financing noted in claim 20254.

IREN's financing record moves through three distinct regimes. The first is the November 2021 IPO, which raised roughly $215.3M net (claim 14390). The second is a 2022–2023 period of incremental equity and asset-backed borrowing: the B. Riley committed equity facility, re-disclosed repeatedly at $25M (claims 13325, 12526, 12204) and later at a larger headline figure (claim 12175); a $300M ATM (claim 12185); and the NYDIG equipment facility (claim 14216). The third, from December 2024 onward, is dominated by convertible notes of rapidly increasing size: $440M (claim 5269), roughly $535M net in June 2025 (claim 4783), a $1.0B zero-coupon issue in October 2025 (claim 21221), and a concurrent $2B 2032/2033 offering in November 2025 (claim 19126). Dilution capacity has scaled accordingly. The ATM ceiling grew from $300M to $1B (claim 5159). The conversion price moved from about $14 (claim 4788) to about $85.63 (claim 21758) within four months, which means each later dollar raised carries less share issuance, while the earlier notes remain a low-strike overhang.

The 2023 period reads as patching. The equipment financing went into lender enforcement with a PwC receiver (claims 12462, 12447) and was resolved only through litigation (claim 12270) and an August 2025 settlement (claim 4429). The later convertible program is more coherent. Each deal pairs capped calls with prepaid forwards, and those hedges track their stated allocations closely: $75.9M becomes $74M (claims 5306, 5274), $40.3M becomes $44M (claims 5305, 5273), and $49.6M becomes $56.7M after the greenshoe (claims 21756, 21224). One June 2025 issuance figure is now marked withdrawn (claim 4782), which requires explanation. Beyond the hedges, most proceeds are described as general corporate purposes. The record does not show how they were deployed, including the 2022 data-center growth purpose (claim 13625).

The December 2025 structure, which raised about $1.632B via a registered direct share offering to repurchase the 2029/2030 notes (claims 18756, 18760), converts debt overhang into realized dilution. Several of its legs are still marked delayed (claims 18762, 4026). A reader should watch whether these close as described and how the $5.8B GPU capex (claim 4160) is funded relative to the disclosed financing capacity. Note also that the claims list provided ends mid-entry in early December 2025, so later events are not assessed here.

The record shows IREN financing its growth mostly through equity-linked instruments, with each round larger than the last. After the IPO's $215.3M of net proceeds (claim #14390), the company relied through 2022 and 2023 on equipment debt and a B. Riley committed equity facility (claims #14216, #13623). That facility is restated at $25M across at least five filings in 2023 (claims #13330, #12580, #12460, #12369, #12203), and no draw amounts are disclosed. Over the same period the equipment financing went into lender enforcement, with a PwC receiver appointed (claims #12447, #12462) and a court ruling that Bitcoin was not collateral (claim #12270). A settlement of those defaults appears only in August 2025 (claim #4431). From late 2023 onward the mix shifts to ATM capacity, at $300M (claim #12185) and then $1.0B (claim #5159), alongside a run of convertibles: $440M (claim #5269), a June 2025 issue whose size is recorded variously as $450M, $500M and a withdrawn $600M (claims #4796, #4790, #4782), and $1.0B closed in October 2025 (claim #21221). Each issue came with capped calls, and the first two also came with prepaid forwards (claims #5306, #4785). This hedging pattern is consistent across issues and reads as a deliberate effort to manage dilution rather than as ad hoc fixes. The initial conversion price moved from roughly $14 (claim #4788) to about $85.63 (claim #21758) within four months. That suggests the later issues were sized against a much higher share price rather than out of financial distress. The November–December 2025 package, $2.0B of 2032 and 2033 notes (claim #19126) alongside a registered direct share sale funding a $1,632M repurchase of the 2029 and 2030 notes (claims #4028, #18756), requires explanation. It effectively swaps in-the-money convertibles for new shares plus longer-dated debt, and the resulting dilution is not quantified in the record. Use-of-proceeds language is generic (general corporate purposes, data centers, per claims #13625 and #21756), and the record contains no deployment disclosures to test it against. The $5.8B GPU capex tied to the Microsoft agreement (claim #4160) is far larger than any single raise listed. The record has no insider transactions, and the December 2025 entries are cut off, so net proceeds for that package are unavailable. A reader should watch the final share count from the registered direct offering, actual ATM usage, how the remaining GPU capex beyond the $400M equipment financing (claim #20254) is funded, and whether the items marked delayed are later confirmed.

IREN's record describes a company that has changed how it funds itself in each phase. Dilution is not a steady drip but a series of accelerating steps. After the IPO, which netted $215.3M per claim #14390, the company used asset-level debt: the US$71M NYDIG equipment facility per claim #14216. That facility later went into lender enforcement with a PwC receiver appointed (claims #12447 and #12462), was litigated over Bitcoin collateral (claim #12270), and was settled in August 2025 (claim #4429). The company then moved to equity. It opened a B. Riley committed equity facility (claim #13623), which later filings repeatedly restate at a 25M-unit level (claims #12526 and #12580; the record is ambiguous between shares and dollars). A $300M ATM followed (claim #12185), then a $1B ATM (claim #5159). From December 2024 the convertible market became the primary channel. The sequence runs from $440M (claim #5269) to the June 2025 notes, where a $600M figure is marked withdrawn (claim #4782) against $535M of net proceeds (claim #4783), a point that requires reconciliation. It continues to $1.0B of zero-coupon 2031 notes (claim #21221), and then roughly $2B of 2032 and 2033 notes (claim #19126). The rising conversion price, about $14 in June 2025 (claim #4788) against about $85.63 in October (claim #21758), shows equity strength being monetized. The recurring capped calls and prepaid forwards (claims #5273, #5274, #4785, #21224) point to a deliberate effort to manage headline dilution rather than opportunistic patching. The December 2025 transaction is the most consequential. It pairs a registered direct share offering with the repurchase of the 2029 and 2030 notes, about $1,632M per claim #18756, which in effect settles earlier convertibles in shares. Several related entries are flagged delayed (claims #18760 and #18761). The disclosed record is truncated partway through these filings, so final share counts cannot be confirmed here. On use of proceeds, the stated purposes are consistently generic, such as growth and data centers (claim #13625) or capped calls plus general corporate purposes (claim #21756). The record contains no evidence tracing deployment, apart from the 2023 miner payments (claims #12141 and #12140). The $5.8B of GPU capex tied to the Microsoft agreement (claims #4160 and #4163) is far larger than the disclosed funding, including the $400M GPU financing (claim #20254). No insider transactions appear on record. A careful reader should watch four things: how the GPU capex gap is funded, how much of the $1B ATM is used, the final share issuance behind the note repurchase, and whether the delayed December items close as described.

IREN's financing record describes a company that moved from sporadic, hardware-linked capital to large-scale convertible funding within roughly a year. For its first two years after the IPO (net proceeds of $215.3M, per claim #14390), the company relied on asset-backed equipment debt and small committed equity lines. The NYDIG equipment facility (claim #14216) was under lender enforcement with a PwC receiver appointed by mid-2023 (claims #12462 and #12447), was litigated over whether bitcoin served as collateral (claim #12270), and reached a settlement only in August 2025 (claim #4429). The B. Riley committed equity facility recurs across nearly every 2023 disclosure (for example claims #13325, #12442 and #12204), which reads as incremental, as-needed funding rather than a planned capital structure. The ATM authorizations then scale from $300M (claim #12185) to $1B (claim #5159).

From December 2024 the mix becomes more deliberate. The company issued a $440M convertible (claim #5269), then a June 2025 issue with net proceeds of $535M (claim #4783) after a $600M figure was withdrawn (claim #4782), then a $1.0B zero-coupon 2031 note (claim #21221). Each issue was paired with capped calls and, in the earlier two, prepaid forwards (claims #5274 and #4785). This is a consistent dilution-management template. The rise in initial conversion price, from about $14 (claim #4788) to about $85.63 (claim #21758), reflects equity appreciation. It also means the earlier notes sit well in the money, and the late-November 2025 package makes that explicit. The company paired new 2032 and 2033 notes (claims #4029 and #4030) with a registered direct share offering to repurchase the 2029 and 2030 notes (claims #4028 and #19124). The repurchase figure is recorded as $1,632 without stated units (claim #18756). In effect, dormant convertible dilution was converted into realized share issuance.

Use of proceeds shifted from growth and data centers (claim #13625) to hedging costs plus general corporate purposes (claim #21756), which is less specific. GPU-related capital expenditure of $5.8B (claim #4160) exceeds the disclosed debt raises plus GPU equipment financing (claim #20254), and the record does not show how the remainder is funded.

The record contains no insider transactions, and the claim list is truncated in early December 2025, so later events cannot be assessed. Readers should watch four things: final settlement terms of the December notes, share count after the registered direct offering, ATM utilization against the $1B authorization, and how the Microsoft agreement's payment terms (claim #4163) align with the GPU capital expenditure schedule.

IREN's financing record shows three phases, with a sharp acceleration in the most recent one. After the November 2021 IPO, which netted roughly $215.3M per claim #14390, the company relied through 2022 and 2023 on asset-level equipment debt, the $71M NYDIG facility per claim #14216, and on small committed equity lines with B. Riley. The $25M facility is restated across many 2023 filings (claims #13330 through #12203), so it reads as one facility disclosed repeatedly rather than serial raises. The equipment-debt path was then abandoned under stress: a receiver was appointed in June 2023 per claim #12447, courts ruled on the limited-recourse structure per claims #12274 and #12270, and defaults were settled in August 2025 per claims #4429 and #4431. From that point the mix moved toward ATM equity, a $300M program per claim #12185 and then a $1B program per claim #5159, and then toward convertibles at a rising cadence. The convertibles came in December 2024 ($440M, claim #5269), June 2025 (net $535M, claim #4783), October 2025 ($1.0B zero-coupon, claim #21221) and December 2025 ($2B across 2032 and 2033 notes, claims #18752 and #18753). Each deal paired capped calls or prepaid forwards with the notes, which points to a deliberate effort to manage conversion dilution rather than improvised patching. The December 2025 transaction goes further. It issued 39,699,102 registered direct shares per claim #18755 to fund about $1,632M of repurchases of the 2029 and 2030 notes per claims #18756, #4018 and #4019. That effectively swaps earlier higher-coupon convertibles for equity plus longer-dated paper, and dilution that was contingent became realized. On use of proceeds, the early growth and data-center language in claim #13625 is generic. The later filings tie capital to GPU capex of $5.8B under the Microsoft agreement per claims #4160 and #4163, but the record does not show deployment against those commitments. Several figures require reconciliation: the October 2025 issuance appears as both $1.0B plus a $125M overallotment (claims #4212 and #4213) and as $1.0B inclusive of the greenshoe (claims #21221 and #21222), and a $600M figure for June 2025 is marked withdrawn per claim #4782. The record contains no insider transactions and is truncated after claim #18745, so the closing of the 2033 notes is not visible here. A careful reader should watch three things: how much of the $1B ATM has been drawn, whether the GPU financing, $400M per claim #20254, scales alongside the Microsoft capex, and whether converts continue to be refinanced into equity.

IREN's financing record describes a company whose capital needs have grown by roughly an order of magnitude in each phase, and whose instrument choice has moved from small, flexible equity lines toward large convertible issuance. The early period rests on the IPO's net proceeds of about $215.3M (claim 14390), the NYDIG equipment facility of $71M (claim 14216), and a series of B. Riley committed equity facilities, including the $100M facility (claim 13623) and several restated $25M tranches through 2023 (claims 13330, 12369, 12203). These were incremental, at-market tools rather than a single structured plan. The 2023 equipment financing went into lender enforcement with a receiver appointed (claim 12447), was litigated over collateral scope (claim 12270), and was settled only in August 2025 (claim 4429). That sequence explains the later shift away from asset-level debt, although the record does not state this directly.

From late 2024 the pattern becomes coherent and escalating. It began with the $440M convertible (claim 5269), continued with a mid-2025 convertible whose stated $600M size is marked withdrawn (claim 4782) against net proceeds of $535M (claim 4783), then a $1B issuance (claim 4212), and then paired $1B 2032 and 2033 notes (claims 4013, 4014). The ATM capacity also scaled, from $300M (claim 12185) to $1B (claim 5159). The December 2025 package matters most. It used 39,699,102 registered direct shares (claim 18755) to fund repurchases of the 2029 and 2030 notes at $1,024.2M and $608.2M (claims 4019, 4018). In effect, this converted older coupon-bearing debt into immediate dilution plus longer-dated, lower-coupon paper.

Use of proceeds is described mainly in terms of hedging rather than operations. Capped call spend rose from $44M (claim 5273) to $56.7M (claim 4214) to $201M (claim 3990), alongside prepaid forwards of $74M to $92M (claims 5274, 4785). Against this, GPU capital expenditure of $5.8B is disclosed (claim 4160), with only $400M of GPU equipment financing on record (claim 20254). The data contains no evidence of actual deployment, so whether proceeds track stated intent cannot be assessed here. There are also no insider transactions in this record.

Several points require explanation. The 2029 note repurchase appears as both $1,024M and $317M (claims 3992, 18746). The closed 2033 issuance is recorded at $1,150M (claim 18745) versus $1,300M (claim 3988). A reader should watch four things: how the $5.8B GPU commitment is funded beyond current facilities, how much of the $1B ATM has been drawn, how the Microsoft agreement payment terms are sequenced (claim 4163), and whether dilution beyond the capped call cap of $82 (claim 18749) becomes relevant.

IREN's record shows a company that has moved from small equity facilities to repeated large convertible raises, with each step larger than the last. After the IPO's $215.3M in net proceeds (claim #14390), the 2022–2023 period relied on a B. Riley share purchase facility (claim #13623), a committed equity facility cited at $25M across several filings (claims #13325, #12460, #12204), and then a $300M ATM (claim #12185), later replaced by a $1B program (claim #5159). The equipment-financing leg ran into trouble. A receiver was appointed over non-recourse facilities (claim #12447), a court ruled that Bitcoin was not collateral (claim #12270), and a settlement of the subsidiary defaults came only in August 2025 (claim #4431). The balance-sheet strategy after that episode favors unsecured convertibles over asset-backed debt.

The convertible cadence is the defining pattern. There was $440M in December 2024 (claim #5269). The June 2025 deal was announced at $450M–$500M (claims #4796, #4790) and reported $535M in net proceeds (claim #4783), while a $600M issuance entry is marked withdrawn (claim #4782); that sequence requires explanation. October 2025 brought $1B plus a $125M overallotment (claims #4212, #4213). In December 2025, $2B of 2032 and 2033 notes (claims #18752, #18753) funded the repurchase of the 2029 and 2030 notes at $1,024.2M and $608.2M (claims #4019, #4018), alongside a registered direct placement of 39,699,102 shares (claim #18755). This has the look of a coherent program rather than patching. Each deal pairs with capped calls or prepaid forwards (claims #5273, #5274, #4785, #4214, #4017), and the December exchange extended maturities while turning part of the earlier debt into equity now rather than later. The cost is that hedging consumed a meaningful share of proceeds, most visibly $174.8M to $201M in capped calls in December alone (claims #4017, #3990).

On deployment, the stated use shifted from growth and data-center build-out (claim #13625) to GPU capex of $5.8B tied to the Microsoft agreement (claims #4160, #4163). Raised capital does not yet cover that figure, even with $400M of GPU equipment financing (claim #20254). Several December figures also conflict with each other: closing sizes of $1,300M per series (claims #3987, #3988) against $1,150M (claim #18745), and a $228M 2030 repurchase (claim #18747) against $608M (claim #3991). The record contains no insider transactions, and it ends mid-entry at claim #3992.

A careful reader should watch three things. The first is how the remaining GPU funding gap is closed, whether through the ATM, further convertibles, or secured GPU debt. The second is reconciliation of the December closing amounts. The third is whether the pace of roughly doubling issuance size can hold without equity absorbing more of the load.

IREN's financing record describes a company that moved from small, facility-based equity toward large and increasingly frequent convertible issuance. After the 2021 IPO (net proceeds of $215.3M, per claim #14390), 2022 and 2023 relied on a B. Riley committed equity facility that appears in at least nine claims at $25M (claims #13325 through #12203), plus NYDIG equipment financing of $71M (claim #14216). The $300M ATM in September 2023 (claim #12185) and the $1B ATM in January 2025 (claim #5159) moved the equity base onto a larger footing. The convertible sequence then accelerated. It began with $440M in December 2024 (claim #5269), was followed by roughly $550M in June 2025 (claims #4790 and #4791) and $1B in October 2025 (claim #4212), and concluded with two $1B tranches due 2032 and 2033 in December 2025 (claims #18752 and #18753). The intervals shortened as the sizes grew. The mix reads as a deliberate strategy rather than patching. Each convertible carried capped calls or prepaid forwards, for example $40.3M and $75.9M (claims #5305 and #5306) and $174.8M (claim #4017), which indicates sustained attention to managing dilution. The December 2025 transaction repurchased the 2029 and 2030 notes ($1,024.2M and $608.2M, per claims #4019 and #4018) and funded that repurchase with a registered direct offering of ordinary shares (claim #18755). In effect, it converted near-dated debt into equity and longer-dated paper. Use of proceeds requires explanation in two areas. First, the 2022 growth and data center language (claim #13625) became, from 2024 onward, substantial allocations to hedges and refinancing. Second, the $5.8B GPU capital expenditure (claim #4160) sits against only a $400M GPU equipment financing (claim #20254) in the record. The earlier equipment financing enforcement, including the PwC receiver (claim #12447) and the August 2025 settlement (claim #4431), appears resolved. Several figures are internally inconsistent. The $600M June 2025 issuance is marked withdrawn (claim #4782). The 2032 and 2033 tranches appear as $1,300M (claims #3987 and #3988) against $1,150M elsewhere (claim #18745). The 2030 repurchase appears as both $228M (claim #18747) and $608M (claim #3991). No insider transactions are on record. A careful reader should watch four things: draws on the $1B ATM, the share count after the December 2025 direct offering, how the remaining GPU commitment is funded, and reconciled final tranche sizes.

IREN's financing record moves through three distinct phases, and each one is larger than the last. The first phase pairs the IPO with equipment debt, namely the $215.3M in net IPO proceeds (claim #14390) and the $71M NYDIG equipment facility (claim #14216). The second phase draws on committed and at-the-market equity, beginning with the B. Riley facility (claim #13623), moving to a $300M ATM (claim #12185) and then to a $1B ATM (claim #5159). The third phase runs almost entirely through convertibles. Those issues grew from $440M in December 2024 (claim #5269) to roughly $1B in October 2025 (claim #4212) and then to about $2.3B in December 2025 (claims #18744 and #18745). The intervals between raises shortened as the size of each raise increased. The record contains no disclosed ATM drawdowns, so the dilution the company has actually realized under those programs cannot be measured from this data.

The shift in the mix appears coherent rather than opportunistic, but it was also forced by events. The equipment financing ended in lender enforcement, a PwC receiver (claim #12447), litigation (claim #12274) and a settlement in August 2025 (claim #4429). That history accounts for the company's move away from secured debt. Each convertible after that point came paired with capped calls and prepaid forwards (claims #5273, #4785 and #3990). This pattern points to a deliberate practice of managing dilution. In December 2025, the new notes and a registered direct offering of 39.7M shares (claim #18755) funded the repurchase of the 2029 and 2030 notes (claims #4019 and #4018). In effect, the company replaced 3.25–3.50% coupons with 0.25–1.00% coupons, and it paid for part of that exchange with equity it issued immediately.

The announced uses of proceeds match the hedging and repurchase allocations disclosed at closing. The record does not follow the proceeds through to the $5.8B in GPU capital expenditure tied to the Microsoft agreement (claim #4160). Several figures also require explanation. Claims #3987 and #3988 give $1,300M for each 2032 and 2033 tranche, while claims #18744 and #18745 give $1,150M. Claims #18746 and #18747 report repurchase amounts that do not reconcile with the face values in claims #4019 and #4018. A $600M June 2025 issuance is marked as withdrawn (claim #4782). The record contains no insider transactions.

A careful reader should watch three things. The first is how much of the $1B ATM gets drawn. The second is whether the GPU capital expenditure requires further convertible issuance before the 2032 and 2033 notes season. The third is how settlement of the prepaid forwards and capped calls changes the net share count.

The record shows IREN's financing moving through three distinct phases, and the scale grew sharply at each transition. The first phase ran from the IPO through 2023. The IPO produced net proceeds of $215.3M (claim #14390), and the company then relied on the B. Riley committed equity facility (claims #13623 and #13330) alongside the NYDIG equipment financing of $71M (claim #14216). That equipment financing later moved into lender enforcement with a receiver appointed (claims #12462 and #12447), and was resolved only by the settlement recorded in August 2025 (claim #4429). The second phase replaced the committed facility with at-the-market programs of $300M (claim #12185) and then $1B (claim #5159). The third phase is convertible-led, and its cadence is the central feature of the record. Issuance escalated from $440M in December 2024 (claim #5269), to roughly $535M net in June 2025 (claim #4783), to $1.125B in October 2025 (claims #4212 and #4213), to two $1.15B tranches in December 2025 (claims #18744 and #18745), all within twelve months. The mix reads as deliberate rather than patched. Each note sale was paired with capped calls, prepaid forwards, or both (claims #5273, #5274, #4785 and #4214), which suggests a consistent effort to manage the dilution optics of conversion. That consistency comes at a rising hedging cost, which reached $174.8M to $201M in the December transactions (claims #4017 and #3990). The December refinancing requires explanation. A registered direct share placement of $1,632M (claim #18748) funded repurchases of the 2029 and 2030 notes. Those repurchases are stated as $1,024.2M and $608.2M (claims #4019 and #4018), but separately as $317M and $228M (claims #18746 and #18747). The record does not reconcile the two sets of figures. Either way, the transaction converted debt into realized share issuance well ahead of maturity. Use-of-proceeds language remains general, citing growth and data centers (claim #13625), while the GPU capital expenditure commitment of $5.8B (claim #4160) exceeds the disclosed raises. Actual deployment is not evidenced in this record, so whether proceeds tracked their stated uses cannot be assessed. The record contains no insider transactions, and the claim list is truncated after claim #3987, so any later events are not visible here. A careful reader should next watch for three things: how the remaining GPU commitment is funded, whether the $1B ATM is drawn, and whether the several offerings still marked delayed, along with the withdrawn $600M tranche (claim #4782), are reconciled in later filings.

IREN's financing record describes a company that moved from small, patchwork capital to large, structured convertible issuance, with dilution increasingly managed rather than simply accepted. After the IPO's $215.3M of net proceeds (claim 14390), the 2022–2023 period relied on NYDIG equipment financing of $71M (claim 14216) and a B. Riley committed equity facility that appears in the record many times at the $25M level (claims 13330, 12442, 12203). That equipment financing then went into lender enforcement with a receiver appointed (claims 12447, 12462), a court ruling on collateral (claim 12270), and a settlement in 2025 (claim 4429). This was a period of opportunistic patching, and it is reasonable to read the later strategy as a response to it.

From December 2024 the pattern becomes coherent and accelerates. The company raised a $440M convertible (claim 5269), a June 2025 convertible with $535M of net proceeds (claim 4783), and an October issuance of $1B plus a $125M overallotment (claims 4212, 4213). Each raise was upsized from its announced size, and each directed a meaningful slice to capped calls or prepaid forwards, such as $74M and $44M in December 2024 (claims 5274, 5273). These are explicit efforts to limit share-count dilution. Alongside them sits a $1B ATM program (claim 5159), which enlarged the $300M program from 2023 (claim 12185). The record does not show how much of either ATM was drawn.

The December 2025 transaction is the inflection. The company issued $2B of 2032 and 2033 notes (claim 19126) and funded the repurchase of the 2029 and 2030 notes with 39,699,102 registered direct shares (claim 4012). In effect it converted earlier debt into realized dilution now in exchange for lower coupons and longer maturities. Several figures in this transaction require explanation. The 2033 tranche is recorded at $1,300 (claim 3988) and at $1,150 (claim 18745). Repurchase amounts are given as $1,024.2M and $608.2M (claims 4019, 4018) and also as $317 and $228 (claims 18746, 18747), which may reflect cash versus principal, though the record does not say. Capped call cost is $174.8M (claim 4017) versus $201 (claim 3990).

The stated uses of proceeds, growth and data centers (claim 13625), are consistent in direction with the $5.8B of GPU-related capital expenditure (claim 4160) tied to the Microsoft agreement (claim 4163). However, the record contains no deployment detail that would allow proceeds to be matched to spend, and it contains no insider transactions. The record as provided also ends mid-entry after December 2025. A careful reader should watch for four things: disclosed ATM utilization, how the remaining gap between convertible proceeds and the $5.8B capex is funded, the Microsoft payment terms, and reconciliation of the December 2025 figures.

IREN's financing record describes a company whose capital sources changed as its circumstances did. After the IPO, which produced net proceeds of $215.3M per claim #14390, the company leaned on asset-backed debt, the $71M NYDIG equipment facility per claim #14216. That debt ended in lender enforcement and a PwC receivership (claims #12462 and #12447), a court ruling on collateral (claim #12270), and a settlement in August 2025 (claim #4429). Through 2023 the company funded itself with a B. Riley committed equity facility, restated at $25M across many filings (claims #13330 through #12203), and then a $300M ATM per claim #12185. That period looks like continuous small-scale share issuance rather than a planned structure.

From December 2024 the pattern becomes coherent and grows quickly. The company sold convertibles in increasing sizes: $440M per claim #5269, $535M net per claim #4783, and $1.125B per claims #4212 and #4213. Each deal paired capped calls with prepaid forwards to limit dilution on conversion. The prepaid forward allocation went from $75.9M (claim #5306) to $92M (claim #4785). The June 2025 capped-call figure was revised from $49M (claim #4794) to $54M (claim #4784), and the $600M issuance line is marked withdrawn (claim #4782). The December 2025 transaction changes the direction. It paired new 2032 and 2033 notes with a registered direct placement of 39,699,102 shares (claim #3989) to repurchase 2029 and 2030 notes, reported at $1,024.2M and $608.2M per claims #4019 and #4018. This turned contingent dilution into issued shares at the same time the company added about $2B of new convertible debt.

The announced use of proceeds and the actual deployment line up only loosely. Early filings cite growth and data centers (claim #13625). The largest later commitment is $5.8B of GPU capex tied to the Microsoft agreement (claims #4160 and #4163), which is far larger than the proceeds on record. Two sets of figures require explanation. First, the repurchase amounts of $317M and $228M in claims #18746 and #18747 do not reconcile with claims #4019 and #4018. Second, the 2033 notes appear as $1,150M in claim #18745 and $1,300M in claim #3988. No insider transactions appear in the supplied record, and the data ends partway through December 2025.

A careful reader should watch four things. How the GPU capex gap is funded. Whether the $1B ATM (claim #5159) is drawn down. Whether hedge costs, which reached $174.8M per claim #4017, keep growing relative to deal size. How the discrepancies above are resolved in audited filings.

IREN's record describes a company whose financing cadence has accelerated sharply in both size and frequency. Through 2023 it relied on small instruments: B. Riley committed equity facilities recorded repeatedly at $25,000,000 (claims #13330, #12442, #12204) and a $300,000,000 ATM (claim #12185). From December 2024 the pattern changes. Convertible issuance grew from the $440M notes (claim #5269) to $535M of net proceeds in June 2025 (claim #4783), $1,000,000,000 in October (claim #4212), and two December tranches recorded at $1,150 each at close (claims #18744, #18745). Over the same period ATM capacity rose from $1,000,000,000 (claim #5159) to $6,000,000,000 (claim #3420). Dilution capacity has therefore scaled by roughly an order of magnitude in about fifteen months. The mix reads as a deliberate program rather than patching. Each convertible carries capped calls and, in the earlier deals, prepaid forwards (claims #5273, #5274, #4785, #4214), which is a consistent effort to defer or buffer share-count impact. The December 2025 transaction deserves attention. The 2029 and 2030 notes were repurchased at $1,024.2M and $608.2M (claims #4019, #4018), funded by a registered direct placement of 39,699,102 shares (claim #18755). That converts the earlier hedged dilution into immediate dilution while resetting coupons lower. The repurchase figures recorded as $317 and $228 (claims #18746, #18747), and the $1,300 tranche sizes (claims #3987, #3988) against $1,150 closings, require explanation, as does the capped call moving from $174.8M (claim #4017) to $201 (claim #3990). Stated use of proceeds has migrated from general growth and data centers (claim #13625) to a specific GPU program: $5.8B of capex (claim #4160) tied to the Microsoft agreement (claim #4163), with $3.6B of GPU financing (claim #16663) and $9.2B of total funding cited (claim #16669). The record does not yet show deployment against these figures. The early NYDIG equipment facility (claim #14216) ended in lender enforcement and a 2025 settlement (claims #12462, #4431), which offers context for how secured hardware financing has fared here. The record contains no insider transactions, and the final claim (#11696) is truncated. Readers should watch the pace of draws on the $6B ATM, disclosed GPU deployment against the $5.8B capex figure, the Dell parent guarantee (claim #3418), and whether conversion-price headroom holds relative to the capped call levels.

IREN's capital history divides into two distinct regimes, and the break between them is sharp. From the 2021 IPO, which netted $215.3M per claim #14390, through 2024, the company funded itself with a modest equipment facility (the $71M NYDIG arrangement, per claim #14216) and a ladder of equity lines. The B. Riley committed facility (claim #12175) gave way to a $300M ATM (claim #12185), then a $1B ATM (claim #5159), and finally a $6B program (claim #3420). Each authorization is roughly an order of magnitude larger than the last, so the program ceilings should be read as signals of scale rather than as dilution already taken. From December 2024 the center of gravity moved to convertibles. They escalate from $440M (claim #5269), to a June 2025 deal with $535M of net proceeds (claim #4783), where a $600M figure was later withdrawn (claim #4782), to $1.125B in October (claims #4212, #4213), $2.6B across 2032 and 2033 notes in December (claims #3987, #3988), and a $2B offering in May 2026 (claim #1364).

The mix looks more deliberate than opportunistic, with one caveat. Every convertible has diverted a meaningful share of proceeds into capped calls or prepaid forwards, and the December 2025 capped call cost was revised from $174.8M to $201M between claims #4017 and #3990. The December refinancing used a 39.7M-share equity raise (claim #3989) to help retire the 2029 and 2030 notes (claims #3992, #3991), which converts debt-like dilution into realized dilution. The May 2026 filings again reference those same notes and prior capped call terminations (claims #9174, #9175, #9176), which suggests the refinancing cycle may be recurring rather than one-time.

The record supports stated use of proceeds, not actual deployment. The $5.8B GPU capex (claim #4160) and $3.6B GPU financing (claim #16663) tie to the Microsoft agreement, but the dataset contains no spend reconciliation. The legacy equipment financing went through receivership, litigation, and a 2025 settlement (claims #12447, #12270, #4431), and that history requires explanation relative to the current debt stack. No insider transactions are on record, so that dimension cannot be assessed.

A careful reader should watch several things next. The first is the pace of draws on the $6B ATM. The second is whether 2029 and 2030 notes remain outstanding after the May 2026 deal. The third is the dollar terms of the NVIDIA warrants (claim #9288), which are ambiguous as recorded. The fourth is the scope of the parent guarantee of the Dell obligations (claim #3418).

IREN's capital history divides into three regimes, and the break points matter more than the totals. From the 2021 listing through 2023 the company funded itself in small bites: $215.3 million of IPO net proceeds (claim 14390), a $71 million NYDIG equipment facility (claim 14216), and a B. Riley committed equity facility that appears on the record repeatedly at the same $25 million size (claims 13325, 12580, 12442, 12369, 12204) between February and September 2023. Repeated re-announcement of an identically sized facility is the signature of incremental drawdown under constraint rather than a planned program, and it coincides with receivership and lender enforcement against the non-recourse equipment financings (claims 12447, 12462), unresolved until settlement in August 2025 (claims 4429, 4431).

The second regime is scale. The ATM moved from $300 million (claim 12185) to $1 billion (claim 5159) to $6 billion (claims 3420, 1035), with sales-agent compensation falling from 3% (claims 12189, 5160) to 1% (claim 3421) — a terms improvement consistent with real institutional demand rather than accommodation. The convertible ladder escalated on a roughly semiannual cadence: $440 million (claim 5269), then net $535 million (claim 4783), then $1.125 billion (claims 4212, 4213), then twin $1.3 billion tranches (claims 3987, 3988) and $2 billion in May 2026 (claim 1364). Each is paired with capped calls and prepaid forwards, $44 million and $74 million initially (claims 5273, 5274), rising to $201 million (claim 3990). That is a consistent structure, not opportunistic patching.

The third feature is refinancing. Most of the December 2025 issuance retired the 2029 and 2030 notes, $1,024 million and $608 million (claims 3992, 3991), and May 2026 repeats the pattern with terminations of prior capped calls (claim 9176). Net new capital is materially smaller than gross issuance implies.

Stated use of proceeds has migrated from growth and data center (claim 13625) and miner prepayments (claims 12141, 12140) to $5.8 billion of GPU capex under the Microsoft agreement (claims 4160, 4163) — a disclosed change of business, with $9.3 billion secured in eight months (claim 11696).

Watch whether the $6 billion ATM is drawn or merely authorized; the parent guarantee of subsidiary Dell obligations (claim 3418), which moves vendor risk up the structure; and the NVIDIA warrants for $2.1 billion of shares (claim 9288), whose dilution sits outside the convertible stack. The record contains no insider transactions, so no conclusion about insider behavior is available.

The cadence divides cleanly at late 2024. From the 2021 listing through 2023 the company raised in increments that look like working capital patching rather than a plan: $215.3M of IPO net proceeds (claim 14390), a $71M NYDIG equipment facility (claim 14216), a $100M B. Riley share purchase arrangement paid partly in commitment shares (claims 13623, 13624), and then a $25M committed equity facility that reappears in essentially identical form across February, May, June, July and September 2023 (claims 13325, 12580, 12442, 12369, 12203). Repeated re-disclosure of the same small facility while non-recourse equipment financings passed into receivership and lender enforcement (claims 12447, 12462) is the signature of an issuer raising against near-term obligations. The $300M ATM of September 2023 (claim 12185) is the first step change.

From December 2024 onward the mix becomes a repeatable template rather than opportunism: a large convertible, a capped call, and a prepaid forward, executed four times at escalating size — $440M with $44M and $74M respectively (claims 5269, 5273, 5274), then $535M net with $54M and $92M (claims 4783, 4784, 4785), then $1.0B with $56.7M (claims 4212, 4214), then paired $1.3B tranches with $201M of capped calls (claims 3987, 3988, 3990). The June 2025 headline of $600M is marked withdrawn (claim 4782) against a $550M execution; that revision requires explanation. Closing dates are recurrently flagged as delayed (claims 5303, 4792, 4237, 4025), which is mechanical but persistent.

Stated use tracks actual deployment better after the pivot than before. The $5.8B GPU capital commitment tied to the Microsoft agreement (claims 4160, 4163) and $3.6B of contract-linked GPU financing (claim 16663) give the 2026 raises a visible destination, consistent with $9.3B secured in eight months (claim 11696). The qualifier is that a material slice of every convertible never leaves the capital structure, funding share-count management instead.

Watch three things: whether the $6B ATM (claim 3432) is actually drawn or held as optionality; the refinancing loop, since the 2029 and 2030 notes were bought back at $1,024.2M and $608.2M (claims 3992, 3991) roughly a year after issuance, shortening effective duration; and the NVIDIA warrant overhang of $2.1B (claim 9288) alongside the parent guarantee of subsidiary Dell obligations (claim 3418), a recourse posture the 2023 structures did not carry. No insider transactions appear on this record, so nothing can be said there.

The shape of this history is a step function rather than a ramp. From the November 2021 listing, which brought in roughly $215M net to the company (claim 14390), IREN spent nearly two years raising in $25M increments through successive committed equity facilities with B. Riley (claims 13330, 12580, 12442, 12369, 12203), a cadence that reads as working-capital patching against miner purchase obligations sized in the single-digit and low-double-digit millions (claims 12146, 12147, 6378). The 2023 equipment financing distress — receivership at the subsidiary level (claims 12447, 12462), the Supreme Court ruling that Bitcoin was not collateral under the limited-recourse structure (claims 12270, 12274), and a settlement only reached in August 2025 (claims 4429, 4431) — is the pivot. After it, the company stopped borrowing against hardware and started borrowing against its own equity.

That second phase is coherent and escalating: a $300M ATM in September 2023 (claim 12185), a $1B ATM in January 2025 (claim 5159), and convertibles that roughly double each time — $440M (claim 5269), $600M gross with $535M net (claim 4783), $1.125B across the October 2025 issuance and overallotment (claims 4212, 4213), then $2.6B in December 2025 split across 2032 and 2033 tranches (claims 3987, 3988), with the December raise explicitly used to retire the 3.50% 2029 and 3.25% 2030 notes at $1,024M and $608M (claims 3992, 3991). That is refinancing, not incremental funding, and it is the strongest evidence of a deliberate liability-management program rather than opportunism.

Two things merit attention. First, a widening slice of gross proceeds never reaches operations: capped calls and prepaid forwards absorbed about $116M in December 2024 (claims 5305, 5306), $146M in June 2025 (claims 4784, 4785), and $201M in December 2025 (claim 3990). Second, the stated destination has consolidated around the Microsoft contract and $5.8B of GPU capex (claims 4160, 4163), with $3.6B of GPU-specific financing (claim 16663) and $9.2–9.3B described as secured (claims 16669, 11696) — figures that require reconciliation against the $6B ATM shelf (claim 3420) and the $2.6B May 2026 upsize, since shelf capacity is not the same as capital raised.

No insider transactions appear in this record, so the mix cannot be assessed on that axis. The withdrawn $600M issuance line (claim 4782) and the repeated delayed tags on closing dates appear mechanical. What a careful reader should track next is ATM utilization against the $6B authorization, the dilution implied by the NVIDIA warrants (claim 9288), and whether the May 2026 convertible again refinances rather than adds.

The cadence falls into three distinct regimes. Through 2023 the company financed itself in $25 million increments off a rolling series of committed equity facilities (claims 13325, 12526, 12442, 12369, 12204) while its non-recourse equipment financing sat under lender enforcement with a receiver appointed (claim 12447). That is drip issuance at the bottom of the capital structure, sized by necessity rather than choice, and it is the only stretch in this record that reads as patching. The September 2023 pivot to a $300 million ATM (claim 12185) alongside a $100 million facility (claim 12175) marks the transition; the December 2024 $440 million convertible (claim 5269) established the template that has governed everything since.

That template has now been run five times, and its repetition argues for strategy rather than opportunism. Each convertible arrives paired with a capped call and a prepaid forward: $44 million and $74 million in December 2024 (claims 5273, 5274), $54 million and $92 million in June 2025 (claims 4784, 4785), $56.7 million in October 2025 (claim 4214), $201 million in December 2025 (claim 3990). Roughly $356 million of cash has been spent on capped calls alone. Management is paying real money to blunt conversion dilution, which is a considered posture, not improvisation.

Use of proceeds tracks less cleanly. The October 2025 offering itemizes $50 million of $856 million in net proceeds (claims 4241, 4239); the balance is not traced in this record. The 2022 framing of proceeds for growth and data center (claim 13625) is no longer discussed in later disclosures. Against the $5.8 billion GPU program (claim 4160), the $9.2 to $9.3 billion raised (claims 16669, 11696) implies substantial prefunding beyond the single announced use.

Two items require explanation. The 3.50% 2029s and 3.25% 2030s were repurchased at $1,024 million and $608 million in December 2025 (claims 3992, 3991), yet are targeted again in May 2026 (claims 9174, 9175). The $600 million June 2025 issuance was withdrawn (claim 4782) against $535 million net stated the same day.

Watch the draw pace on the $6 billion ATM (claim 3420), where commission compressed from 3% to 1% (claims 12189, 3421); the parent guarantee of subsidiary Dell obligations (claim 3418); and whether the capped call terminations (claim 9176) return cash.

The trajectory here is one of two distinct capital regimes, and the discontinuity matters more than any single instrument. From the 2021 IPO's $215.3M net to the company (claim #14390) through 2023, financing was granular and reactive: a $71M NYDIG equipment facility (claim #14216), a B. Riley share purchase facility with commitment shares attached (claims #13623, #13624), and then a striking repetition of $25M committed equity facilities recurring through February, May, June, July and September 2023 (claims #13330, #12580, #12526, #12442, #12460, #12369, #12203). Serial same-size top-ups at that cadence read as patching rather than plan, and the simultaneous appointment of a receiver over non-recourse equipment financing (claims #12447, #12462), litigated to a ruling that Bitcoin was not collateral (claims #12270, #12274) and settled only in August 2025 (claims #4429, #4431), suggests the prior structure had outrun its ability to service itself.

From December 2024 the mix becomes genuinely coherent. The $440M convertible (claim #5269), the ~$535M net June 2025 issue (claim #4783), the $1.125B October 2025 issue (claims #4212, #4213) and the paired $1.3B 2032 and 2033 tranches (claims #3987, #3988) are all executed on the same template: notes, capped calls, prepaid forward. That is a repeatable playbook, not opportunism. It is also expensive in a way worth naming — $201M of capped call cost in December 2025 alone (claim #3990) — and a large share of gross proceeds never reaches operations, with $92M to a prepaid forward in June 2025 (claim #4785) before any deployment.

Use of proceeds tracks better in the second regime than the first. The $5.8B GPU capital expenditure tied to the Microsoft agreement (claims #4160, #4163) and the $3.6B GPU financing (claim #16663) give the 2026 raises a visible destination, consistent with the $9.2B and $9.3B aggregates claimed (claims #16669, #11696).

Three things merit watching. First, drawdown pace on the $6B ATM (claim #3420), which is the directly dilutive instrument and currently the least visible. Second, the December 2025 repurchases of the 2029 and 2030 notes (claims #3992, #3991) — whether refinancing older paper with new paper becomes a recurring necessity. Third, the expected termination of portions of prior capped calls (claim #9176), which unwinds protection already paid for and requires explanation. Several 2026 items remain marked delayed, including the registered direct share offering (claim #4026).

The record divides cleanly into two regimes. Through 2021-2023 the company funded itself in small, reactive increments: roughly $215M of IPO net proceeds (claim 14390), a $71M NYDIG equipment facility (claim 14216), and then a long string of B. Riley committed equity facilities drawn and re-papered in $25M units (claims 13330, 12580, 12442, 12369, 12203) alongside a $100M facility (claim 13623) and a $300M ATM at a 3% agent commission (claims 12185, 12189). That cadence, combined with the equipment financing default, the PwC receivership (claim 12447) and the eventual court holding that Bitcoin was not collateral (claim 12270), reads as opportunistic patching under stress rather than a designed capital structure. The matter was not finally settled until August 2025 (claims 4429, 4431).

From December 2024 the mix becomes highly systematic. Each convertible issuance follows the same template: size, then capped calls and a prepaid forward funded out of proceeds to blunt conversion dilution, visible at $440M (claim 5269) with $44M and $74M sidecars (claims 5273, 5274), at $550M in June 2025 with $54M and $92M (claims 4784, 4785), at $1.125B in October 2025 with $56.7M of capped calls (claim 4214), and at the paired $1.3B 2032 and 2033 notes with $201M of capped calls (claims 3987, 3988, 3990). Notably, the December 2025 and May 2026 rounds are substantially refinancing rather than new money, repurchasing $1,024M of the 2029s and $608M of the 2030s (claims 3992, 3991) and terminating portions of prior capped calls (claim 9176). The ATM commission falling from 3% to 1% (claim 3421) against a $6B program (claim 3420) is consistent with improved access, not deterioration.

Use of proceeds is mostly stated in general terms; the only concrete deployment anchor is the $5.8B GPU capital expenditure disclosed alongside the Microsoft agreement (claims 4160, 4163). The record contains no drawdown reporting against the $300M or $1B ATMs (claims 12185, 5159), so actual deployment cannot be verified here, and no insider transactions appear at all, which limits any read on alignment.

Watch three things: whether the $6B ATM is drawn or held as optionality against the $9.3B already raised (claim 11696); the dilution overhang from NVIDIA warrants of about $2.1B (claim 9288); and the parent guarantee of subsidiary Dell obligations (claim 3418), which moves structural risk upward. The withdrawn $600M figure (claim 4782) and several delayed closings are routine revisions but merit tracking to final terms.

IREN's record shows dilution capacity rising by roughly an order of magnitude at each step. After an IPO that netted $215.3M (claim #14390), the company relied through 2023 on a B. Riley committed equity facility, which recurs across many filings (claims #13330, #12442, #12203). It then moved to a $300M at-market program (claim #12185), a $1B program (claim #5159), and, in March 2026, a $6B prospectus supplement (claim #3420). Convertible issuance scaled on the same curve: $440M in December 2024 (claim #5269), $1B plus a $125M overallotment in October 2025 (claims #4212, #4213), paired 2032 and 2033 notes recorded at $1.3B each in December 2025 (claims #3987, #3988), and a $2B offering in May 2026 that a later entry records as upsized to $2.6B (claims #9171, #9160).

The mix looks deliberate rather than improvised, and the structure repeats from deal to deal. Almost every convert carries capped calls and, earlier on, prepaid forwards (claims #5273, #4785, #3990), which suggests a standing effort to manage conversion dilution. The December 2025 package retired the 2029 and 2030 notes ($1,024M and $608M, claims #3992, #3991), funded partly by a registered direct equity sale (claim #3989). The May 2026 filings again reference those series and the expected termination of prior capped calls (claims #9174, #9175, #9176). That is a refinancing loop in which new notes and new shares replace older notes, and it pushes maturities out. The one clear break in the record is the NYDIG equipment facility (claim #14216), which moved to lender enforcement and a PwC receivership (claims #12447, #12462) before being settled in 2025 (claim #4429).

For use of proceeds, the record gives stated intentions, such as the GPU capital expenditure of $5.8B tied to the Microsoft agreement (claims #4160, #4163) and $9.3B raised over eight months (claim #11696). It does not show actual deployment, so whether spending tracks those intentions cannot be tested from these data. Some entries also conflict or are ambiguous in their units. The June 2025 offering appears at $450M, $500M and a withdrawn $600M (claims #4796, #4790, #4782).

A careful reader should watch four things: how much of the $6B program is actually drawn, whether the May 2026 upsizing is confirmed at closing, the terms of the NVIDIA warrants and share purchase rights (claims #9288, #11078), and the parent guarantee of the Dell obligations (claim #3418). The record provided ends mid-entry, so later developments are not captured here.

IREN's financing record falls into two distinct regimes. Through 2023 the company relied on its IPO (net proceeds of $215.3M, per claim 14390), a $71M NYDIG equipment facility (claim 14216), and a B. Riley committed equity facility that recurs across many filings (claims 13623, 13330, 12203), supplemented by a $300M ATM (claim 12185). The equipment financing later moved into lender enforcement and receivership (claims 12462, 12447) and a collateral dispute (claim 12270). It was settled only in August 2025 (claim 4429).

From December 2024 the mix shifts decisively to convertibles, and each offering is larger than the last. The first was $440M (claim 5269). The June 2025 deal came in at roughly $500M plus a $50M option, after a $600M figure was withdrawn (claims 4790, 4791, 4782). October 2025 brought $1.125B including overallotment (claims 4212, 4213), and December 2025 paired 2032 and 2033 tranches (claims 4013, 4014). In May 2026 the company launched a $2B 2033 offering, reported as upsized to $2.6B with that status marked delayed (claims 9171, 9160).

Equity capacity scaled in parallel, from a $1B ATM (claim 5159) to a $6B program (claim 3420). The structure is consistent rather than improvised. Nearly every note carries capped calls, and the earlier notes also carry prepaid forwards (claims 5273, 5274, 4784, 4785). The December 2025 transaction used a registered direct of 39,699,102 shares (claim 3989) alongside new notes to repurchase $1,024.2M of 2029 notes and $608.2M of 2030 notes (claims 4019, 4018). This is managed dilution, but conversion exposure is being rolled forward and rebased rather than retired.

On use of proceeds, the disclosures specify allocations to hedges and refinancing precisely. Operational deployment appears only as intent. The $5.8B GPU capital expenditure (claim 4160) and the $9.3B raised over eight months (claim 11696) are not reconciled anywhere in the record. No insider transactions appear in the data provided. Several figures are recorded without a unit scale, and the most recent claims are truncated, so per-share dilution conclusions should be held loosely.

A careful reader should watch several items. First, whether the May 2026 upsizing is confirmed at close. Second, whether the 2029 and 2030 notes referenced again in that offering (claims 9174, 9175) are repurchased with new equity, and how the prior capped calls are unwound (claim 9176). Third, the pace of draws on the $6B ATM and the terms of the NVIDIA warrants (claim 9288). Finally, whether the Microsoft agreement payment terms (claim 4163) begin to reduce reliance on further issuance, which the record does not yet show.

IREN's financing history falls into three phases, and the size of each raise has grown by roughly an order of magnitude from one phase to the next. After the IPO's $215.3 million in net proceeds (claim #14390), the company relied on small tools: the NYDIG equipment facility (claim #14216) and a B. Riley committed equity facility that it repeatedly re-disclosed through 2023 (claims #13623, #12175). That facility was joined by a $300 million ATM (claim #12185). The second phase began with the $440 million convertible (claim #5269) and a $1 billion ATM (claim #5159). The third phase moved to billion-dollar scale: the $1 billion plus $125 million October 2025 notes (claims #4212, #4213), the paired 2032 and 2033 tranches (claims #4013, #4014), a $6 billion ATM (claim #3420), and a May 2026 offering upsized from $2 billion to $2.6 billion (claims #9171, #9160).

The mix looks more deliberate than patched together, though its logic has changed over time. Early convertible proceeds went in large part to capped calls and prepaid forwards, such as the $74 million forward (claim #5274) and the $92 million forward (claim #4785). That suggests management was actively trying to limit near-term share-count pressure. By December 2025, new notes and a registered direct share offering (claim #3989) funded the repurchase of $1,024.2 million of 2029 notes and $608.2 million of 2030 notes (claims #4019, #4018). The May 2026 offering again references those same series and expected capped call terminations (claims #9174, #9175, #9176). In practice, the company is rolling earlier convertibles into longer-dated paper while converting part of the obligation into equity. The equipment-financing enforcement, court ruling, and eventual settlement (claims #12447, #12270, #4429) show a clear turn away from secured asset-level debt.

On use of proceeds, the record states intentions, such as growth and data center build-out (claim #13625) and $5.8 billion of GPU-related capital expenditure (claim #4160). It does not show deployment against those figures, so any link between the two cannot be verified from this data. Two points require explanation. First, the December 2033 and 2032 issuance amounts of $1.3 billion each (claims #3987, #3988) do not reconcile with a $1 billion base plus a $150 million option. Second, capped call costs appear as both $174.8 million and $201 million (claims #4017, #3990). No insider transactions appear in the record, and the data provided ends partway through the May 2026 disclosures.

A careful reader should watch four things: actual draws under the $6 billion ATM; whether the remaining 2029 and 2030 notes are retired with cash or with shares; the dilution implied by the NVIDIA warrants (claim #9288); and the contingent exposure created by the parent guarantee of the Dell obligations (claim #3418).

IREN's record shows a company whose financing has changed in kind as well as in size. The early phase was modest and mixed. There were IPO net proceeds of $215.3M (claim #14390), the NYDIG equipment facility of $71M (claim #14216), and a small B. Riley committed equity facility that recurs across several 2023 filings at $25M (claims #13325, #12460, #12203). The equipment debt did not hold. A receiver was appointed (claim #12447), the collateral was litigated (claim #12270), and a settlement arrived only in August 2025 (claim #4429). After that episode, secured asset-level debt no longer appears in the record.

From late 2023 the mix narrows to two instruments, and each one grows sharply. At-the-market capacity went from $300M (claim #12185) to $1B (claim #5159) to $6B (claim #3420). Convertible issuance followed the same path: $440M (claim #5269), then a June 2025 deal announced at $450M (claim #4796) with net proceeds of $535M (claim #4783), then $1B plus a $125M overallotment in October (claims #4212, #4213), then two $1B tranches in December (claims #4013, #4014). Most of these deals carried capped calls or prepaid forwards (claims #5273, #4214, #4017), which suggests dilution is being managed on purpose rather than accepted passively. The December refinancing repurchased 2029 and 2030 notes for $1,024.2M and $608.2M (claims #4019, #4018). A registered direct equity offering helped fund those repurchases (claim #4012). The pattern is coherent, but refinancing is increasingly carried out through new issuance.

Several inconsistencies require explanation. A $600M June 2025 issuance figure is marked withdrawn (claim #4782). The December tranches later appear at $1.3B each (claims #3987, #3988), above the $1.15B implied by the base amounts plus options (claims #4015, #4016). Capped call cost moved from $174.8M to $201M (claims #4017, #3990). The May 2026 upsizing to $2.6B remains delayed (claim #9160). The same filing again references the 2029 and 2030 notes and the termination of prior capped calls (claims #9174, #9175, #9176).

The record states how proceeds will be used better than it shows how they were deployed. The company discloses $5.8B of GPU capex (claim #4160) and $9.3B raised over eight months (claim #11696). No claim reconciles the two figures. No insider transactions appear in this dataset, so that part of the pattern cannot be assessed.

A careful reader should watch four things. The first is how fast the $6B ATM is drawn. The second is whether the $2.6B upsize closes. The third is what remains outstanding of the 2029 and 2030 notes. The fourth is the terms behind the NVIDIA warrant and share-purchase claims (claims #9288, #11078).

IREN's financing record describes a company that moved from small, constrained patching to a large, deliberate capital machine in about three years. The early period consists mainly of repeated restatements of the same $25 million B. Riley committed equity facility (claims #13330, #12580, #12442, #12369, #12203). Alongside it sat an equipment financing arrangement that ended in lender enforcement and a PwC receivership (claims #12447, #12462), and it took until the settlement noted in claim #4429 to close out. Dilution in that phase was modest in size but defensive in character. The inflection runs through the ATM ladder, from $300 million (claim #12185) to $1 billion (claim #5159) to a $6 billion supplement (claim #3420). It also runs through the convertibles: $440 million in December 2024 (claim #5269), roughly $1 billion in October 2025 (claims #4212, #4213), $2.6 billion across 2032 and 2033 notes (claims #3987, #3988), and a further $2 billion upsized to $2.6 billion in May 2026 (claims #1364, #9160). The company itself cites $9.3 billion raised in eight months (claim #11696).

The mix now looks like a coherent strategy rather than improvisation. Almost every convertible carried capped calls or prepaid forwards (claims #5305, #5306, #4794, #4795, #4017). The December 2025 package repurchased the 2029 and 2030 notes (claims #4019, #4018), funded partly by a registered direct offering (claim #3989). In effect, near-dated, higher-coupon paper was refinanced into 0.25% and 1.00% longer-dated notes. The May 2026 terms again reference the 2029 and 2030 notes and the partial termination of earlier capped calls (claims #9174, #9175, #9176), which suggests the rolling is ongoing. The trade-off is layered, overlapping hedges whose net dilution protection is hard to reconstruct from the claims alone.

On use of proceeds, the stated purposes are clearest for hedging costs. For the underlying capital program, such as the $5.8 billion GPU commitment (claim #4160) and the Microsoft agreement (claim #4163), the record does not show actual deployment, so tracking cannot be confirmed. The withdrawn $600 million figure (claim #4782) against $535 million in net proceeds (claim #4783) requires explanation.

This record contains no insider transactions, and the list appears truncated after claim #9167, so this read is partial. A careful reader should watch four things: the pace of draws on the $6 billion ATM, whether any 2029 and 2030 notes remain outstanding, how the NVIDIA warrants (claim #9288) enter the share count, and the scope of the parent guarantee of Dell obligations (claim #3418).

IREN's record divides into two distinct regimes. Through 2023 the company funded itself in small increments. It relied on a B. Riley committed equity facility repeatedly referenced at $25,000,000 (claims #13330, #12203) and a $300M at-market program (claim #12185). This happened while subsidiary equipment financing was under lender enforcement with a receiver appointed (claim #12447), a matter that was litigated (claim #12270) and not settled until August 2025 (claim #4429). Financing in that period reads as patching around a constrained balance sheet rather than executing a plan. From December 2024 the pattern becomes coherent and much larger. Convertible issuance roughly doubles at each step: $440M (claim #5269), net proceeds of $535M after a $600M figure was withdrawn (claims #4783, #4782), $1.125B including overallotment (claims #4212, #4213), paired $1.3B tranches due 2032 and 2033 (claims #3987, #3988), and a $2B 2033 offering with a $300M option (claims #1364, #1365). The at-market capacity expanded in parallel, from $1B (claim #5159) to $6B (claim #3420). Each convertible is paired with capped calls and prepaid forwards (for example claims #5273, #4785, #3990). The company describes these as repurchase mechanisms (claims #4091, #4092). The effect is that dilution is being deferred and shaped rather than avoided. The December 2025 repurchase of the 2029 and 2030 notes (claims #4018, #4019) was funded partly by a registered direct share offering (claim #4012). The May 2026 filings again reference those same series alongside partial capped-call terminations (claims #9174, #9175, #9176). Together these indicate a rolling refinancing in which older, higher-coupon notes are retired into longer-dated paper. Allocations to hedging are consistent between announcement and pricing, with $92M for the prepaid forward in both claims #4795 and #4785. Broader use of proceeds is harder to verify. Stated purposes moved from growth and data centers (claim #13625) to $5.8B of GPU capital expenditure tied to the Microsoft agreement (claims #4160, #4163), but the record contains no deployment figures against which to test them. No insider transactions appear in the data provided, and the input appears truncated. A careful reader should watch four things. The first is the drawdown pace on the $6B at-market program. The second is whether further repurchases of the 2029 and 2030 notes require additional share issuance. The third is the exercise terms of the NVIDIA warrants (claims #9288, #11078). The fourth is the parent guarantee of subsidiary Dell obligations (claim #3418), which requires explanation as contingent exposure grows.

IREN's record shows a capital structure that has changed character, not just grown. Through 2023 the company relied on small, repeated draws of committed equity with B. Riley, recorded at $25 million across several filings (claims 13325, 12203), and a $300 million at-market program (claim 12185). In the same period, lenders were enforcing on non-recourse equipment financings, with a PwC receiver appointed (claim 12447). The court finding that bitcoin was not collateral for those facilities (claim 12270) and the August 2025 settlement (claim 4431) appear to close that episode. From December 2024 the cadence shifts to convertible notes of rising size: $440 million (claim 5269), about $535 million in net proceeds in June 2025 (claim 4783), $1 billion in October 2025 (claim 4212), paired 2032 and 2033 tranches recorded at $1.3 billion each (claims 3987, 3988), and a 2033 offering upsized from $2 billion to $2.6 billion (claim 9160). At-market ceilings rose alongside, from $300 million to $1 billion (claim 5159) to $6 billion (claim 3420).

The mix looks more deliberate than opportunistic. Each convertible carried capped calls and, in the earlier deals, prepaid forwards sized at $74 million and $92 million (claims 5274, 4785), which limits near-term dilution optics. The December 2025 package used new notes and an equity offering to repurchase the 2029 and 2030 notes, at $1,024.2 million and $608.2 million (claims 4019, 4018). This reads as a rolling refinance into longer, lower-coupon paper. The May 2026 offering again references the 2029 and 2030 notes and the partial termination of prior capped calls (claims 9174, 9175, 9176), so that cycle appears to be repeating.

On use of proceeds, the record itemizes hedging costs precisely but says little about operating deployment. The $5.8 billion GPU capital expenditure (claim 4160) and the Microsoft agreement (claim 4163) are the stated anchor. However, no claim links specific note proceeds to that spend, and the $9.3 billion raised in eight months (claim 11696) cannot be reconciled to deployment from this data. That gap requires explanation. The record contains no insider transactions, and the claim list is truncated after claim 1346.

A careful reader should watch four things: actual draws against the $6 billion at-market program, the closing status of the 2033 upsizing (marked delayed), and the dilution terms of the NVIDIA warrants (claim 9288). The fourth is the scope of the parent guarantee of subsidiary Dell obligations (claim 3418), which moves hardware commitments onto the parent balance sheet.

The record traces two distinct phases. Through 2023 the company relied on small, stopgap sources: a $25 million committed equity facility with B. Riley (claims #12580, #12203), and equipment financing that went into lender enforcement with a receiver appointed (claims #12447, #12462). That dispute ran through a court ruling (claim #12270) and was not settled until August 2025 (claim #4429). The $300 million ATM in September 2023 (claim #12188) marks the pivot. After it, ATM capacity rose to $1 billion (claim #5159) and then to $6 billion (claim #3420).

From December 2024 the convertible market became the main instrument, and each deal grew in size and was usually upsized. The $300 million offering became $440 million (claims #5300, #5269). The June 2025 notes were marketed at $450 million (claim #4796), and a $600 million issuance figure is marked withdrawn (claim #4782), which requires explanation. October 2025 brought $1 billion plus a $125 million overallotment (claims #4212, #4213). December 2025 brought $1 billion each of 2032 and 2033 notes (claims #4013, #4014). May 2026 brought $2.6 billion, upsized from $2 billion (claim #9160). Management's own tally of $9.3 billion raised in eight months (claim #11696) is consistent with this acceleration.

The mix looks deliberate rather than improvised. Nearly every convertible was paired with capped calls or prepaid forwards (claims #5273, #5274, #4017, #1348), which suggests an ongoing effort to limit conversion dilution. The December 2025 repurchase of the 3.50% 2029 notes for $1,024.2 million (claim #4019) stands against an original offering of roughly $450–500 million (claims #4796, #4797). That gap indicates the earlier paper was retired well above par, partly funded by a registered direct share issuance (claim #3989). The dilution has been refinanced rather than avoided.

The stated uses of proceeds are dominated by hedging and liability management. The operational use, $5.8 billion of GPU-related capex tied to the Microsoft agreement (claims #4160, #4163), is not reconciled against specific tranches in the record, and actual deployment cannot be verified from these claims. Several figures carry ambiguous units, including the NVIDIA warrant and purchase-right amounts (claims #9288, #11078). The input also appears truncated, and it contains no insider transaction data, so no read on insider behavior is possible.

A careful reader should track four things. The first is how much of the $6 billion ATM is drawn. The second is the effect of terminating portions of prior capped calls (claim #9176). The third is whether the May 2026 closing, currently flagged delayed (claim #1346), completes as described. The fourth is how the NVIDIA warrants add to the fully diluted share count.

IREN's financing record describes a company that has moved in roughly three years from small, contingent equity lines to repeated multi-billion-dollar convertible issuance, and the scale of each step has grown faster than the prior one. The 2023 base was a $25 million committed equity facility with B. Riley (claim 12203), set against equipment financing under lender enforcement with a receiver appointed (claims 12447, 12462). That dispute was only formally closed by a settlement in August 2025 (claim 4429). ATM capacity then expanded from $300 million (claim 12185) to $1 billion (claim 5159) to $6 billion (claim 3432). Over the same period convertible sizes rose from $440 million (claim 5269) to $1 billion including the overallotment (claims 4212, 4213), then to paired 2032 and 2033 notes of $1 billion each (claims 4013, 4014), and then to an offering upsized from $2 billion to $2.6 billion (claims 9171, 9160).

The mix looks more deliberate than opportunistic. Nearly every convertible carried capped calls or prepaid forwards meant to soften conversion dilution (for example claims 5305, 5306, 4794, 4795). The December 2025 transaction went further: it used a registered direct offering of 39,699,102 shares (claim 3989) alongside new notes to repurchase 2029 and 2030 notes (claims 4019, 4018). That extended maturities while turning some contingent dilution into issued shares. The May 2026 offering again references the 2029 and 2030 notes and the termination of portions of prior capped calls (claims 9174, 9175, 9176), which suggests the refinance-and-rehedge cycle is becoming routine. The record contains some inconsistencies that require explanation. The 2032 and 2033 notes appear at $1,300 each (claims 3987, 3988), which exceeds the $1 billion base plus the $150 million option cited in claims 4015 and 4016. Capped call costs for that deal appear as both $174.8 million (claim 4017) and $201 (claim 3990). The June 2025 issuance is also marked withdrawn (claim 4782), even though net proceeds are separately stated (claim 4783).

The record does not let us test whether announced use of proceeds tracks deployment. Stated uses are mostly hedging allocations, and the only capital-deployment figure is GPU-related capex of $5.8 billion tied to the Microsoft agreement (claims 4160, 4163), with no follow-up on spend. Insider transaction data does not appear in the portion of the record provided, and the claim list is truncated, so no read on insider behaviour is offered.

A careful reader should watch four things. The first is how much of the $6 billion ATM (claim 3420) is actually drawn. The second is the terms and exercise of the NVIDIA warrants and purchase rights (claims 9288, 11078). The third is whether further repurchases of the 2029 and 2030 notes are again funded with ordinary shares. The fourth is whether disclosures begin to reconcile the $9.3 billion raised (claim 11696) with installed capacity.

IREN's record describes a financing pattern that grew by an order of magnitude in under three years. In 2023 the company relied on a $25M committed equity facility (claim #12203), a $300M ATM (claim #12188), and cash on hand for miner purchases, with $64M cited as available reserves (claim #12140). From December 2024 the primary instrument became the convertible note, and each deal grew between launch and close. The first went from $300M (claim #5300) to $440M (claim #5269). The October 2025 deal went from $875M (claim #4261) to $1B plus a $125M overallotment (claims #4212, #4213). The May 2026 deal went from $2B (claim #1364) to a $3.0B close (claim #9124). Alongside these, ATM capacity rose from $300M to $1B (claim #5159) and then to $6B (claim #3420). Potential dilution has therefore been building across several channels at once rather than in single discrete steps.

The mix looks deliberate more than improvised. Nearly every convertible was paired with capped calls, and the earlier ones also carried prepaid forwards. Examples are the $74M forward (claim #5274) and the $201M capped call cost in December 2025 (claim #3990). Those structures indicate the company is actively managing share-count exposure. The December 2025 refinancing also points to planning. The company issued 2032 and 2033 notes of $1.3B each (claims #3987, #3988), repurchased $1,024M of 2029 notes and $608M of 2030 notes (claims #3992, #3991), and ran a concurrent share offering (claim #3989). One item requires explanation: a settlement over subsidiary equipment-financing defaults (claim #4429), which shows at least one earlier layer of the capital stack did not perform as structured.

Hedge allocations track closely from announcement to close, for example the capped call figures in claims #4784 and #4214. Deployment of the general-purpose proceeds cannot be verified from this record. The $5.8B GPU capex figure (claim #4160) and the $9.3B raised over eight months (claim #11696) are both stated. No claims here reconcile one against the other. The use-of-proceeds disclosure for the May 2026 offering (claim #9173) carries no quantified breakdown.

The record contains no insider transactions, and it is truncated after claim #1334, so this read is incomplete. A careful reader should watch four things. First, how much of the $6B ATM is actually drawn. Second, whether the May 2026 offering again retired 2029 and 2030 notes, given that it references them (claims #9174, #9175) and expects prior capped calls to be partly terminated (claim #9176). Third, the dilution terms of the NVIDIA warrants (claim #9288). Fourth, how the parent guarantee of Dell obligations (claim #3418) interacts with the Microsoft payment terms (claim #4163).

IREN's record describes a company whose financing has grown in scale by more than an order of magnitude over three years while remaining essentially one instrument. The 2023 entries are modest: a $25 million committed equity facility with B. Riley (claim 12204) and a $300 million at-the-market program (claim 12185), used around miner prepayments. From December 2024 the convertible note becomes the primary tool. The $440 million issue (claim 5269) is followed by roughly $535 million in net proceeds in June 2025 (claim 4783), a $1.0 billion initial issuance plus a $125 million overallotment in October 2025 (claims 4212, 4213), two $1.0 billion tranches in December 2025 (claims 4013, 4014), and a $3.0 billion close in May 2026 (claim 9124). ATM capacity has scaled on the same curve, from $1 billion (claim 5159) to $6 billion (claim 3420). The cadence has shortened from annual to roughly quarterly, and each step is larger than the last.

The mix reads as a deliberate structure rather than patching. Nearly every convertible carries capped calls, and the earlier deals also carry prepaid forwards (claims 5273, 5274, 4785). Both instruments are designed to soften conversion dilution. The December 2025 transaction goes further. It repurchases the 2029 and 2030 notes for $1,024.2 million and $608.2 million (claims 4019, 4018), funded in part by a registered direct offering tied to 39,699,102 shares (claim 3989). That swaps higher-coupon paper for lower-coupon, longer-dated paper and realizes dilution in shares now. The May 2026 offering again references the 2029 and 2030 notes and the termination of prior capped calls (claims 9174, 9175, 9176), which suggests a second round of the same recycling.

Use of proceeds cannot be checked against deployment from this record. Disclosed allocations cover only hedge costs, such as $201.3 million for capped calls (claim 9129). The $5.8 billion GPU capital expenditure tied to the Microsoft agreement (claims 4160, 4163) is the evident destination, but no claim reconciles spending with the $9.3 billion said to have been raised over eight months (claim 11696). The August 2025 settlement of subsidiary equipment-financing defaults (claim 4429) is not discussed further. Several figures are recorded without consistent units, and claims 3987 and 3988 list $1,300 per tranche against $1,000 plus $150 options, which requires explanation. No insider transactions appear in the record.

A reader should watch for three things: how much of the $6 billion ATM is actually drawn, whether the NVIDIA warrants (claim 9288) are exercised, and whether revenue from the Microsoft contract begins to replace the convertible market as the funding source.

IREN's capital raising pattern reveals a company chasing accelerating capex requirements with escalating convertible issuance, then gradually shifting toward non-dilutive and hybrid structures when primary equity markets appeared constrained. From November 2023 through January 2025, capital needs were modest—$4 million for deferred equipment (claim 6491), $10 million for options (claim 6378)—before pivoting to a $440 million convertible offering in December 2024 (claim 5269). Within weeks, a $1 billion at-the-market program launched (claim 5159), signaling confidence in sustained equity access. The cadence accelerated thereafter: $500-600 million convertibles in June 2025 (claim 4783), $1 billion in October 2025 (claim 4212), and two concurrent $1.3 billion issuances in December (claims 3987, 3988). By March 2026, the company announced $9.3 billion raised in eight months (claim 11696), with a $6 billion ordinary shares prospectus pending (claim 3432).

The instrument mix shows intentional dilution management atop escalating capital needs. Every major convertible offering paired with prepaid forward and capped call hedges—$92-100 million and $49-56 million respectively through mid-2025 (claims 4794-4796, 4213-4214)—which explicitly offset shareholder dilution from conversion. The December 2025 refinancing (retiring $1.632 billion of near-term convertibles while issuing $2.6 billion of longer-dated paper) was a term-out: no deleveraging, but extended maturities. Only in 2026 did the company materially diversify: NVIDIA warrants valued at $2.1 billion (claim 9288), a $1.5 billion delayed draw term loan, and $2.1 billion in senior notes (claims 1264-1265), plus accelerated ATM offerings (claim 3432). This diversification suggests either deliberate capital planning or market resistance to further subordinated issuance.

Use of proceeds diverged from actual deployment through 2025. Proceeds ostensibly flowed to hedging vehicles, not operations, until late-year. GPU capex ($5.8 billion, claim 4160) appeared in late 2025 guidance, but financing remained primarily convertible notes. By mid-2026, customer contracts ($2.8 billion, claim 8098), customer prepayments covering 45 percent of capex (claim 8107), and dedicated facilities (term loan, senior notes, $3.6 billion GPU financing per claim 8560) suggest actual capex execution and diversification away from pure equity.

Watch: restricted cash of $1.72 billion as of June 2026 (claim 1956) against total debt of $7.59 billion (claim 1954) requires explanation, as does whether customer prepayments sustain capex run-rate given $2.1 billion NVIDIA warrant dilution pending.

IREN's capital structure tells a story of escalating dilution driven by infrastructure ambitions that outpaced initial financing readiness. Through late 2024, the company raised modestly via convertibles ($300–440M, per claims 5300–5301), but the cadence accelerated sharply in 2025–2026. The company issued $1.125B in convertibles in October 2025 (claim 4261), then in December 2025 executed a refinancing that issued $2B in 2033 notes and $1B in 2032 notes while simultaneously repurchasing $1.632B of older convertibles due 2029–2030 (claims 4013–4014, 4018–4019). This was followed by an even larger blitz in May 2026, closing $3B in new convertibles (claim 9124) after upsizing a $2B offering (claim 9160). By June 2026, total debt stood at $7.59B against cash of $5.9B (claim 1954).

The instrument mix reveals both intent and strain. Convertibles dominate, but the company has consistently hedged dilution via capped calls (consuming $40–201M per tranche) and prepaid forwards, and in May 2026 pivoted to straight debt—a $1.5B delayed-draw term loan and $2.1B senior notes (claims 1264–1265)—signaling reduced appetite for equity underwriting. Two delayed closings on equity offerings (claims 4012, 4026) and equipment financing settlement agreements in August 2025 (claims 4429, 4431) suggest operational friction.

The announced use of proceeds maps to a $5.8B GPU capital expenditure (claim 4160), consistent with the financing trajectory, but the execution path is revealing. Customer contracts of $2.8B were signed by July 2026 (claim 8098), and management notes customer prepayments covering portions of GPU capex (claim 8107). Separately, the company has secured customer-specific GPU financing ($3.6B noted, claim 8560). This reliance on customer funding to underwrite infrastructure spending is material and distinct from traditional balance-sheet capex.

Sophisticated readers should monitor the maturity ladder—convertibles and term debt mature 2029–2033—and watch whether conversion risks accelerate as the stock moves. The equity overhang from the $6B ATM program (claim 3432) remains available. More pressingly, the tight cash position relative to debt, combined with the apparent dependency on customer prepayments and third-party GPU financing, means any slowdown in customer demand or financing availability becomes a binding constraint. The 2025 equipment defaults suggest operational execution risk remains latent.

IREN's capital structure shifted from equipment-stage financing ($4 million deferred, claim 6491) to systematic GPU infrastructure funding, deploying over $10 billion in convertible debt between late 2024 and May 2026. The dilution trajectory accelerated sharply: initial $300-400 million convertible offerings (claim 5300) expanded to a $3 billion closing (claim 9124) within eighteen months, a ten-fold increase in annual raise velocity. This pattern shows strategic coherence rather than opportunism. Management consistently paired convertible issuances with capped call transactions—initially $40 million (claim 5305), escalating to $201.3 million in May 2026 (claim 1336)—to systematically cap equity dilution. A December 2025 refinancing crystallizes the approach: the company repurchased $1.024 billion of 2029 convertibles and $608.2 million of 2030 notes (claims 4019, 4018), immediately replacing them with $2 billion of lower-coupon 2032 and 2033 instruments (claims 4013, 4014). This reflects active liability management, not distress.

The stated use-of-proceeds narrative underwent material revision in 2026. Through 2025, disclosed allocations were narrow—capped calls and prepaid forwards absorbed most proceeds, masking actual deployment. By mid-2026, GPU infrastructure emerged as the focal point: $5.8 billion in reported procurement (claim 8559), $3.6 billion in new financing secured (claim 8560), and $2.8 billion in customer contracts signed (claim 8098). Microsoft payment terms (claim 4163) and $2.1 billion in NVIDIA warrants (claim 9288) indicate genuine partnerships anchoring the capex cycle.

Sophisticated readers should track three elements. First, conversion mechanics embed material dilution: the May 2026 convertible carries a $32 conversion premium (claim 9127) against capped calls at $110 (claim 9128), implying significant equity assumption. Second, customer concentration has sharpened—Microsoft and NVIDIA partnerships now fund primary capex, creating dependency risk. Third, execution slippages recurred throughout the record (delayed closings on multiple offerings and equity sales), requiring verification that announced capital translates to actual deployment rather than remaining promised but unfunded.

IREN has executed a coherent financing architecture rather than reactive patching. From December 2024 onwards, the company issued escalating convertible tranches—$440M (claim 5269), $450M (claim 4796), and $3B by May 2026 (claim 1334)—each paired with capped call hedges that synthetically cap shareholder dilution. The May 2026 offering layered $3B in convertibles with $201M in capped call costs struck at $110 (claims 1334, 1336, 9128). This structure is deliberate capital architecture, not distress financing. The actual capital deployment tracks announced use closely. IREN authorized a $6B at-market equity program (claim 3420) but has drawn minimally, instead layering convertibles with traditional debt: $1.5B in term loans and $2.1B in senior notes (claims 1264, 1265). The capital is flowing to a clear purpose: $5.8B in GPU procurement (claim 8559) supported by $3.6B in GPU-specific financing (claim 8560) and customer prepayment commitments (claim 8107). This is not general corporate use; it is capex-driven and vendor-financed. A late 2025 refinancing revealed management's longer-term intent. IREN issued $1.3B each of 2032 and 2033 convertibles (claims 3987, 3988) to repurchase $608M of 2030 notes and $1.024B of 2029 notes (claims 3991, 3992). Term extension and repricing discipline, not panic. However, compression in equity pricing signals pressure. The May 2026 conversion premium of 32% (claim 9127) tightened versus prior offerings. If stock approaches conversion prices or exceeds the $110 capped call cap, dilution accelerates sharply. Watch the 2029 maturity wall as near-term obligations require refinancing or balance-sheet deployment; cash on hand stood at $5.9B as of June 2026 (claim 1955), adequate but not abundant relative to capex pace. The presence of NVIDIA warrants and parent guarantees of Dell financing (claims 11078, 1300) embeds supplier concentration. GPU capex closure timing and customer payment schedules require monitoring against stated use of proceeds.

IREN's financing trajectory from late 2024 through mid-2026 reveals a capital strategy refined for massive, technology-dependent capex deployment rather than classic opportunistic financing cycles. The company issued approximately $10.1 billion in convertible notes (per claims 5269, 4796, 4212, 4213, 3987, 3988, 1334), with parallel $2.1 billion senior notes (claim 1265) and structured GPU financing facilities totaling $3.65 billion (claim 8287). The convertible issuance cadence accelerated sharply—from $440 million in December 2024 (claim 5269) to $3 billion by May 2026 (claim 1334)—yet the company employed systematic dilution hedging at each step: capped call transactions totaling $174.8 million in costs by December 2025 (claim 4017) and prepaid forwards functioning as equity collars. This mix is coherent, not opportunistic. Every convertible included embedded hedges, and the December 2025 refinancing—retiring $1.632 billion of 2029 and 2030 convertibles (claims 4018-4019) while issuing $2.6 billion in longer-dated 2032-2033 paper with lower coupons (claims 4013-4014)—reflects disciplined debt maturity management. The company also acquired $30 million in NVIDIA share purchase rights and $2.1 billion in NVIDIA warrants (claims 11078, 9288), expanding financing flexibility but signaling dependency on NVIDIA GPU allocation.

Announced use of proceeds tracks deployment reasonably well. The $5.8 billion GPU procurement (claim 8559) maps to the $5.8 billion capital expenditure target announced in November 2025 (claim 4160). Customer prepayments of approximately $2.8 billion (claim 8098) and structured GPU financing of $3.6 billion (claim 8560) offset balance sheet burn. By June 2026, cash stood at $5.9 billion against $7.59 billion in debt (claims 1955, 1954), suggesting a 1.3x gross-debt-to-cash position—high but serviceable for a capex-stage company with customer prepayment backing.

What requires monitoring: the 2029-2033 convertible maturity wall contains over $8 billion in stated principal; refinancing risk materializes if GPU utilization fails to scale. The $6 billion amended ATM facility (claim 3420) remains largely undrawn, a cushion that depletes quickly in deceleration scenarios. Equipment financing settlement in November 2025 (claims 4429, 4431) showed operational stress already present; the parent's $1.6 billion Dell guarantee (claim 1300) extends balance-sheet exposure to supplier risk. The narrowing spread between cash position and near-term capex commitments means the next two quarters will determine whether customer deployment accelerates or capital constraints begin binding.

IREN's capital structure activity reveals a transition from tactical to structurally incoherent over eighteen months. Through mid-2025, the company followed a recognizable pattern: convertible issuances ($440M closed December 2024, $535M net June 2025, per claims #5269, #4783) paired with standardized dilution hedges—capped calls and prepaid forwards. The August 2025 settlement agreements for equipment financing defaults (claims #4431, #4429), filed despite $2B+ in prior capital proceeds, signaled that deployment was lagging obligations.

The November–December 2025 sequence exposed a structural problem. IREN announced $4B in new convertibles (2032 and 2033 tranches per claims #4029, #4030) while simultaneously repurchasing $1.632B of prior convertibles maturing 2029–2030 (claims #3991, #3992), funded partly through a $39.7M equity offering (claim #3989). This is liability refinancing dressed as growth financing. The company was not expanding its balance sheet but restacking the maturity profile: new instruments carried 0.25%–1.00% coupons versus 3.25%–3.50% on retired debt. The stated use-of-proceeds remained hedging—capped calls and prepaid forwards—not business deployment.

By May 2026, the financing architecture broke. A $3B convertible closed (claim #9124), yet the company simultaneously accessed $1.5B in delayed-draw term loans and $2.1B in investment-grade senior notes (claims #1264, #1265). Convertibles alone no longer sufficed. GPU infrastructure capex, announced at $5.8B (claim #4160) in November 2025, was by August 2026 being financed through investment-grade debt and customer prepayments ($2.8B contracts per claim #8098), not convertibles. The divergence between announced funding sources and actual deployment requires explanation.

A sophisticated reader should monitor three items. First, whether customer prepayments remain available; $2.8B covers less than half of stated capex. Second, the $6B ATM program (claim #3420) remains largely undrawn; heavy utilization would signal return to equity dilution as primary funding. Third, the company's net debt position (gross $7.59B per claim #1954 against $5.9B cash per claim #1955, though $1.72B is restricted per claim #1956) leaves limited optionality if capex remains elevated. The shift from pure convertible financing to mixed instruments in mid-2026 suggests either converter investor hesitancy or deliberate optimization for lower coupons on longer tenure. The mechanism warrants scrutiny.

IREN's capital trajectory over the past 18 months reflects a company systematically pre-funding a hardware build-out cycle. The arc begins with opportunistic convertible issuances in late 2024 ($440M, per claim 5269), accelerates through three major convertible offerings across 2025 totaling roughly $2.4B in gross proceeds (claims 5300, 4790, 4261), then shifts structurally in November-December 2025 toward refinancing lower-coupon instruments ($2B at 0.25%-1.00% due 2032-2033, per claims 4013, 4014) while repurchasing higher-coupon debt ($1.6B of 3.25%-3.50% notes due 2029-2030, per claims 4018, 4019). This mix—convertibles, capped calls positioned as hedges, and a $6B ATM authorization (expanded from $1B, per claim 3432)—suggests neither pure equity issuance strategy nor pure debt. Rather, the company appears to have secured investor appetite for convertibles at declining coupon rates, signaling confidence in equity upside or revenue trajectory.

The stated deployment tells a narrower story than the capital raise itself. Beyond hedging costs ($201M in capped calls on the $3B May convertible, per claim 9129), disclosed use-of-proceeds language is sparse until November 2025, when a $5.8B GPU capital expenditure is announced (claim 4160). By June 30, 2026, the balance sheet carries $5.9B cash and $1.7B restricted cash against $7.6B total debt (claims 1955, 1956, 1954), with customer contracts of $2.8B signed by July (claim 8098) and customer prepayments estimated to cover 45% of GPU capex (claim 8107). This suggests actual capex deployment is customer-order-backed, not speculative.

Two features require explanation. First, multiple convertible closings show delayed status (claims 4792, 4237, 4026, 4025), and equipment financing defaults were resolved via settlement in August 2025 (claims 4429, 4431), which raises questions about execution discipline and supplier relationships during the 2025 ramp. Second, the introduction of straight senior debt ($2.1B, claim 1265) and a term loan facility ($1.5B, claim 1264) in May-June 2026 breaks the convertible pattern and may indicate either rising borrowing costs for convertibles or a judgment that near-term equity dilution is disfavored. The mix of customer prepayment funding, debt leverage, and equity optionality via convertibles suggests a deliberate capital stack for infrastructure deployment, but the delayed closings and prior defaults warrant monitoring whether execution keeps pace with funding commitments.

IREN's financing strategy bifurcates into two phases. From late 2024 through mid-2025, the company raised capital via convertible instruments—$440 million in December 2024, $450 million in June 2025, and $1 billion in October 2025 (claims 5269, 4796, 4212). Each issuance paired with substantial capped call hedges ranging from $44 million to $201 million (claims 5273, 9129) that absorbed capital but limited dilution. Use of proceeds remained largely undisclosed beyond hedging costs. The October offering stated only $50 million of intended deployment despite $856 million in net proceeds (claim 4239), suggesting capital was being held opportunistically rather than deployed.

December 2025 marked a deliberate refinancing pivot. IREN issued $2 billion in new convertibles due 2032 and 2033 (claims 4013, 4014) while repurchasing $1.024 billion of 2029 notes and $608 million of 2030 notes (claims 4019, 4018), shifting toward lower coupon liabilities. A $39.7 million equity offering funded this (claim 3989), with $174.8 million in capped call costs (claim 4017). This was disciplined liability management, not reactive patching.

Spring 2026 reoriented the entire strategy. When GPU capital expenditure of $5.8 billion crystallized (claim 4160), IREN deployed multiple coordinated funding sources: $6 billion from an at-the-money equity draw (claim 1035), a $3 billion convertible (claim 9124), traditional debt totaling $3.6 billion via a $1.5 billion term loan and $2.1 billion senior notes (claims 1264, 1265), and GPU-specific financing at $3.65 billion (claim 8287). Critically, customer prepayments were stated to cover 45 percent of capex (claim 8107), with Microsoft's facility structure reaching 95 percent funding coverage (claim 8278). By June 30, 2026, total debt was $7.59 billion against $5.9 billion cash and $1.72 billion restricted cash (claims 1954, 1955, 1956).

The strategy exhibits high coherence. IREN built liquidity via convertibles and ATM access when capex was undefined, refinanced toward lower coupons once existing debt was sized, then deployed multiple sources once capex crystallized. Risks warrant monitoring: whether capex converts to revenue before debt service accelerates, and whether customer concentration—Microsoft at 95 percent of GPU infrastructure funding—creates renegotiation exposure. Watch covenant compliance on the $2.1 billion senior notes and any signaling hesitation from GPU financing partners.

IREN's financing trajectory from late 2024 through mid-2026 reflects a decisive pivot from mining to GPU infrastructure, executed through a systematic capital structure strategy rather than reactive patching. The company issued convertible notes at escalating scale: $440M (claim 5269, December 2024), $500M (claim 4790, June 2025), $875M plus $125M option (claims 4261–4262, October 2025), then dual $1B tranches (2032 and 2033 maturities, claims 4014–4013, December 2025), culminating in a $3B convertible closing in May 2026 (claim 1334, upsized from $2B). This represents roughly $6.8B in convertible proceeds over eighteen months. The mix shows coherence: convertibles fund growth capex, capped call transactions systematically hedge dilution (claims 4017, 4214, 1336, 9129), and prepaid forwards provide additional downside protection (claims 4795, 4785). In December 2025, the company deployed new issuance to actively refinance higher-coupon notes, repurchasing $608M of 3.25% 2030 notes and $1.024B of 3.50% 2029 notes (claims 4018–4019), extending the maturity wall and reducing near-term coupon burden. This behavior signals forward planning, not distress. Announced GPU capex—$5.8B procurement (claim 8559) with $3.65B investment-grade financing secured (claim 8287)—matches the timing and scale of capital raises, demonstrating deployment coherence. Equity supplements this mix through a $1B ATM program (claim 5159, January 2025) later expanded to $6B (claim 3432, March 2026) and a direct offering (claim 3989, December 2025), providing optionality without forced issuance. By June 2026, the company held $5.9B cash (claim 1955) against $7.59B total debt (claim 1954), alongside $1.72B restricted cash (claim 1956) and customer contracts covering $2.8B (claim 8098). The maturity profile concentrates between 2029 and 2033, requiring refinancing discipline. Sophisticated readers should monitor: whether capped call protection survives if equity dilutes below conversion price thresholds; whether customer prepayments accelerate to shrink the capex funding gap; and whether restricted cash constraints tighten operational flexibility as GPU deployments scale.

IREN's financing narrative bifurcates into two distinct phases: pre-GPU and GPU-era capital deployment. Through 2024, the company raised convertibles ($440M closing in December, per claim 5269) with modest hedging ($76M prepaid forward per claim 5306, $44M capped call per claim 5273), focused on miners and equipment. The pattern began shifting in late 2024 when a $1B ATM program was authorized (claim 5159), signaling willingness to tap equity markets opportunistically. By mid-2025, convertible issuance accelerated markedly—$500M in June (claim 4790) followed by $1.125B in October-November (claims 4212, 4213)—while equipment financing defaults were settled (claims 4429, 4431), suggesting prior capex programs encountered execution friction.

The December 2025 refinancing crystallizes IREN's strategy: $2.6B in new convertibles ($1.3B due 2032 at 0.25%, $1.3B due 2033 at 1.0%, per claims 3987, 3988) replaced $1.6B in older convertibles (2029 and 2030 notes, claims 3992, 3991), extending maturity while locking refinance risk into the outer years. Capped call transactions totaling $201M+ (claim 3990) function as share buybacks, offsetting equity dilution. This is coherent: de-risk near-term maturities, hedge share count, and fund capex without massive equity raises. A $39.7M equity offering in December (claim 4012) remained trivial—not the sign of a dilution-happy management.

The announced use of proceeds tracks GPU capex ($5.8B procurement, claim 8559), with $3.6B in equipment financing secured (claim 8560) and customer prepayments covering roughly 45% (claim 8107). Announced customer contracts of $2.8B (claim 8098) fall short of capex, implying IREN is funding a $3B+ inventory bet on future demand. Cash balances and restricted funds ($5.9B and $1.72B respectively, claim 1955, 1956) suggest borrowings are real but not yet deployed.

A sophisticated reader should monitor three vectors: (1) whether customer order backlog expands to justify $5.8B in procured inventory, as any shortfall forces asset write-downs or extended holding costs; (2) refinancing conditions for the 2029-2030 convertibles ($1.6B notional) now due in three to four years, as deteriorating credit conditions could force costly refinance or equity dilution; (3) whether capped call hedges remain economically effective if share price compresses, potentially forcing unanticipated share issuance to fund buyback obligations.

IREN's capital strategy pivoted from mining equipment to GPU infrastructure, executing an aggressive convertible program that reshaped its balance sheet between late 2024 and mid-2026. The cadence reveals disciplined scaling: initial offerings of $300–440 million (claims 5300, 5301, 5269) evolved into $600–875 million tranches (claims 4796, 4261) before dual $1 billion convertibles in December 2025 (claims 4013, 4014), culminating in a $3 billion offering in May 2026 (claim 9124). Early proceeds financed miner purchases and option downpayments (claims 6491, 6378); by mid-2025, capital shifted decisively to GPU procurement ($5.8 billion, claim 8559) underwritten by layered funding: $3.6 billion in GPU-specific financing (claim 8560), investment-grade debt (claim 8287), and $2.1 billion in NVIDIA warrants (claim 9288).

The instrument mix signals calculated capital-structure management, not opportunistic patching. Capped call and prepaid forward transactions consistently hedged dilution post-issuance—allocations escalated from $40–50 million (claims 5305, 5306) to $174–201 million tranches by late 2025 (claims 4214, 9129). These derivatives cap dilution at specified conversion prices, suggesting management either expects equity valuations to hold or desires protection against downside. The $6 billion ATM program authorized in March 2026 (claim 3420) provided optionality for equity funding at cheaper valuations if market conditions permitted.

Use-of-proceeds tracking shows general alignment between announced intent and deployment, though multiple offering closures were delayed (claims 5303, 4792, 4237, 4025), and equipment financing defaults required settlement in August 2025 (claim 4431). The December 2025 refinancing—repurchasing $1.6 billion of 2029 and 2030 convertibles (claims 4019, 4018) with new 2032 and 2033 tranches (claims 4013, 4014)—extended maturities but increased total debt service without reducing leverage. The company now carries $7.6 billion in convertible debt with maturity concentration 2029–2033.

Sophisticated readers should track: whether $2.8 billion in signed customer contracts (claim 8098) and 45 percent customer prepayment coverage (claim 8107) materialize sufficiently to justify $5.8 billion GPU capex deployment; whether capped-call hedges remain effective if equity valuations decline, since expired calls convert to unhedged dilution; whether subsidiary equipment financing stress (claim 4431) signals operational constraints beyond financing access. The critical question remains whether GPU utilization and revenue scaling will generate sufficient cash to service convertible interest accruals or trigger additional capital raises.

IREN's capital strategy coalesces around a GPU capex cycle beginning in late 2024, marked by deliberate refinancing of older convertible debt into a lower-coupon stack. From December 2024 to May 2026, the company issued $7.6 billion in convertible debt (claims 5269, 4796, 4212-4213, 3987-3988, 9124), drew a $6 billion at-the-market equity program (claims 5159, 3432, 1035), received $2.1 billion in NVIDIA warrants (claim 9288), and secured $3.6 billion in debt facilities (claims 1264-1265, 8287). The centerpiece refinances 3.25% and 3.50% convertibles due 2029-2030 ($1.63 billion repurchased, claims 4018-4019) into 0.25% and 1.0% convertibles maturing 2032-2033 ($2.6 billion issued, claims 3987-3988). This swap reduces interest expense and extends maturity, accepting conversion dilution at 32% premium (claim 9127).

Proceeds deployment separates convertible and capex financing structurally. Convertible offerings allocate 15 to 20 percent of net proceeds to hedging instruments (capped calls, prepaid forwards), leaving the bulk unallocated in disclosure. Meanwhile, external claims document $5.8 billion in GPU procurement (claim 8559), $3.6 billion in GPU financing (claim 8560), and $2.8 billion in customer contracts (claim 8098). Prior equipment financing defaults (claims 4429, 4431, August 2025) suggest piecemeal capex funding preceded integration. By June 30, 2026, IREN held $5.9 billion in unrestricted cash and $1.72 billion restricted against $7.59 billion in total debt (claims 1955-1956, 1954). Customer prepayments cover 45 percent of GPU capex (claim 8107), substantially reducing unhedged exposure.

Conversion dilution and capex continuity warrant monitoring. The $3 billion May 2026 convertible at $110 cap price (claims 9124, 9128) implies roughly 27 million share equivalents upon conversion. The $6 billion ATM fully drawn by April 2026 adds continuous pressure. Watch whether customer-backed capex sustains through 2026-2027, whether restricted-cash covenants tighten as debt approaches $8 billion, and whether stock appreciation triggers conversion of the 0.25%/1.0% stack.

IREN conducted a convertible-debt-led financing campaign from December 2024 through May 2026 that tracks an orderly escalation in GPU infrastructure conviction, though the company's recent history includes a friction point worth noting. The company issued convertible notes in a rising staircase: $440 million in December 2024 (claim 5269), $600 million in June 2025 (claim 4783), $1.125 billion in October 2025 including a $125 million overallotment (claims 4212, 4213), $2.6 billion in December 2025 after upsizing from $1 billion announced per maturity (claims 3987, 3988), and $3 billion in May 2026 (claim 9124). Across six issuances, IREN raised roughly $7.7 billion of convertible debt—equivalent to its entire June 2026 debt stock—and in each offering deployed capped calls consuming 3-7 percent of proceeds (claims 5305, 4794, 4214, 4017, 1336) to mechanically hedge dilution. The company also established a $1 billion ATM program in January 2025 (claim 5159) and expanded it to $6 billion in March 2026 (claim 3432), available but apparently undrawn.

The capital structure choice reveals intentionality. Convertibles front-loaded the sequence, while senior debt arrived late and modest: $2.1 billion in May 2026 (claim 1265) and $1.5 billion delayed-draw (claim 1264). Equity proper stayed minimal: a $39.7 million registered direct in December to fund convertible repurchases (claims 4012, 3989), not a capital-raise. The pattern suggests management expected to service convertible coupons from operations rather than dilute shareholders through equity.

Yet proceeds deployment invites scrutiny. IREN announced $5.8 billion of GPU-related capex in November 2025 (claim 4160), then disclosed in August 2026 $5.8 billion GPU procurement (claim 8559) and $3.6 billion financing secured (claim 8560). The company also guaranteed a subsidiary's $1.6 billion Dell purchase (claim 1300). Customer contracts totaled $2.8 billion by July 2026 (claim 8098)—materially smaller than capex or the guarantee's contingent size. Separately, the company settled equipment financing defaults in August 2025 (claims 4429, 4431), friction arriving only months before the GPU commitment was announced.

A sophisticated reader should monitor three dynamics: whether the $3 billion May 2026 convertible approaches conversion parity, whether customer revenue and stated prepayments (claim 8107) scale to justify the Dell contingency, and how much of the $6 billion ATM actually gets deployed, given an undrawn buffer of that size remains both reassurance and unfinished capital structure work.

IREN's capital-raising arc from late 2023 through August 2026 reflects accelerating financial intensity tied to a stated $5.8 billion GPU procurement announced in November 2025 (claim 4160). The pattern is neither purely opportunistic nor a single coherent master plan, but rather successive rounds of convertible and debt issuances timed to meet capex milestones, with material support from customer prepayments.

Convertible notes comprise the dominant instrument. Starting with $440 million closed in December 2024 (claim 5269), the company escalated to $1.125 billion in October 2025 (claims 4212-4213) and $3 billion in May 2026 (claim 1334). Coupons show a clear refinancing arc: the June 2025 vintage carried approximately 4 percent interest (claim 4786); the May 2026 issuance, 1 percent (claim 9126). In December 2025, IREN issued $2 billion in new 2032-2033 convertible notes (claims 4029-4030) and immediately retired $1.024 billion of 2029 notes and $608 million of 2030 notes at higher rates (claims 4019, 4018), evidence of deliberate refinancing rather than fresh dilution for operations.

Two details complicate the narrative of straightforward capex funding. First, subsidiary equipment-financing defaults requiring settlement agreements preceded the GPU capex announcement by months (August 2025, claims 4429-4431). Second, and more materially, customer prepayments are funding capex directly: $2.8 billion in contracts announced July 2026 (claim 8098), with management noting prepayments cover GPU capital expenditure (claim 8107). This suggests the capital raises are backstopping customer revenue, not replacing it.

Leverage has risen sharply. By June 30, 2026, total debt stood at $7.59 billion against $5.9 billion in cash (claims 1954-1955). In May-August 2026, IREN added $2.1 billion in senior notes (claim 1265), $3.65 billion in investment-grade financing (claim 8287), and a $6 billion ATM equity program (claim 7926), with the ATM substantially undeployed as of August. Capped call transactions—consistently sized at roughly 3-4 percent of convertible issuances (claims 4214, 1336)—show disciplined dilution hedging, but the cumulative effect is debt dominance and rising leverage. The key watch: whether customer prepayments sustain this capital intensity or whether deployment timelines compress faster than proceeds can be drawn.

IREN's capital structure from late 2024 through mid-2026 reflects a deliberate shift from financial engineering toward infrastructure financing, but the earlier phase and transition warrant scrutiny. Through October 2025, the company issued a series of convertible notes—$300–440M in December 2024 (per claim 5300, upsized per claim 5301, then $440M per claim 5269), $450M in June 2025 (claim 4796), and $875M in October 2025 (claim 4261)—with stated uses of proceeds that emphasized prepaid forwards and capped calls ($40–49M per offering, per claims 5305, 4794) rather than business deployment. This pattern suggests opportunistic capital stacking rather than a coherent strategy; the company was locking in convertible capacity ahead of anticipated dilution or equity issuance. By November 2025, the cadence accelerated sharply: dual $1B convertible offerings (claims 4013, 4014) at significantly lower coupons (0.25% and 1.00%) than prior tranches, coupled with repurchase of $1.024B in 3.50% 2029 notes and $608M in 3.25% 2030 notes (claims 4019, 4018). This debt recycling implies either improved credit quality or a capital structure repositioning—neither is self-evident from the data.

The structural inflection occurs in May 2026, when stated capex needs for GPU infrastructure ($5.8B, per claim 8559) became explicit. A $3B senior convertible offering (claim 1334) with net proceeds of $2.96B (claim 1335) is followed immediately by $3.65B in investment-grade financing tied to a Microsoft contract (claim 8287) and a $6B ATM equity facility (claim 7926). By June 2026, the company carried $7.59B in total debt (claim 1954) against $5.9B in unrestricted cash (claim 1955), with $1.72B restricted by creditor controls (claim 1956). Customer prepayments of $45M (per claim 503) and $2.8B in signed contracts (claim 8098) do not materially offset the $5.8B procurement. The divergence between early convertible proceeds—deployed into derivatives rather than capex—and later proceeds, which explicitly fund GPU infrastructure, suggests initial capital raises were exploratory or speculative, only later anchored to a concrete use case. Sophisticated readers should monitor three items: whether the Microsoft contract remains binding and whether payment terms adjust (the $3.65B financing is contract-dependent per claims 1030, 1759); conversion mechanics across the seven convertible tranches outstanding (2029–2033 maturity ladder) and whether stock appreciation triggers dilution that strains customer economics; and whether restricted cash of $1.72B, which is beyond management's discretion, tightens operational flexibility faster than revenue scales. The absence of disclosed revenue figures contemporaneous with capital raises—June 2026 reported $5.9B cash against $7.59B debt, but no corresponding quarterly revenue—requires explanation.

IREN's capital structure narrative shifted from opportunistic equity hedging to strategic capex financing between late 2024 and mid-2026. Through 2025, the company executed a disciplined convertible ladder—$300M in December 2024, scaling to $875M (plus $125M overallotment) by October 2025, then doubling down with simultaneous $2B offerings for 2032 and 2033 maturities in November 2025 (per claims 4029, 4030)—while using capped calls and prepaid forwards to suppress dilution. The December 2025 repurchase of $608M in 2030 convertibles and $1.024B in 2029 convertibles (claims 4018, 4019) signals liability refinancing rather than balance-sheet tightening, consistent with a company deploying capital aggressively elsewhere. By mid-2026, the financing mix fundamentally reoriented: a $6B ATM program (claim 3432) provided ongoing equity flexibility, but GPU capex became the organizing principle. A $6.5B GPU financing facility closed in late August 2026 (claim 2736), while $3.65B in investment-grade debt tied directly to Microsoft contract infrastructure (claim 1759) replaced earlier reliance on convertible issuance. The company disclosed $5.8B in GPU procurement (claim 8559) and $3.6B in GPU financing secured (claim 8560), suggesting deployment is tracking announced capex, though restricted cash of $1.72B at June 30 (claim 1956) indicates funds remain locked or reserved. A parent guarantee of subsidiary Dell obligations worth $1.6B (claim 1300) adds leverage cascading risk. Two settlement agreements for equipment financing defaults in August 2025 (claims 4429, 4431) required attention but lack follow-up disclosure. Going forward, the critical question is whether customer prepayments (claim 8107: $45M as of July) and project-specific financing stay ahead of the capex burn rate, particularly as the ATM program represents incremental dilution capability, and whether the $7.59B debt load (claim 1954) at June 30 becomes constraining given quarterly revenue of $137M (claim 2802).

IREN's capital structure from 2024-2026 reveals a three-stage progression that maps cleanly to operational needs rather than opportunistic patching. The 2024 phase established a convertible foundation—the $300M and $400M offerings (claims 5300, 5301) built flexibility without immediate dilution, with capped calls ($40-44M, claims 5305, 5273) consistently used to cap share creep. That approach persisted through 2025 as the ATM program launched (claim 5159, $1B) and convertible issuance accelerated ($450M upsized to $600M, then $875M+, claims 4796, 4234), each coupled with hedging mechanisms ($49-56.7M capped calls).

The 2026 pivot is where strategic coherence becomes evident. Rather than further convertible escalation, IREN layered in $3.65B investment-grade debt (claim 8287) and $6.5B GPU financing (claim 2736), both collateralized by actual customer commitments—Microsoft ($2.8B in signed contracts, claim 8098) and customer prepayments covering 45% of GPU capex (claim 8107). The $6B ATM offering (claims 1035, 7926) sits mostly dry-powder, suggesting confidence that customer cash flow can fund near-term needs. By June 2026, the balance sheet showed $5.9B cash, $1.72B restricted (claim 1956), against $7.59B debt (claim 1954)—a funded balance, not overleveraged.

Where close reading matters: the company has not revised its stated use of proceeds. The $5.8B GPU procurement (claim 4160) announced in November 2025 remains consistent with actual capital deployed and financing secured. However, revenue ($137M in Q4 2026, claim 2802) against a $7.59B debt stack requires uninterrupted customer growth and contract renewal. The Microsoft agreement payment terms (claim 4163) merit monitoring—if that anchor contract restructures, capital needs shift dramatically. IREN's debt maturity profile across convertibles (2029-2033) and new senior notes (2026 facilities) also needs watching; refinancing risk emerges if growth stalls or customer concentration tightens further. The near-zero EPS (claim 2801) despite large revenue scales signals a capex-heavy period, which is rational given infrastructure buildout, but also signals how narrowly funded the runway sits if deployment slows.

IREN's capital stack evolved deliberately from a convertible-heavy foundation into a GPU-capex financing machine, with three distinct phases marked by strategic coherence rather than opportunistic patching. From late 2024 through mid-2025, the company issued approximately $4 billion in convertible notes across four offerings—$440M in December 2024 (per claim 5269), $600M in June 2025 (per claim 4783 net proceeds), $1.125B in October 2025 (per claims 4212, 4213), and $2B in December 2025 (per claims 4013, 4014 dual issuance). Each offering included substantial hedging via capped calls and prepaid forwards, consuming $200M+ in proceeds; this capital was explicitly allocated to reduce dilution contingent on upside conversion, per claims 4017 and 1336. The pattern shows not reactive fundraising but intentional extension of duration and accumulation of dry powder.

In December 2025, following public announcement of $5.8B GPU capex (claim 4160), IREN executed a refinancing of older convertibles—buying back $608M of 2030 notes and $1.024B of 2029 notes (claims 4018, 4019) and issuing new 0.25%-1.00% notes due 2032-2033 (claims 4013, 4014). This was a clear signal: lock in lower coupons before deployment accelerated the capital structure. Simultaneously, the ATM program was expanded from $1B to $6B (claims 5159, 3420, 3432). By May 2026, the company closed a $3B convertible offering (claim 1334) and layered on $2.1B senior notes plus a $1.5B delayed draw term loan (claims 1265, 1264), backed by explicit $1.6B parent guarantee for Microsoft-related Dell purchases (claim 1300). In August 2026, a $3.65B GPU-specific financing facility closed, explicitly tied to Microsoft contract cash flows (claims 1030, 1759). At June 30, 2026, cash stood at $5.9B with total debt of $7.59B (claims 1955, 1954); restricted cash of $1.72B reflects prepayments and dedicated GPU financing (claim 1956).

The use-of-proceeds narrative tracks deployment: $2.8B in customer contracts have been signed (claim 8098), with customer prepayments covering approximately 45% of the stated GPU capex (claim 8107). Q4 2026 revenue reached $137M (claim 2802), confirming GPU business activation. A sophisticated reader should monitor three vulnerabilities. First, convertible-to-equity conversion at $3B+ nominal value will require either sustained stock performance or future dilutive equity issuance, with the $6B ATM as the release valve. Second, investment-grade debt service on $2.1B notes and $1.5B term loan depends on GPU revenue reaching stated contract rates—the $137M Q4 result implies material margin compression if scaling stalls. Third, the $3.65B GPU financing is implicitly collateralized by Microsoft contract flows; any volume reduction or repricing reverberates through the entire capital structure, which was built for high-velocity deployment.

IREN's capital stack tells the story of a company pivoting from small-scale mining equipment to GPU-scale infrastructure, and the financing arc reflects both strategic coherence and mounting urgency. Between December 2024 and May 2026, IREN completed five major convertible offerings escalating from $440M (claim 5269) to $3B (claim 1334), each substantially larger than its predecessor—a pattern consistent with underestimated capex demands rather than orderly scaling. The convertible instrument choice is deliberate: IREN largely avoided straight equity dilution (one $39.7M offering in December 2025, per claim 3989), instead layering convertibles with capped calls ($175M–201M per issuance, claims 4017, 1336) to purchase optionality and defer the dilution math. The $1B ATM program announced January 2025 (claim 5159), expanded to $6B by March 2026 (claim 3432), functions as a backstop but shows limited utilization in disclosed activity.

Stated use-of-proceeds tracking diverges from deployment. Early convertibles itemize only prepaid forwards and capped call costs (claims 5305–5306), omitting capex detail. December 2025's $39.7M equity offering (claim 3989) was explicitly sized to fund $201.3M in capped call costs (claim 1336), revealing financing-of-financing mechanics. May's $3B convertible issuance (claim 1334) lists no specific allocation. What is unambiguous: the company deployed approximately $6.5B through GPU financing facilities (claims 1328, 1759) backed by $2.8B in Microsoft GPU contracts (claim 8098), with customer prepayments covering $45M of capex (claim 8107). June 2026 cash stood at $5.9B (claim 1955), suggesting rapid burn relative to proceeds raised.

The capital structure reveals both refinancing discipline and potential constraints. December 2025 debt repurchases—$608M of 3.25% convertibles due 2030 and $1.024B of 3.50% convertibles due 2029 (claims 4018, 4019)—indicate coordinated rate reduction, not equity preservation. However, August 2026's non-cash asset impairment (claim 2917) requires explanation, as do prior equipment financing settlements from August 2025 (claims 4429, 4431), which suggest capital allocation discipline may have faltered. A reader should closely monitor whether Microsoft customer concentration, now funding the entire structure, remains exclusive or diversifies.

IREN's capital-market footprint evolved from opportunistic patch-work to coordinated strategy. The early claims—a deferred T21 miner payment in late 2023, equipment option downpayments—read as inherited obligations. The pattern shifted decisively in December 2024 when IREN launched its first major convertible offering ($300M upsized to $440M, per claims 5301, 5269), followed four weeks later by a $1B ATM program (claim 5159), signaling confidence in sustained equity appetite.

What follows reveals deliberate dilution management rather than ad-hoc fundraising. The June 2025 convertible ($500M, per claim 4790) paired capped call transactions ($49M, claim 4794) and prepaid forwards ($92M, claim 4795)—defensive overlays designed to cap shareholder dilution if stock appreciated. The October offering ($875M base plus $125M greenshoe, claims 4234, 4262) repeated the pattern. December 2025 marked a refinancing inflection: IREN issued $2B in new convertibles (claims 4013, 4014) while repurchasing $1.6B of older debt maturing in 2029–2030 (claims 4018, 4019), extending its maturity ladder with smaller capped-call expenses ($201M, claim 4017). May 2026's $3B convertible (claims 1334, 1344) plus senior notes ($2.1B) and term lending ($1.5B, claims 1265, 1264) completed the picture: a graduated mix of equity-linked and straight debt.

The use-of-proceeds narrative tracked deployment. Settlement agreements for equipment-finance defaults (August 2025, claims 4431, 4429) preceded disclosure of Microsoft GPU-service obligations (November 2025, claim 4163). By mid-2026, the capital story crystallized around data-center infrastructure. IREN announced $2.8B in customer contracts (claim 8098) and raised $3.65B debt explicitly tied to a Microsoft contract (claims 8287, 782), with customer prepayments financing 45% of GPU capital expenditure (claim 8107). The $6B ATM equity offering (claims 3432, 7897) in March–August 2026 targeted GPU expansion.

A sophisticated reader should monitor three vectors. First, the $25B FY 2027 capex target (claim 1974) against $5.9B cash and $6.5B closed GPU financing reveals an unmet $13B+ funding gap, likely dependent on customer prepayments holding and asset-backed financing from Microsoft or others maturing. Second, convertible maturity clustering (2029–2033, per recent issuances) creates refinancing risk absent sustained revenue and cash generation—Q4 2026 reported losses (claim 2801). Third, the claimed $36B investment-grade GPU financing (claim 1329) far exceeds announced closures ($6.5B actual, claim 1328), requiring reconciliation of pipeline versus deployed capital.

IREN's financing pattern from November 2023 through August 2026 reflects a deliberate pivot from cheap capital raising to capex-driven debt and equity deployment, anchored to a single customer contract. The first twelve months establish the playbook: convertible notes—$700 million announced across two tranches in December 2023, per claims 5300 and 5301—paired with capped call transactions to limit equity dilution. The $386 million net proceeds from claim 5304, combined with $76 million in prepaid forward protection per claim 5306, suggest management's early confidence that the stock would move higher and convertible conversion could be pushed beyond funded capex needs.

By mid-2025, the strategy clarifies. IREN issued a $1 billion ATM program (claim 5159, January 2025) and continued convertible issuance through June and October, reaching $3 billion in gross convertible debt across the second half of 2025 (claims 4213 and 4212 show $1.125 billion closed in October alone). Critically, the November 2025 announcements reveal a tactical shift: the company repurchased $1.632 billion of older, higher-rate convertibles (3.25% 2030 and 3.50% 2029, per claims 4018 and 4019) and issued $2 billion of new ultra-low-rate convertibles (0.25% 2032 and 1.00% 2033, per claims 4013 and 4014). This is not opportunistic but structural—extending maturity and locking in favorable rates before capex demand forced borrowing.

The capex thesis emerges publicly in November 2025: $5.8 billion in GPU capital expenditure announced (claim 4160). By August 2026, the funding stack is complete: a $6 billion ATM offering (claim 7926), a $3.65 billion loan facility tied to the Microsoft contract (claim 782), customer prepayments of $45 million (claim 503), and $2.8 billion in signed customer contracts (claim 8098). The company now holds $5.9 billion cash against $7.59 billion total debt (claims 1955 and 1954), though restricted cash of $1.72 billion (claim 1956) reduces deployable liquidity by 29 percent. Net proceeds from the May 2026 convertible offering of $2.96 billion (claim 1335) flowed directly into capex and capped call hedges worth $201 million (claim 1336).

Three dynamics require close attention. First, capex intensity vastly exceeds near-term revenue: Q2 2026 revenue of $137 million (claim 2802) annualizes to $548 million against $5.8 billion in announced spend—a multi-decade payback if customer contracts do not ramp materially. The Microsoft contract appears real, but execution delays or scope reduction would cascade through covenant compliance. Second, equity dilution from convertible conversion is stock-price-dependent. A $6 billion ATM authorization combined with $3 billion in convertible issuance creates sensitivity to valuation; underperformance forces involuntary conversion and dilutes existing shareholders. Third, the debt maturity schedule now concentrates at 2032-2033 (two $1 billion tranches, per claims 4013 and 4014), five to six years out; near-term compliance depends entirely on Microsoft revenue materializing and capex deploying on plan.

IREN has employed an escalating convertible notes program since late 2024, raising $700 million by mid-2025, then accelerating to $3 billion by May 2026 (per claims 5300, 4790, 4261, 1334). This reflects a confidence-based capital strategy: low coupons (0.25%-1% by December 2025, per claims 4013–4014) are priced only if management expects stock appreciation above conversion levels. The hedge budget mirrors this calculus—capped calls and prepaid forwards grew from $40–44 million in December 2024 (claims 5305, 5273) to $201 million by May 2026 (claim 1336), showing proportional commitment to protecting against dilution at each stage.

The capital mix pivoted in mid-2026. Traditional debt entered the stack: $2.1 billion senior notes and $1.5 billion delayed-draw term facility (claims 1265, 1264). More notably, August saw $6.5 billion GPU-specific financing (claim 1328) and $3.65 billion tied to Microsoft contract cash flows (claim 1030). This suggests a maturing asset-backed strategy for infrastructure capex, distinct from convertible equity raises. The $6 billion ATM offering authorized March 2026 (claim 3432) has seen minimal deployment—only $39.7 million equity drawn (claim 4012)—indicating management preference for convertibles and secured debt over dilutive common.

Stated use of proceeds tracks deployment with a lag but becomes coherent by mid-2026. Early tranches lacked transparency; by June, $5.8 billion GPU capex was acknowledged (claim 4160), though customer prepayments of $45 billion (claims 8107, 503) are cited without obvious balance-sheet verification. Cash position June 30: $5.9 billion unrestricted, $1.72 billion restricted (claims 1955–1956), with higher figures mentioned elsewhere (claim 366)—discrepancies suggest earmarking for specific obligations.

A material gap remains: stated fiscal 2027 capex of $25 billion-plus (claim 1974) versus $6.5 billion GPU financing plus $5.8 billion deployed implies approximately $13 billion unfinanced. Equipment financing defaults settled in August 2025 (claims 4429, 4431) indicate prior capex financed via subsidiary structures encountered covenant stress. Watch whether Microsoft prepayments truly fund forward capex, whether the ATM offering deploys if convertible markets tighten, and whether additional project-level debt fills the capex gap. Conversion prices relative to stock performance will signal whether holders face imminent dilution or hold in-the-money positions.

IREN's capital structure reveals a coherent three-year transition from equipment financing (miner purchase, 2023–24) to a scaled convertible debt strategy paired with equity hedging, culminating in GPU infrastructure financing. The company issued $440 million in convertible notes in December 2024, per claim 5269, followed by $500 million in June 2025 (claim 4790) and $1.125 billion in October 2025 (claims 4212–4213), establishing a clear cadence: semi-annual convertible tranches, each paired with capped call transactions (totaling $201.3 million in costs by May 2026, per claim 1336) to offset dilution. This is not opportunistic patching; the November 2025 announcement of $5.8 billion GPU capital expenditure (claim 4160) retroactively justifies the convertible accumulation.

The deployed capital tells a clearer story. Rather than spend down the convertible proceeds directly, IREN retired higher-coupon older convertibles—$1.632 billion of 2029 and 2030 notes repurchased in December 2025 (claims 4018–4019)—while issuing new 2032 and 2033 tranches at 0.25–1.00 percent (claims 4013–4014), extending maturity and lowering costs. Simultaneously, the company layered in asset-backed financing: a $3.65 billion GPU-specific loan facility closed in August 2026 (claim 1328), backed by the Microsoft contract, plus $2.4 billion in equipment financing (claim 1951). As of June 30, 2026, the company held $5.9 billion in cash (claim 1955) against $7.59 billion in total debt (claim 1954), with $2.8 billion in customer contracts signed (claim 8098) to justify capex.

A sophisticated reader should track conversion mechanics: the $1.3 billion 2032 and 2033 convertibles (claims 4014, 4030) will dilute shareholders if the stock appreciates past their conversion prices. The capped calls are explicit hedges against this tail risk. Second, monitor the monetization of customer contracts—$2.8 billion in signed commitments must convert to cash flow to service the debt load. Third, the $6 billion At-the-Market program authorized in March 2026 (claim 3432) remains largely unutilized in the data, suggesting either that the company has sufficient funding through convertibles and asset financing, or that equity issuance is held in reserve. Finally, the equipment-backed financing facility ($3.65 billion, claim 1328) is asset-efficient and likely favorable, but it is also the most inflexible source of capital if the contract terms change or customer demand shifts.

IREN scaled convertible issuances from $440 million in December 2024 (claim 5269) to $3 billion by May 2026 (claim 1334), hedging each tranche with capped calls and prepaid forwards costing over $500 million in economic dilution. Through mid-2025, stated use of proceeds cited hedging instruments themselves rather than operational deployment, suggesting opportunistic capital structure arbitrage. Equipment financing defaults materialized in August 2025 (claims 4431, 4429), signaling operational strain beneath the financing narrative. Late 2025 brought tactical clarity: management repurchased $1.6 billion in higher-coupon convertibles (3.50% and 3.25%, claims 4019, 4018) and reissued at 0.25% and 1.00%—a straightforward refinancing. The strategy crystallized in November 2025 around the Microsoft contract (claim 4163) and $5.8 billion GPU capex target (claim 4160). By August 2026, IREN had closed a $6.5 billion GPU financing facility (claim 1328) and announced a $30 billion fiscal 2027 AI capex plan (claim 5437), retrospectively unifying prior issuances into a single investment thesis. As of June 30, 2026, total debt stood at $7.6 billion (claim 1954) against $5.9 billion cash and $1.7 billion restricted cash (claims 1955, 1956). Direct equity dilution remained limited; a $39.7 million equity offering (claim 4012) and the $6 billion ATM program (claim 3432) constituted the primary equity raises, with convertibles and hedges bearing the structural load. A reader should track execution velocity on the $30 billion capex target, debt service against GPU-derived revenue (Q2 2026 revenue was $137 million with a 5% sequential contraction, per claims 2791, 2792), and whether $500 million in capped call premiums—economic dilution paid upfront—prove justified by infrastructure returns. The shift from opportunistic hedged convertible issuance to Microsoft-contract-backed GPU deployment is material; whether it represents strategic foresight or ex-post rationalization will clarify IREN's capital discipline through 2027.

IREN's capital structure shows a deliberate progression from legacy mining finance through an operational pivot to AI infrastructure, though with timing inconsistencies that merit scrutiny. From late 2024 through mid-2025, the company executed a systematic convertible issuance strategy—$440M in December 2024 (per claim 5269), $600M in June 2025 (per claim 4783 confirming net proceeds), and $1.125B in October 2025 (per claims 4212–4213)—paired at each step with capped call transactions and prepaid forwards totaling $170M–$210M per tranche. This hedging intensity suggests management anticipated material share price appreciation and wanted to suppress dilution, though disclosed use of proceeds for these offerings remained vague until 2026.

The November–December 2025 refinancing marked a strategic inflection: IREN repurchased $1.63B of its 2029 and 2030 convertible notes (per claims 4018–4019) while simultaneously issuing $2B in new 2032 and 2033 convertibles (claims 4013–4014), paired with another $200M in capped call costs (claim 4017). This suggests management either faced maturity pressure or took advantage of market receptiveness to extend duration. A single $39.7M equity offering (claims 4012, 3989) provided minimal cushion, indicating reluctance to dilute via common stock.

The capital deployment narrative remained opaque until May 2026. Initial convertible proceeds lacked stated operational uses; by May, IREN disclosed a $5.8B Microsoft GPU commitment (claim 4160) and closed a $3B convertible offering plus $3.65B in secured debt facilities (claims 1334, 1264–1265). The shift to asset-backed financing—$6.5B GPU facility closed (claim 2736) and a $1.6B parent company guarantee of Dell obligations (claim 1300)—demonstrates infrastructure commitment but also tightens credit conditionality. The inconsistency here requires explanation: if Microsoft-contracted GPU demand was clear by May 2026, why were 2024–2025 convertibles not explicitly allocated to that use?

Revenue trends raise a second concern. Q2 2026 showed $137M sequentially with a 5% decline (claim 2792), yet the company announced a $25B–$30B AI capex plan for fiscal 2027 (claims 5437, 1974). Capped call transaction costs now exceed $200M per offering cycle (claim 1336)—a substantial expense that hedges dilution but reflects management's anxiety about conversion mechanics. A sophisticated reader should track whether actual share conversions materialize, whether secured facilities extend as capex matures, whether customer prepayments (currently $45M, per claim 503) scale with deployment, and whether the parent guarantee of Dell obligations introduces contingent credit risk if that contract underperforms.

IREN's capital strategy evolved from opportunistic piecemeal financing into a coherent infrastructure play, though execution risk remains material. Through mid-2025, the company raised convertible debt ($440M in December 2024, per claim 5269; $450M due 2029 in June 2025, per claim 4796) while systematically hedging conversion risk through capped calls and prepaid forwards—typically consuming $40-50M of each offering's proceeds and signaling management uncertainty about share trajectory. This pattern reversed sharply in November-December 2025. The company issued $1B of 0.25% 2032 convertibles and $1B of 1.00% 2033 convertibles (claims 4013, 4014)—extraordinarily cheap financing—then immediately repurchased $1.024B of higher-coupon 2029 convertibles and $608M of 2030 convertibles (claims 4019, 4018), funded by a $39.7M direct equity offering (claim 3989). This refinancing move implies either genuine profitability visibility or collateral strength sufficient to command single-digit coupons.

The 2026 capital stack reflects calculated layering. A $6B ATM program authorized March 2026 and reportedly drawn by April (claims 3420, 1035) was paired with a $3B convertible offering in May (claim 1334), $2.1B senior notes plus $1.5B delayed-draw term loan (claims 1265, 1264), and a $3.65B investment-grade loan tied to the Microsoft GPU contract (claim 782). Customer prepayments of $45M appear in August (claim 503), alongside a $6.5B equipment financing facility close (claim 2736). Announced GPU capex escalated from $5.8B in November 2025 (claim 4160) to $30B for fiscal 2027 (claim 5437), yet stated funding for infrastructure expansion totals $22B (claim 5440)—a gap requiring either phased deployment or material customer prepayment reliance.

Dilution has been partially blunted: capped calls and prepaid forwards reduced net proceeds from each convertible offering, and the aggressive 2029-2030 repurchase consolidated remaining obligations into lower-coupon 2032-2033 notes. However, the $6B equity raise represents substantial primary dilution. Warranting scrutiny: equipment financing defaults settled in August 2025 (claims 4429, 4431) predate this confidence; repeated offer delays through 2025 (claims 4792, 4237, 1346) suggest execution friction; a non-cash impairment noted August 2026 (claim 2917) coincides with peak financing activity; and restricted cash tied to Microsoft covenants ($1.72B, per claim 1956) indicates narrow operating margins before covenant breach. The company has access to capital but limited margin for capex miss or revenue delay.

IREN's financing arc from late 2024 through mid-2026 shows a transition from speculative capital markets arbitrage to contract-backed infrastructure financing. Through 2024 and into 2025, the company ran a disciplined convertible strategy, raising approximately $3.5 billion across three offerings (per claims 5269, 4796, 4782) with aggressive hedging via prepaid forwards and capped calls. The tactic was transparent: convert debt-like instruments into equity while locking downside via derivatives, compressing dilution. By December 2025, management began refinancing this early issuance, repurchasing $1.63 billion of 3.25-3.50 percent convertibles (claims 4018, 4019) and replacing them with 0.25-1.0 percent 2032 and 2033 notes (claims 4013, 4014). This is disciplined deleveraging or market-timing—not opportunistic patching.

The August 2025 equipment financing defaults (claims 4429, 4431) appear to have marked an inflection. By spring 2026, IREN announced a $6 billion at-the-market equity offering (claim 7926) alongside a $3 billion convertible (claim 1334) and $2.1 billion in senior notes (claim 1265), plus a $1.5 billion delayed draw term loan (claim 1264). This was no longer convertible arbitrage; it was balance-sheet building for capex. The subsequent $3.65 billion loan facility explicitly tied to a Microsoft GPU contract (claim 1030) and the $6.5 billion GPU financing facility closure (claim 1328) confirm the pivot: raised capital now flows into infrastructure, not financial engineering. Stated use of proceeds in 2024-2025 convertible offerings (prepaid forwards, capped calls totaling roughly $100-200 million per deal) eventually supported the $5.8 billion GPU capex announced in November 2025 (claim 4160), though the routing is complex.

By June 2026, IREN's balance sheet showed $5.9 billion in cash (claim 1955) against $7.59 billion in total debt (claim 1954), with $1.72 billion restricted (claim 1956). The disclosed capex plan of $25-30 billion for fiscal 2027 (claims 1974, 5437) is material relative to current cash and raises the question whether the $36 billion investment-grade facility ceiling (claim 1329) will be tapped. Capped call costs have escalated materially—from $40 million in December 2024 to $201.3 million on the May 2026 convertible (claim 1336)—reflecting the mechanical drag of dilution hedging as capital needs grow. The refinancing of higher-coupon convertibles into lower-coupon ones was well-timed; the refinancing execution and contract-backed debt raises demonstrate planning rather than crisis. What to monitor: whether customer prepayments (currently $45 million, claim 503) accelerate meaningfully, how quickly the available $36 billion facility gets drawn, and whether the maturity ladder—now spanning 2029 through 2033 on convertibles alone—creates refinancing pressure if capex returns remain lumpy.

Over eighteen months from late 2023 through mid-2026, IREN executed a disciplined capital strategy designed to fund GPU-intensive AI infrastructure. The financing mix and timing reveal two phases: defensive refinancing through mid-2025, followed by contract-backed buildout from late 2025 onward.

The convertible trajectory charts this evolution. December 2024 offerings initially sized at $300M, then $400M (claims 5300–5301), ultimately closed at $440M (claim 5269). The pace accelerated in 2025: June offering of $500M plus $50M option (claims 4790–4791), October issuance of $875M plus $125M overallotment executed as $1.125B (claims 4261–4262, 4212–4213), then November-December refinancing retiring higher-coupon 2029/2030 notes ($1.632B aggregate, per claims 4019–4018) with ultralow-coupon issuances at 0.25% and 1.00% (claims 4013–4014). May 2026 brought a $3B convertible (claim 1334) with $2.96B net proceeds (claim 1335). The arc—from $440M to $3B—reflects methodical capital raising, not panic.

The deployment requirement surfaced in November 2025: $5.8B GPU capex plan (claim 4160), later expanded to $25–30B for fiscal 2027 (claims 1974, 5437). This coincided with a $1B ATM program (claim 5159) amplified to $6B in March 2026 (claims 3420–3431), with $6B deployed by April (claim 1035). A $3.65B GPU-specific secured loan closed in August 2026 (claim 1030). The match between capital raised and stated capex illustrates strategic coherence. Convertibles provide predictable dilution; ATM enables opportunistic equity issuance; secured lending anchors GPU-related contract revenue.

Yet gaps warrant scrutiny. Equipment financing defaults settled in August 2025 (claims 4429, 4431) required a $1.6B corporate guarantee for subsidiary Dell obligations (claim 1300). Revenue at June 2026 was $137M (claim 2791), while capex targets exceed $25B—entirely dependent on Microsoft contract ramp and customer prepayments. Many convertible offerings lack stated use-of-proceeds, and prepaid forward and capped call transactions totaling $200M+ (claims 1336, 3990) raise questions about management conviction on valuation. Restricted cash of $1.72B (claim 1956) signals heavy collateralization. Multiple delayed closings on convertibles and equity offerings suggest execution friction.

A sophisticated reader should monitor three indicators: Microsoft contract milestones and actual capex realization; debt service sustainability on $7.59B total debt (claim 1954) against near-term revenue; and the pace of convertible conversion if stock appreciates, which would accelerate dilution beyond ATM issuance alone. The financing architecture is coherent but entirely contingent on successful infrastructure deployment.

IREN's capital strategy from April through September 2026 reveals a disciplined, asset-backed funding model that eschews equity dilution in favor of project-level financing anchored to customer contracts. The company raised an ATM offering of $6 million in April (claim 1035), then pivoted decisively to the capital markets with a $3 billion convertible senior notes issuance in May (claim 1334), netting $2.96 billion after $201.3 million in capped call hedging costs (claims 1335, 1336). This convertible, followed immediately by $2.1 billion in senior notes (claim 1265) and a $1.6 billion guarantee on Dell purchase obligations (claim 1300), established a financing foundation while Q2 cash reserves reached $5.9 billion (claim 1955).

The true inflection arrived in late August with a $6.5 billion GPU financing facility (claim 2736), structured as three tranches: $3.6 billion for the Microsoft contract, $2.4 billion for non-investment-grade customers, and $2.8 billion for other deployments (claims 1949, 1950, 1951). Customer prepayments covered 96 percent of Microsoft GPU capex (claim 2030), signaling strong contract-level demand capture. This mix—convertible plus asset-backed equipment loans—appears coherent rather than opportunistic; the capital instruments track a capex-intensive, contract-secured business model rather than a company piecing together survival funding.

However, deployment tracking reveals stress. The $25 billion FY2027 capex plan (claim 1974) against $19 billion in secured funding over the past twelve months (claim 2003) leaves an $8 billion gap (claim 1999), with GPU financing now representing 90 percent of total debt (claim 2824). Q2 2026 revenue stood at $137 million (claim 2791) with an operating loss (claim 2790), meaning the company is in early capex deployment phase, not yet generating cash returns. The $2 billion in Microsoft-related restricted cash (claim 2830) and $1.72 billion in restricted funds more broadly (claim 1956) confirm funds are locked into performance obligations rather than freely available.

Insider activity warrants attention: July 2025 saw $115.7 million in insider sales (7 transactions), but July 2026 shifted to insider buys (9 transactions), potentially signaling conviction post-Q2 results. Sophisticated readers should monitor whether the remaining $8 billion gap closes via equity (dilution reversal) or more project debt, the pace of GPU deployment against capex guidance, and whether revenue accelerates fast enough to service a $7.59 billion debt load (claim 1954) built ahead of revenue maturity.

IREN's signed-customer record is far thinner before November 2025 than its announcement volume suggests. Most early named counterparties are suppliers or service providers, not customers: Bitmain hardware contracts (14392, 6374, 6152), BC Hydro power (12182, 12198), Kraken for liquidation (12183), and WEKA as a software partner (12005). The first named compute customer is Poolside AI (6285), which IREN described as a GPU cloud service commencement without disclosing value. Through 2025, the AI pipeline was described in pre-contract language. Horizon 1 and Sweetwater were framed as due diligence, pricing negotiations, letters of intent and 'multiple potential customers' (31515, 31448, 29802, 29819, 31516). Prince George B200 interest was described only as 'engagement' (31445). The $28m run-rate came from white-labelled compute for unnamed 'leading US AI cloud providers' (29810, 29824, 28325). The Morgan Stanley termination (31471) signals that the Sweetwater monetisation path was still unresolved. The Microsoft agreement changed the picture. It is a signed, filed contract with a named counterparty: $9.7bn over five years with a 20% prepayment (20307, 4157, 20245, 1266). IREN then named Together AI, Fluidstack and Fireworks AI (20247), and later Perplexity, Figure AI, Fal AI and Hume AI (8101). The latter list still includes an unidentified 'new leading AI developer', and the July $2.8bn tranche (360) does not name its counterparties in the quoted record. The NVIDIA $3.4bn contract (11076, 9180) is signed. The accompanying 5GW 'strategic partnership' (11077, 9289) is framework language, not a revenue commitment. Diligence should also note the circularity: NVIDIA is simultaneously a customer, a $2.1bn investor whose rights vest on GPU deployment (9179), and the upstream GPU supplier routed through Dell (4161, 1297). The arithmetic broadly reconciles. $9.7bn over five years gives the stated ~$1.9bn ARR (8996), and $3.4bn over five years gives ~$0.68bn. Adding Prince George's ~$0.4bn (16665) brings the total to ~$3.0bn, leaving roughly $0.1bn for smaller contracts against the $3.1bn ARR claim (9183). The >$0.5bn British Columbia figure was still under negotiation (16666) and should not be counted. Concentration is acute: Microsoft alone is about 60% of contracted ARR, and Microsoft plus NVIDIA is above 80%. Dell purchase obligations of $5.8bn plus $1.6bn, carrying a parent guarantee (4164), absorb a large share of contract value up front.

IREN's customer record splits into a small set of signed, named contracts and a larger body of framework, negotiation and unidentified-counterparty language. The signed, named core is clear. The Microsoft GPU services agreement dated November 2, 2025 is documented at the subsidiary level, with a $9.7 billion value, a five-year average term and a 20% prepayment (claims 20307, 4157, 20245, 1266). The $3.4 billion five-year NVIDIA AI Cloud contract was signed in May 2026 (claims 11076, 9180, 9287), although later filings recast it as a "managed services" contract (claims 8995, 1299). Together AI, Fluidstack and Fireworks AI are named as multi-year contract counterparties (claim 20247). Perplexity, Figure AI, Fal AI and Hume AI later appear in a customer roster rather than in contract announcements (claim 8101). Dell (claims 4161, 1297) and Bitmain (claims 14392, 6152) are suppliers, not revenue counterparties, and WEKA was a technology partner (claim 12005).

Much of the rest is either softer than a contract or unnamed. The Horizon 1 workstreams were never more than due diligence, LOIs and negotiations with unidentified parties (claims 31515, 31448, 29802). The $28 million white-labelled run-rate came from "leading US AI cloud providers" who were never named (claims 29824, 28325). The British Columbia ">$0.5bn ARR" is explicitly still under negotiation (claim 16666). The NVIDIA 5GW "strategic partnership" is a framework with no stated economics (claims 11077, 9289). The $2.8 billion of July 2026 contracts is attributed only to "leading AI developers" (claim 360), and the roster includes "a new leading AI developer" with no name (claim 8101).

Concentration has moved from fragmented, unnamed GPU cloud users to two dominant counterparties. Microsoft's expected $1.9 billion of annual revenue (claims 20245, 8996) is roughly 61% of the $3.1 billion of ARR under contract (claim 9183). Adding NVIDIA's implied ~$0.68 billion per year takes the two to about 83%. NVIDIA's position is also circular: it is a customer, a supplier-aligned partner, and holder of $2.1 billion of investment rights that vest only as NVIDIA infrastructure is deployed (claim 9179).

The arithmetic broadly reconciles. Microsoft at about $1.9 billion, NVIDIA at about $0.68 billion and Prince George at about $0.4 billion (claim 16665) sum to roughly $2.98 billion. That leaves about $0.1 to $0.2 billion for the named neocloud and AI-native contracts, which is plausible against the $3.1 billion figure. The July $2.8 billion postdates that ARR figure and has no disclosed term, so its annual contribution cannot be checked. The weighted-average-term disclosure is truncated in the record (claim 8108).

IREN's disclosure record falls into three phases with very different evidentiary weight. Before late 2025, the named counterparties were almost entirely suppliers and service providers, not revenue customers: Bitmain for miners (14392, 6374, 6152), BC Hydro for power (12198, 12182), and Kraken for liquidation (12183). The early AI pivot named WEKA as a partner (12005) and Poolside AI as a GPU cloud customer (6285), but neither came with a contract value. Through most of 2025, the Horizon 1 and Sweetwater narrative relied on pre-contract language. That language included technical due diligence, letters of intent, negotiations with multiple potential customers, and single-tenant opportunities that continue to progress (31515, 31516, 31448, 29802, 29819), none of them identified. The $28 million run-rate from white-labelled compute was attributed only to leading US AI cloud providers, who were never named (29810, 29824, 28325). The Morgan Stanley termination on Sweetwater (31471) shows that a sale process did not close. The record changes with the signed Microsoft contract of 2 November 2025, which names the counterparty and specifies the contracting subsidiary, Childress tranches, a five-year term and 20% prepayment (20307, 4157, 20245, 1266). The NVIDIA $3.4 billion five-year contract followed in May 2026 (11103, 9180, 9287). Named smaller customers then appeared: Together AI, Fluidstack and Fireworks AI (20247), and later Perplexity, Figure AI, Fal AI and Hume AI (8101). Two items should be read separately from the contract figures. The 5GW NVIDIA strategic partnership is framework language (11077, 9289). NVIDIA's $2.1 billion investment vests only as deployments occur (9179), so NVIDIA is customer, investor and architecture sponsor at once. The July 2026 $2.8 billion of contracts (360) is truncated in the record, and the related new leading AI developer (8101) is unnamed. Concentration is acute. Microsoft represents roughly 61% of the roughly $15.9 billion of announced contract value, and Microsoft plus NVIDIA exceeded 80% before July. The arithmetic largely reconciles. $9.7 billion over five years implies about $1.94 billion a year, consistent with the stated $1.9 billion (20245, 8996). NVIDIA's $3.4 billion adds about $0.68 billion, and Prince George adds about $0.4 billion (16665). Together these come to about $3.0 billion against the stated $3.1 billion under contract (9183). That leaves little room for the further $0.5 billion in British Columbia, which was still under negotiation (16666). The Dell agreements are capital commitments, not revenue (4161, 20308, 1297, 8992). They total $5.8 billion against the $9.7 billion Microsoft contract and $1.6 billion against the $3.4 billion NVIDIA contract, and IREN's parent guarantee adds balance-sheet exposure (4164).

IREN's record splits cleanly into two eras, and the line between signed customer contracts and softer language falls almost exactly at November 2025. Before then, most named counterparties were suppliers, not buyers. Bitmain (14392, 6374, 6152), BC Hydro (12182, 12198), Kraken (12183), WEKA (12005) and later Dell (4161, 1297) are hardware, power, liquidity or software relationships. The only named AI customer in that period was Poolside AI (6285). Through mid-2025, the Horizon 1 and Sweetwater pipeline was described only in pre-contract terms: technical due diligence, pricing negotiations and letters of intent (31515), 'colocation negotiations' with 'multiple potential customers' (31448), and 'several customers undergoing due diligence' (29819, 29802). The $28 million run-rate came from white-labelled compute for 'leading US AI cloud providers' who were never identified (29810, 29824, 28325).

The Microsoft agreement is the first large, plainly signed and fully specified customer contract. It carries $9.7 billion over five years, tranche-based GPU Services at Childress, and a 20% prepayment (20307, 4157, 20245, 1266). Together AI, Fluidstack and Fireworks AI followed as named multi-year contracts (20247). The NVIDIA $3.4 billion five-year contract (11076, 9287, 9180) is also stated as signed. Readers should separate it from the '5GW strategic partnership' (11077, 9289), which is framework language with no committed volume. NVIDIA's $2.1 billion investment right also vests only as NVIDIA infrastructure is deployed (9179), so the same party is customer, supplier ecosystem and prospective investor. The later customer roster (8101) adds Perplexity, Figure AI, Fal AI and Hume AI, but also 'a new leading AI developer' left unnamed. The July $2.8 billion tranche (360) is not tied to identified counterparties in the record.

Concentration is high. Microsoft's expected $1.9 billion annual revenue (20245, 8996) is roughly 61% of the $3.1 billion ARR under contract (9183). Adding NVIDIA at about $0.68 billion a year brings the two counterparties to more than 80%.

The arithmetic broadly reconciles. $9.7 billion over five years is about $1.94 billion a year, and $3.4 billion over five years is about $0.68 billion. Adding the roughly $0.4 billion contracted at Prince George (16665) gives about $3.0 billion, with the remainder plausibly from smaller cloud customers. The '>$0.5 billion ARR' in British Columbia (16666) was explicitly still under negotiation and should not be counted. The roughly 4-year weighted average term (8108) is shorter than the five-year headline deals, which implies shorter-dated smaller contracts. Capex commitments to Dell, of $5.8 billion and $1.6 billion, absorb a large share of contract value.

IREN's record divides cleanly into signed agreements with named counterparties and a looser layer of negotiation and partnership language. The signed core is substantial. It includes the Microsoft GPU services contract, a five-year, $9.7 billion agreement dated November 2, 2025 (claims 20307, 4157, 1266), and a five-year, $3.4 billion AI Cloud contract with NVIDIA (claims 9180, 11076). Both are backed by matching Dell hardware purchase agreements of roughly $5.8 billion and $1.6 billion (claims 4161, 20308, 1297, 8992). Named multi-year contracts with Together AI, Fluidstack and Fireworks AI (claim 20247) and the earlier Poolside service commencement (claim 6285) are also identifiable. Earlier Bitmain, BC Hydro and Kraken items (claims 14392, 6152, 12198, 12183) are supplier and vendor relationships, not customers. The WEKA 'partnering' (claim 12005) is a technology alliance.

The unidentified layer is persistent. The $28 million run-rate came from unnamed 'leading US AI cloud providers' (claims 29810, 29824, 28325). Horizon 1 and Sweetwater were described for over a year only as LOIs, due diligence and 'multiple potential customers' (claims 31515, 31448, 29802, 31516). The ~$0.4 billion Prince George ARR 'under contract' and more than $0.5 billion in British Columbia negotiations carry no counterparty names (claims 16665, 16666). The July 2026 customer list ends with 'a new leading AI developer' (claim 8101). The quoted text of the $2.8 billion July announcement does not identify who signed it (claim 360). The NVIDIA 5GW 'strategic partnership' is a framework, not a revenue contract (claims 11077, 9289). NVIDIA's $2.1 billion investment vests only as its GPU infrastructure is deployed (claim 9179). That makes NVIDIA customer, supplier-aligned partner and prospective shareholder at once, which readers should weigh when assessing arm's-length terms.

Concentration moved from diffuse, small, unnamed white-label revenue to heavy dependence on two counterparties. The arithmetic broadly reconciles. $9.7 billion over five years implies about $1.94 billion a year, matching the $1.9 billion expected average (claims 20245, 8996). NVIDIA's $3.4 billion implies about $0.68 billion. Together with the $0.4 billion Prince George contracts, that is roughly $3.0 billion against the $3.1 billion ARR under contract (claim 9183). That leaves only about $0.1-0.2 billion for all other named customers. On that basis Microsoft and NVIDIA represent over 80% of contracted ARR. That ARR figure is forward-looking: NVIDIA capacity targets a ramp only from early 2027 (claim 9287). The July $2.8 billion postdates the $3.1 billion figure and should be treated as additive only once its counterparties and ARR contribution are disclosed.

IREN's record divides cleanly into two eras. Through mid-2025, almost nothing on the AI side was a signed, named customer agreement. Poolside AI (claim 6285) and WEKA (claim 12005) were identified, but WEKA is a software partner rather than a buyer, and the $28 million run-rate (claims 29824, 28325) came from white-labelled compute sold to unnamed 'leading US AI cloud providers' (claim 29810). Horizon 1 and Sweetwater were described entirely in pipeline language: due diligence, pricing negotiations, letters of intent, and 'multiple potential customers' (claims 31515, 31448, 29802, 31516). The end of the Morgan Stanley sale mandate (claim 31471) shows Sweetwater's monetisation path was still undecided. Bitmain, Dell, BC Hydro and Kraken are supplier or service counterparties, not revenue sources (claims 14392, 6152, 12198, 12183). The Microsoft contract of November 2025 is the first unambiguous signed, named, sized customer agreement, filed with a counterparty, date and site (claims 20307, 4157, 1266). It was followed by named but unsized contracts with Together AI, Fluidstack and Fireworks AI (claim 20247), and by the $3.4 billion NVIDIA contract (claims 11103, 9180). Investors should keep that NVIDIA contract separate from the 5GW 'strategic partnership' (claims 11077, 9289), which is a framework. The same goes for the $2.1 billion NVIDIA investment, which vests only as deployments occur (claim 9179), and the 'Preferred Partner' label (claim 20300). The further '>$0.5bn ARR' in British Columbia was still in negotiation (claim 16666), and the customer list in claim 8101 trails off into an unidentified new name. Concentration is high. Microsoft's roughly $1.9 billion average annual contribution (claims 20245, 8996) is about 61% of the $3.1 billion ARR under contract (claim 9183). Adding NVIDIA's implied ~$0.68 billion a year brings two counterparties to roughly 83%. The July 2026 $2.8 billion of contracts (claim 360) at a ~4-year weighted term (claim 8108) would add perhaps $0.7 billion of ARR and pull Microsoft's share toward half, but no per-customer values are disclosed. The arithmetic broadly reconciles. Microsoft (~$1.94 billion), NVIDIA (~$0.68 billion) and Prince George (~$0.4 billion, claim 16665) sum to about $3.0 billion, leaving roughly $0.1 billion for the smaller named contracts. Supplier commitments track customer value consistently: $5.8 billion of Dell GPUs against $9.7 billion from Microsoft (claim 4161) and $1.6 billion against $3.4 billion from NVIDIA (claim 1297). Both are parent-guaranteed exposures (claim 4164).

IREN's record splits cleanly into signed agreements and softer framing. The firm, named commitments are the Microsoft GPU services contract, $9.7bn over five years at Childress with a 20% prepayment (claims 20307, 4157, 1266, 20245), and the five-year $3.4bn NVIDIA AI Cloud contract (claims 11076, 9180), which later filings recharacterize as a managed-services arrangement (claims 1299, 8995). Earlier signed relationships are supplier or utility contracts, not revenue customers: Bitmain hardware purchases (claims 14392, 6374, 6152), BC Hydro power (claim 12198), Kraken liquidation (claim 12183), and Dell GPU purchases of $5.8bn and $1.6bn (claims 4161, 1297). The WEKA 'partnering' (claim 12005), the 5GW NVIDIA 'strategic partnership' (claims 11077, 9289) and the NVIDIA Preferred Partner label (claim 20300) are frameworks, not revenue contracts. Mirantis (claim 8768) is a technology partner, not a customer. Before November 2025, every AI customer reference was anonymous or pre-contractual. The $28m run-rate came from unnamed 'leading US AI cloud providers' (claims 29824, 28325), and Horizon 1 sat in 'due diligence and contractual negotiations' (claims 29802, 31448). Counterparties named since include Poolside (claim 6285), Together AI, Fluidstack and Fireworks AI (claim 20247), and Perplexity, Figure AI, Fal AI and Hume AI (claim 8101). The $0.4bn Prince George ARR has no named counterparty (claim 16665), and the further >$0.5bn was still in negotiation (claim 16666). The 'new leading AI developer' (claim 8101) and the frontier lab IREN 'couldn't name' (claim 29143) also remain unidentified. Concentration moved from a handful of anonymous resellers to a single hyperscaler. Microsoft's $1.9bn expected annual revenue (claim 8996) was about 61% of the $3.1bn ARR under contract (claim 9183), and Microsoft plus NVIDIA together are roughly 83%. NVIDIA is now simultaneously customer, GPU ecosystem supplier and an investor whose $2.1bn rights vest only as its infrastructure deploys (claim 9179), which is a circularity worth watching. The arithmetic broadly reconciles. $9.7bn over five years implies about $1.94bn, consistent with claim 20245. NVIDIA's $3.4bn implies about $0.68bn a year, and adding Prince George's $0.4bn brings the total to roughly $2.98bn, leaving only about $0.1bn for the November mid-size contracts. The July $2.8bn tranche (claims 360, 8117) carries a weighted term of about four years (claim 8108), which would add roughly $0.7bn, but no refreshed ARR figure is on record to confirm it.

IREN's customer record falls into three phases. In the bitcoin-mining era, the named counterparties were suppliers and service providers rather than revenue customers: Bitmain hardware contracts (14392, 6374, 6152), BC Hydro power (12182, 12198) and Kraken for liquidation (12183). The 2023-24 pivot added a WEKA software partnership (12005), which is vendor positioning, and one named GPU customer, Poolside AI (6285). By May 2025 the AI cloud business still rested on unnamed counterparties. Revenue of a $28 million run-rate came from white-labelled compute for 'leading US AI cloud providers' (29810, 29824, 28325), and Horizon 1 was only 'several customers undergoing due diligence and contractual negotiations' (29802, 29819), which is pipeline language.

The November 2025 Microsoft agreement is the clearest signed, named contract. It is a $9.7 billion, five-year deal with a 20% prepayment (4157, 20307, 20245). Microsoft's accepted delivery of Horizon 1 (35, 8673) moves it from paper to performance. The arithmetic reconciles: $9.7 billion over five years is about $1.94 billion a year, consistent with the $1.9 billion ARR guidance (20245, 8996). The $5.8 billion Dell purchase (4161, 20308), guaranteed by the parent (4164), is a capital commitment, not revenue.

NVIDIA requires more care. The $3.4 billion five-year contract is signed (11076, 9180) and implies about $0.68 billion a year. The '5GW strategic partnership' (11077, 9289), however, is framework language with no stated economics. NVIDIA is also a $2.1 billion investor whose rights vest only as NVIDIA infrastructure is deployed (9179). That makes NVIDIA customer, supplier ecosystem and capital provider at once, a circularity worth noting. The Microsoft and NVIDIA streams, about $2.62 billion combined, plus roughly $0.4 billion contracted at Prince George (16665), approximate the stated $3.1 billion ARR under contract (9183). The additional '>$0.5bn' in British Columbia was still under negotiation (16666) and should not be counted.

The July 2026 $2.8 billion tranche (360) has weaker attribution. IREN's own list names Perplexity, Figure AI, Fal AI and Hume AI alongside earlier customers Together AI, Fluidstack and Fireworks AI (20247), but it also includes 'a new leading AI developer' (8101). The August frontier-lab contract is likewise unnamed (29143). The deal-to-name mapping in 8117 comes from secondary reporting. Concentration is easing but remains heavy. Of roughly $15.9 billion in disclosed contract value, Microsoft accounts for about 61% and NVIDIA about 21%. The record supplied is truncated at claim 8541, so later disclosures are not reflected here.

IREN's record separates cleanly into a small set of signed, named customer contracts and a larger layer of supplier agreements, framework language and unnamed counterparties. The anchor is the five-year, $9.7bn Microsoft GPU services contract dated November 2, 2025 (claims 20307, 4157, 1266), which is substantiated by an 8-K-style description and by Microsoft's acceptance of Horizon 1 (claims 35, 8673). The second is the five-year, $3.4bn NVIDIA AI Cloud contract (claims 9180, 9287, 11076). The NVIDIA contract should be kept apart from the '5GW strategic partnership' (claims 11077, 9289), which is collaboration language without stated economics, and from NVIDIA's $2.1bn investment right, which vests only as deployment milestones are met (claim 9179). Dell (claims 4161, 1297), Bitmain (claims 14392, 6152), WEKA (claim 12005), BC Hydro (claim 12198) and Kraken (claim 12183) are suppliers or service venues, not revenue counterparties. Mirantis's 1,500 enterprise customers (claim 8768) belong to Mirantis, not IREN.

Identification has been uneven. The 2025 white-labelled revenue came from 'leading US AI cloud providers' who were never named (claims 29810, 28325). Horizon 1 was pitched as 'several customers undergoing due diligence' (claim 29802) before Microsoft surfaced. Together AI, Fluidstack and Fireworks AI are named as multi-year contracts (claim 20247), but the July $2.8bn tranche is attributed only to 'leading AI developers' (claim 360), and the named list comes from a secondary 'companies like' characterization (claim 8117). The customer roster also includes 'a new leading AI developer' (claim 8101) and an unnamed frontier lab (claim 29143). The British Columbia figures mix ~$0.4bn ARR 'under contract' with >$0.5bn still in 'negotiations' (claims 16665, 16666).

Concentration has moved from a single $28m white-label run-rate (claim 29824) to heavy Microsoft dependence. The $1.9bn expected ARR (claims 20245, 8996) is roughly 61% of the $3.1bn ARR under contract reported in May 2026 (claim 9183). The NVIDIA and July contracts dilute this only partially.

The arithmetic broadly reconciles. $9.7bn over five years is about $1.94bn a year, and $3.4bn over five years is about $0.68bn. Adding ~$0.4bn for Prince George gives about $3.0bn, leaving a modest residual for smaller contracts against the $3.1bn total. The July $2.8bn cannot be converted to ARR without a disclosed term; the ~4-year weighted average (claim 8108) implies roughly $0.7bn. Some amounts in the record are inconsistently normalized (claims 8112, 9292), and the record is truncated at claim 361.

IREN's record divides into a small number of signed, dollar-denominated customer contracts and a larger body of framework, negotiation and unnamed-counterparty language. The signed core is the Microsoft GPU services contract of November 2, 2025, a five-year, $9.7bn deal with a 20% prepayment and a guided $1.9bn in average annual revenue (claims 20307, 4157, 20245, 8996). Microsoft has since accepted delivery of Horizon 1 (claims 35, 8673), which moves that contract from signature to performance. The second signed contract is NVIDIA's five-year, $3.4bn AI Cloud agreement (claims 11076, 9287, 9180). Its companion 5GW strategic partnership is collaboration language with no committed capacity or price (claims 11077, 9289). NVIDIA's $2.1bn investment rights vest only as NVIDIA infrastructure is deployed (claim 9179), so NVIDIA is customer, investor and supplier ecosystem at once, and that circularity deserves scrutiny. The Dell agreements of $5.8bn and $1.6bn (claims 4161, 1297) are supply-side obligations, guaranteed by the parent (claim 4164). They are not revenue.

Attribution has been uneven over time. The May 2025 disclosures cite unnamed 'leading US AI cloud providers' generating $28m of run-rate revenue (claims 29824, 29810, 28325). The same disclosures describe Horizon 1 customers only as in 'due diligence and contractual negotiations' (claims 29802, 29819). Together AI, Fluidstack and Fireworks AI are named but carry no contract values (claim 20247). The Prince George figure of roughly $0.4bn ARR is described as under contract without counterparties, and the further >$0.5bn is explicitly still in negotiation (claims 16665, 16666). The July 2026 $2.8bn tranche is attributed to a list that includes 'a new leading AI developer' (claims 360, 8101, 8117). An August contract with a frontier lab is disclosed but unnamed (claim 29143). Mirantis's 1,500 enterprise customers belong to Mirantis, not IREN (claim 8768).

On concentration, the business shifted from merchant Bitcoin sold through Kraken (claim 12183) to two counterparties. Microsoft's $1.9bn makes up roughly 61% of the $3.1bn ARR under contract stated in May 2026 (claim 9183). NVIDIA's $3.4bn over five years adds about $0.68bn, so the two together account for about 83%.

The arithmetic broadly reconciles. Microsoft's $1.9bn, NVIDIA's roughly $0.68bn and Prince George's roughly $0.4bn sum to about $2.98bn, which leaves around $0.1bn for the named neocloud customers. That depends on Prince George not overlapping with those customers, and the disclosures do not settle the point. The July $2.8bn tranche implies roughly $0.7bn of additional ARR at the stated four-year weighted term (claim 8108). No updated ARR total appears on record to confirm it.

IREN's record divides cleanly into a small set of signed, named commercial agreements and a larger layer of partnership and pipeline language. The signed, identified contracts are the five-year, $9.7bn Microsoft GPU services agreement (claims 20307, 4157, 1266), which reached a verifiable milestone when Horizon 1 was delivered to and accepted by Microsoft (claims 35, 8673). The others are the five-year, $3.4bn NVIDIA AI Cloud contract (claims 11103, 9180), and the supplier-side Dell purchase agreements of roughly $5.8bn and $1.6bn (claims 4161, 1297), the first backed by a parent guarantee (claim 4164). The earlier mining-era Bitmain contracts are described as binding (claims 14392, 6152), and the BC Hydro supply contracts carry defined terms (claim 12198).

Softer language sits around these. The WEKA 'partnering' (claim 12005), 'NVIDIA Preferred Partner' status (claim 20300) and the 5GW NVIDIA 'strategic partnership' (claims 11077, 9289) are frameworks, not revenue contracts. NVIDIA's $2.1bn investment vests only as deployments occur (claim 9179). Poolside was named, but only the 'commencement' of service was disclosed, with no terms (claim 6285). Unidentified counterparties recur across the record. In May 2025, white-labelled compute went to unnamed 'leading US AI cloud providers' (claims 29810, 29824, 28325), and 'several customers' were described as in due diligence (claims 29802, 29819). The July 2026 list includes 'a new leading AI developer' (claim 8101), and in August IREN cited an unnamable 'frontier AI lab' (claim 29143). Together AI, Fluidstack and Fireworks AI are named (claim 20247), but contract values are not broken out for them individually. The Mirantis items (claims 8768, 8769) describe Mirantis's own credentials rather than an IREN customer relationship. Claim 8541 is truncated and cannot be assessed.

Concentration has moved from a bitcoin model with a single liquidation venue, Kraken (claim 12183), to heavy hyperscaler dependence. Microsoft's expected $1.9bn of annual revenue (claims 20245, 8996) is about 61% of the $3.1bn ARR under contract stated in May 2026 (claim 9183). NVIDIA's $3.4bn over five years adds roughly $0.68bn, so the two together account for about 83%.

The arithmetic broadly reconciles. $9.7bn over five years is about $1.94bn a year, consistent with the $1.9bn figure. The residual of roughly $0.5bn fits the ~$0.4bn Prince George ARR under contract (claim 16665) plus smaller customers. The >$0.5bn of British Columbia negotiations (claim 16666) has no confirmed conversion on record. The July $2.8bn tranche (claims 360, 8117) implies about $0.7bn of additional ARR at the stated ~4-year average term (claim 8108). However, no updated ARR total has been disclosed against which to check that figure.

IREN's disclosure record falls into three tiers. The first is signed, named contracts. The second is named relationships described in partnership or framework language. The third is unnamed counterparties. Before 2025 the named links were supply-side rather than revenue-side: binding hardware contracts with Bitmain (claims 14392, 6374, 6152), BC Hydro power contracts (12198), and Kraken as a liquidation venue (12183). The WEKA tie-up was described only as 'partnering' (12005). Poolside AI was the first named compute customer, though it was disclosed as a service commencement with no contract value (6285). By May 2025 the AI cloud business still rested on unnamed 'leading US AI cloud providers' generating $28m run-rate (29824, 28325), with 'several customers undergoing due diligence and contractual negotiations' (29802, 29819). That is pipeline language, not signed business.

The Microsoft agreement changed the picture. It is a signed, named, $9.7bn five-year contract with a 20% prepayment and $1.9bn expected ARR (20307, 4157, 20245, 8996). It is backed by a guaranteed $5.8bn Dell purchase obligation (4161, 4164), and Horizon 1 has since been delivered and accepted (35, 8673). Together AI, Fluidstack and Fireworks AI followed as named multi-year contracts, but without values (20247). The NVIDIA relationship needs to be split in two. The $3.4bn five-year contract is signed (9180, 9287). The '5GW strategic partnership' (11077, 9289) and 'Preferred Partner' status (20300) are framework language. NVIDIA's $2.1bn investment vests only as NVIDIA infrastructure is deployed (9179), which makes NVIDIA a customer, supplier and prospective shareholder at once. Unnamed exposure remains: 'a new leading AI developer' (8101) and an unnamed frontier lab (29143). In British Columbia, about $0.4bn of ARR is contracted (16665) against more than $0.5bn still in negotiation (16666).

Concentration was extreme after November 2025. Microsoft's $1.9bn was about 61% of the $3.1bn ARR under contract reported in May 2026 (9183). The arithmetic broadly reconciles. Microsoft at $1.9bn, NVIDIA at roughly $0.68bn ($3.4bn over five years) and Prince George at about $0.4bn sum to about $2.98bn, leaving a modest residual for the smaller named contracts. The July 2026 $2.8bn tranche (360, 8117) has no disclosed updated ARR total in this record. At the stated roughly four-year weighted term (8108), it would imply about $0.7bn of additional ARR, which would lower Microsoft's share toward the high-40s percent. That figure should be treated as an inference, not a company disclosure. The supplied record cuts off at claim 8553, so any later reconciliation could not be reviewed.

IREN's record divides cleanly between a small number of executed, named customer contracts and a wider layer of framework, supplier and unattributed language. The executed core is Microsoft, a five-year, $9.7 billion GPU services agreement signed November 2, 2025 through IE US Hardware 3 (claims 4157, 20307, 1266). It has passed a hard milestone: Horizon 1 has been delivered to and accepted by Microsoft (claims 35, 8673). The second is NVIDIA's five-year, $3.4 billion AI cloud contract (claims 11076, 9180, 1299). Everything else in that announcement deserves different weight. The '5GW strategic partnership' is collaboration language with no stated consideration (claims 11077, 9289). The $2.1 billion NVIDIA investment vests only as GPU infrastructure is deployed (claim 9179). 'NVIDIA Preferred Partner' is a program designation, not a customer relationship (claim 20300). Dell, Bitmain, WEKA, Mirantis and Kraken are suppliers or vendors, and the Dell purchase commitments of $5.8 billion and $1.6 billion are cost obligations backing the two revenue contracts (claims 4161, 1297, 4164).

Attribution has improved but remains incomplete. The 2025 white-labelled revenue came from unnamed 'leading US AI cloud providers' (claims 29810, 29824, 28325). Horizon 1 was described only as 'several customers' in diligence (claims 29802, 29819). The July 2026 roster names Perplexity, Figure AI, Together AI, Fluidstack, Fireworks AI, Fal AI and Hume AI, yet it also includes an unidentified 'new leading AI developer' (claim 8101), and an August contract is with a frontier lab IREN 'couldn't name' (claim 29143). Poolside AI, an early named customer (claim 6285), is absent from that list.

Concentration has moved from a single offtaker, Bitcoin liquidated via Kraken (claim 12183), to Microsoft dominance. Of roughly $15.9 billion in disclosed contract value across Microsoft, NVIDIA and the July tranche, Microsoft is about 61 percent, and Microsoft and NVIDIA together exceed 80 percent.

The arithmetic largely reconciles. $9.7 billion over five years is about $1.94 billion a year, consistent with the stated $1.9 billion ARR (claims 20245, 8996). NVIDIA adds roughly $0.68 billion, and Prince George adds about $0.4 billion (claim 16665). Together these approach the $3.1 billion ARR under contract (claim 9183), with the smaller named contracts (claim 20247) plausibly covering the remainder. Two items need caution. The British Columbia '>$0.5bn ARR' was still under negotiation (claim 16666), so it should not be treated as contracted. The July $2.8 billion carries no stated ARR, and the press characterization of it as a single 'deal' (claim 8117) is looser than the company's own wording of multiple contracts (claim 360).

IREN's contract record splits cleanly into a small set of signed, named customer agreements and a wider surround of supplier, framework and pre-signing language. The anchor is the Microsoft GPU services contract signed November 2, 2025: five years, $9.7bn, at Childress, with a 20% prepayment and $1.9bn of expected annual revenue (claims 20307, 4157, 20245, 8996). Microsoft's acceptance of Horizon 1 in August 2026 shows that contract is now delivering, not just booked (claims 35, 8673). The second signed anchor is the five-year, $3.4bn NVIDIA AI Cloud contract of May 2026 (claims 11076, 9180, 9287). The broader 5GW NVIDIA "strategic partnership" is a separate thing. It is framework language with no stated revenue (claims 11077, 9289), and NVIDIA's $2.1bn investment rights vest only as its infrastructure is deployed (claim 9179). That makes NVIDIA customer, investor and supplier-ecosystem partner at once, a circularity worth watching. Bitmain, Dell, BC Hydro, Kraken, WEKA and Mirantis are suppliers or vendors, not revenue counterparties (claims 14392, 4161, 1297, 12198, 12183, 12005, 8768).

The unnamed layer is persistent. The 2025 "white-labelled" capacity went to "leading US AI cloud providers" who were never identified (claims 29810, 29824, 28325). Horizon 1 was described as having "several customers" in diligence before Microsoft was named (claims 29802, 29819). The "broad range of customers" line is purely descriptive (claim 20306). The customer roster of July 2026 includes "a new leading AI developer" (claim 8101), and August brought a contract with a frontier lab IREN "couldn't name" and gave no value for (claim 29143). The British Columbia figure of >$0.5bn ARR is explicitly "remaining contract negotiations," not signed revenue (claim 16666). Smaller named counterparties are real but thinly quantified: Poolside, Together AI, Fluidstack, Fireworks AI, Figure AI and Hume AI (claims 6285, 20247, 8117).

Concentration moved from Bitcoin self-mining to a single hyperscaler. Against the $3.1bn ARR under contract (claim 9183), Microsoft's $1.94bn ($9.7bn over five years) is about 62%. Adding NVIDIA's roughly $0.68bn takes the two to about 84%.

The arithmetic broadly reconciles. Microsoft ($1.94bn), NVIDIA ($0.68bn) and Prince George (~$0.4bn, claim 16665) sum to about $3.0bn, which leaves roughly $0.1bn for the smaller contracts and legacy white-label revenue. The July contracts add $2.8bn of total value (claim 360). At the disclosed ~4-year weighted term (claim 8108), that implies roughly $0.7bn of incremental ARR. IREN has not stated that figure itself, and the unnamed frontier-lab deal cannot be sized at all.

IREN's signed-customer record is narrower than its partnership vocabulary suggests. The anchor is the five-year Microsoft GPU services contract dated November 2, 2025 (claims 20307, 4157, 1266). It is valued at $9.7bn with a 20% prepayment and $1.9bn expected ARR (20245, 8996). It is the only contract with a disclosed operational milestone: Horizon 1 delivered and accepted (35, 8673). The second is a five-year, $3.4bn NVIDIA contract (11076, 9180), later described as 'managed services' (8995, 1299). NVIDIA is also a $2.1bn investor whose rights vest only as NVIDIA infrastructure deploys (9179), so that customer relationship carries circularity. The surrounding '5GW strategic partnership' (11077, 9289) is framework language with no committed dollars. Dell (4161, 1297), Bitmain (14392, 6374, 6152), BC Hydro (12182), Kraken (12183), WEKA (12005) and Mirantis (8768) are suppliers or vendors, not revenue counterparties. Several revenue claims lack named counterparties. The $28m white-labelled run-rate came from unnamed 'leading US AI cloud providers' (29824, 28325). Horizon 1 'customers undergoing due diligence' were never itemised (29802, 29819). The ~$0.4bn Prince George ARR is 'under contract' without names, while a further >$0.5bn remains in negotiation (16665, 16666). There is also a 'new leading AI developer' (8101) and a 'frontier AI lab it couldn't name' (29143). Together AI, Fluidstack and Fireworks AI are named (20247), as are the July cohort (8101), but values appear only as a $2.8bn aggregate (360); claim 8117 is a secondary paraphrase. Concentration has moved from Bitcoin sales through a single exchange, to unnamed white-label resale, to a Microsoft-dominated book. On a straight-line basis Microsoft ($1.94bn, consistent with $9.7bn over five years) plus NVIDIA (~$0.68bn) plus Prince George (~$0.4bn) gives roughly $3.0bn. That reconciles reasonably with the $3.1bn ARR under contract reported in May 2026 (9183), and implies Microsoft was about 60% of contracted ARR. The July $2.8bn, at the ~4-year weighted term (8108), would add roughly $0.7bn ARR and dilute Microsoft to about half. That figure is derived, not disclosed, and the NVIDIA ARR depends on a ramp targeted only from early 2027 (9287). Claim 8541 is truncated in the record and was not assessed.

IREN's record splits cleanly into a small number of signed, named, valued contracts and a wider set of unnamed or framework-level counterparty language. The anchor is the Microsoft GPU services agreement dated November 2, 2025: $9.7bn over five years, with a 20% prepayment and roughly $1.9bn of expected annual revenue (claims 4157, 20307, 20245, 8996, 1266). It is the only relationship on record with an operational milestone, the delivery and acceptance of Horizon 1 (claims 35, 8673). The NVIDIA $3.4bn five-year AI cloud contract is also signed and specified, with air-cooled Blackwell systems at Childress ramping from early 2027 (claims 11076, 9180, 9287). The adjacent 5GW 'strategic partnership' is framework language, not a revenue commitment (claims 11077, 9289). NVIDIA's $2.1bn investment, whose rights vest only as NVIDIA infrastructure is deployed, makes NVIDIA customer, supplier ecosystem and prospective shareholder at once, and that interlock deserves scrutiny (claim 9179). Dell, Bitmain, BC Hydro, Kraken and WEKA are suppliers or vendors, not customers (claims 4161, 1297, 14392, 6152, 12198, 12183, 12005), and the Mirantis figure of 1,500 customers describes Mirantis's base, not IREN's (claim 8768).

On identification, the pre-Microsoft AI revenue rested on 'leading US AI cloud providers' who were never named (claims 29810, 29824, 28325), and Horizon 1 interest was described only as 'several customers' in diligence (claims 29802, 29819). Together AI, Fluidstack and Fireworks AI are named, but no individual contract values are disclosed (claim 20247). The July $2.8bn tranche is reported as a single aggregate across named developers (claims 360, 8117). The roster also includes 'a new leading AI developer' and a 'frontier AI lab it couldn't name' (claims 8101, 29143).

Concentration moved from a $28m run-rate spread across anonymous resellers to a book dominated by Microsoft, with NVIDIA a distant second. The arithmetic broadly reconciles. Microsoft's $9.7bn over five years implies about $1.94bn a year, matching the $1.9bn guidance. NVIDIA's $3.4bn over five years implies about $0.68bn, and Prince George adds about $0.4bn (claim 16665). Together these come to roughly $3.0bn against the stated $3.1bn of ARR under contract (claim 9183), leaving only about $0.1bn for the smaller named customers. The further >$0.5bn of British Columbia ARR is explicitly still in negotiation (claim 16666). The July $2.8bn post-dates the $3.1bn figure, and no reconciled ARR update appears in this record, whose final entry (claim 361) is truncated.

IREN's signed-customer record is far shorter than its partnership record, and most of it dates from November 2025. Before then, the named counterparties were suppliers and service providers, not revenue customers. These include Bitmain hardware contracts (14392, 6374, 6152), Kraken for Bitcoin liquidation (12183) and BC Hydro power supply (12182, 12198). The WEKA 'partnering' (12005) is vendor language with no economics attached. Poolside AI (6285) was the first named GPU cloud customer, though it was announced only as a 'commencement' with no contract value. The pivotal signed agreement is the Microsoft contract. It is identified by counterparty, entity (IE US Hardware 3), site, term and value at $9.7bn (20307, 4157, 1266). It is supported by a $5.8bn Dell purchase agreement guaranteed by the parent (4161, 4164), and its execution is evidenced by Microsoft's acceptance of Horizon 1 (35, 8673). The NVIDIA $3.4bn five-year contract is also described as signed (9180, 9287) and is backed by a $1.6bn Dell order (1297, 8992). However, it sits beside a '5GW strategic partnership' (11077, 9289) that is framework language without committed volume. It also sits beside a $2.1bn NVIDIA investment whose rights vest only as NVIDIA infrastructure deploys (9179), which creates customer-investor circularity that readers should weigh. Unidentified counterparties recur throughout. They include 'leading US AI cloud providers' behind the $28m white-label run-rate (29810, 29824, 28325), 'several customers' in Horizon 1 due diligence (29802), a 'new leading AI developer' (8101), an unnamed frontier lab (29143), and British Columbia negotiations said to support more than $0.5bn of ARR that remain negotiations, not contracts (16666). Concentration has moved from effectively zero external customers to heavy Microsoft dependence. At $1.9bn of expected ARR (20245, 8996) against $3.1bn under contract (9183), Microsoft represented roughly 60% of contracted revenue in May 2026. Partial dilution followed through NVIDIA and the July $2.8bn tranche spread across Together AI, Fluidstack, Fireworks, Figure, Hume and others (20247, 360, 8101, 8117). The arithmetic broadly reconciles. The Microsoft contract at $9.7bn over five years implies about $1.94bn a year. The NVIDIA contract at $3.4bn over five years implies about $0.68bn. Prince George adds about $0.4bn (16665). Together these total roughly $3.0bn, which is consistent with $3.1bn once the smaller late-2025 contracts are included. The July $2.8bn, at the roughly four-year weighted term (8108), implies about $0.7bn of incremental ARR, but no updated consolidated ARR figure appears on record to confirm it. Claim 362 is truncated in the source and was not assessed.

IREN's record divides cleanly into signed agreements and softer framing, though several signed items are supply contracts rather than customer revenue. The Bitmain contracts (14392, 6374, 6152), the BC Hydro power contracts (12198, 12182) and the two Dell GPU purchases of $5.8bn (4161, 20308) and $1.6bn (1297, 8992) are binding, but they are cost-side commitments. Kraken (12183) and WEKA (12005) are service and vendor relationships, not buyers. The first named compute customer is Poolside (6285). The anchor signed contract is Microsoft's $9.7bn, five-year GPU services agreement (20307, 4157, 1266), and it has moved past paper: Horizon 1 has been delivered and accepted (35, 8673). NVIDIA's $3.4bn five-year contract is reported as signed (9180, 11076). The surrounding '5GW strategic partnership' (11077, 9289), however, is framework language, and the $2.1bn NVIDIA investment vests only as NVIDIA infrastructure is deployed (9179). That makes NVIDIA simultaneously customer, supplier ecosystem and investor, a circularity readers should weigh. Much of the earlier and later language is unidentified. The 'leading US AI cloud providers' buying white-labelled compute (29810, 29824, 28325) are never named. Horizon 1 'customers undergoing due diligence and contractual negotiations' (29802, 29819) are not contracts. British Columbia carries more than $0.5bn of ARR still in negotiation (16666), against $0.4bn signed (16665). A 'new leading AI developer' (8101) and an unnamed frontier AI lab (29143) are also undisclosed. Together AI, Fluidstack and Fireworks AI are named (20247) but have no disclosed values. The $2.8bn July tranche (360) is attributed to specific names only by secondary press using 'companies like' phrasing (8117). Mirantis's 1,500 enterprise customers (8768) are Mirantis's, not IREN's. Concentration moved from no true customers in the mining era, to a single named GPU client, to heavy Microsoft dependence. The arithmetic broadly reconciles. Microsoft's $9.7bn over five years matches the stated $1.9bn ARR (20245, 8996). NVIDIA's $3.4bn over five years adds about $0.68bn, and Prince George adds $0.4bn, for roughly $2.98bn against the $3.1bn ARR under contract (9183). That leaves about $0.1bn for the remaining smaller customers. Microsoft is therefore around 61% of contracted ARR at that date. The July $2.8bn, at a roughly four-year weighted term (8108), would add about $0.7bn and dilute Microsoft toward half, but no updated ARR total appears on record. The claim list also ends truncated at claim 361, so later disclosures could not be checked.

IREN's record splits into a small number of signed, named customer contracts and a larger body of supplier, framework, and unnamed-counterparty language. The anchor is the Microsoft GPU services agreement dated November 2, 2025, announced as signed (20307, 4157, 1266). Its $9.7bn, five-year value implies roughly $1.9bn a year, matching management's stated contribution (20245, 8996). It moved beyond paper when Microsoft accepted delivery of Horizon 1 in August 2026 (35, 8673). The $3.4bn NVIDIA contract is also described as signed (9180, 9287). The surrounding '5GW strategic partnership' (11077, 9289), however, is a collaboration framework with no stated revenue, and NVIDIA's $2.1bn investment rights vest only as its GPUs are deployed (9179). That makes NVIDIA customer, supplier, and prospective shareholder at once, and readers should weigh that circularity. Several named relationships are not customers at all. Bitmain (14392, 6374, 6152) and Dell (4161, 20308, 1297, 8992) are equipment vendors, and IREN guarantees the Dell obligations (4164). BC Hydro supplies power (12182, 12198), Kraken is a liquidation venue (12183), and WEKA is a software partnership (12005). Mirantis's 1,500 customers belong to Mirantis, not IREN (8768). On identification, Poolside AI (6285), Together AI, Fluidstack and Fireworks AI (20247) are named. The 2025 white-label buyers were only 'leading US AI cloud providers' (29810, 28325), and the 'several customers' in Horizon 1 diligence were never named before Microsoft emerged (29802, 29819). The July 2026 $2.8bn tranche went to 'leading AI developers' (360). Press reports name some of them (8117), but the company's own list still includes 'a new leading AI developer' (8101), and an August contract is with a frontier lab IREN could not name (29143). Concentration moved from a $28m diffuse white-label base to extreme reliance on one customer. In May 2026, Microsoft's $1.9bn was about 61% of the $3.1bn ARR under contract (9183), and Microsoft plus NVIDIA was roughly 83%. That arithmetic broadly reconciles. Microsoft ($1.9bn), NVIDIA ($3.4bn over five years, about $0.68bn a year) and Prince George (about $0.4bn, 16665) sum to about $3.0bn, leaving little for the earlier smaller contracts. The further >$0.5bn in British Columbia was still under negotiation (16666) and should be excluded. If the July $2.8bn runs near the 4-year weighted term (8108), it adds about $0.7bn a year and dilutes Microsoft toward half. Contract values are carried in inconsistent units across claims, and the record is truncated at claim 361, so later announcements are not assessed here.

IREN's signed, named customer agreements are few, and its broader counterparty narrative leans heavily on suppliers, strategic framing and unidentified buyers. Two customer contracts are clearly executed and identified. The first is the five-year, $9.7bn Microsoft GPU services agreement (claims 20307, 4157, 1266), whose Horizon 1 tranche has been delivered and accepted (claims 35, 8673). The second is the five-year, $3.4bn NVIDIA AI Cloud contract (claims 11076, 9180), which only targets ramp from early 2027 (claim 9287). Many other named counterparties are suppliers or service providers rather than customers: Bitmain (claims 14392, 6374, 6152), BC Hydro (claim 12198), Kraken (claim 12183), WEKA (claim 12005) and Dell, whose $5.8bn and $1.6bn purchase agreements are capex obligations guaranteed by the parent (claims 4161, 4164, 1297). The NVIDIA '5GW strategic partnership' (claims 11077, 9289), the 'Preferred Partner' label (claim 20300) and the vesting-contingent $2.1bn NVIDIA investment (claim 9179) are framework language, not revenue commitments. Mirantis's 1,500 customers (claim 8768) belong to Mirantis, not IREN. Unidentified counterparties recur throughout: 'leading US AI cloud providers' buying white-labelled compute (claims 29810, 28325), 'several customers' in due diligence (claims 29802, 29819), 'a new leading AI developer' (claim 8101) and an unnamed frontier lab (claim 29143). Smaller AI names, including Poolside, Together AI, Fluidstack and Fireworks AI, appear as contract wins (claims 6285, 20247), but no values are attached to them individually. The July $2.8bn tranche is attributed to named firms only via secondary reporting ('companies like', claim 8117), while the company's own wording is 'leading AI developers' (claim 360). Concentration has moved from a commodity Bitcoin market to a single hyperscaler and then to a hyperscaler-plus-supplier structure. Microsoft's $1.9bn expected ARR (claims 20245, 8996) is about 61% of the $3.1bn ARR under contract (claim 9183). NVIDIA's $3.4bn over five years implies roughly $0.68bn a year, so the two together account for about 83% of contracted ARR. The arithmetic broadly reconciles: $1.9bn plus $0.68bn plus Prince George's $0.4bn (claim 16665) equals about $2.98bn, leaving around $0.1bn for smaller AI Cloud contracts. However, 'ARR under contract' includes capacity not yet deployed. The July $2.8bn at a roughly four-year weighted term (claim 8108) would add about $0.7bn, but the company has not disclosed how it overlaps with the British Columbia negotiations (claim 16666). The claim record supplied was truncated at claim 361, so later disclosures are not assessed here.

IREN's signed-customer record is narrow and recent. Before November 2025, most named counterparties were suppliers or service vendors rather than customers: Bitmain for miners (14392, 6374, 6152), BC Hydro for power (12182, 12198), Kraken for Bitcoin liquidation (12183), and WEKA for software (12005). The only named AI customer in that period was Poolside AI, described as a service commencement with no disclosed value (6285). The May 2025 AI cloud revenue, a $28 million run-rate, came from unnamed "leading US AI cloud providers" through white-labelled compute (29810, 29824, 28325). At the same time, Horizon 1 customers were described only as being in "due diligence and contractual negotiations" (29802, 29819).

The first named, signed, and valued contract is Microsoft's: $9.7 billion over five years, carried by a subsidiary with a 20% prepayment (20307, 4157, 20245, 1266). Delivery and acceptance of Horizon 1 in August 2026 is the strongest performance evidence in the file (35, 8673). The NVIDIA $3.4 billion five-year contract is also described as signed (9180, 11076). The surrounding "5GW strategic partnership" is collaboration language with no stated economics, and the $2.1 billion NVIDIA investment vests only as deployments occur (11077, 9289, 9179). "NVIDIA Preferred Partner" is a program designation, not a customer contract (20300). Mirantis's 1,500 customers are Mirantis's own, not IREN's (8768). The Dell agreements of $5.8 billion and $1.6 billion are capital commitments that IREN guarantees, not revenue (4161, 4164, 1297). Several counterparties remain unidentified: "a leading frontier AI lab it couldn't name" (29143), "a new leading AI developer" (8101), and British Columbia negotiations said to support more than $0.5 billion of ARR (16666).

Concentration has eased but remains high. Microsoft's expected $1.9 billion of average annual revenue (8996) was about 61% of the $3.1 billion of ARR under contract in May 2026 (9183).

The revenue arithmetic broadly reconciles. Microsoft's $1.9 billion times five years gives $9.5 billion against the $9.7 billion headline, with the gap plausibly explained by the ramp. NVIDIA's $3.4 billion over five years adds about $0.68 billion a year. Together these leave roughly $0.5 billion, consistent with about $0.4 billion at Prince George (16665) plus legacy contracts. The July 2026 $2.8 billion (360) has no disclosed ARR conversion. At the roughly four-year weighted term cited (8108), it implies about $0.7 billion a year, which would bring Microsoft's share down to around half. One inconsistency stands out. A secondary source names Fluidstack and Fireworks AI as part of the July deal (8117), yet both were already announced in November 2025 (20247). IREN's own wording names no counterparties for that deal (360). The record supplied ends mid-entry at claim 361, so any later disclosures are not reflected here.

IREN's customer record splits cleanly into a small number of signed, identified contracts and a wider layer of unnamed or framework language. The anchor is the Microsoft GPU services agreement dated November 2, 2025, disclosed with a contracting subsidiary, site and term (claims 20307, 4157, 1266). It has moved beyond paper: Horizon 1 was delivered to and accepted by Microsoft in August 2026 (claims 8673, 35). The NVIDIA $3.4bn five-year AI Cloud contract is likewise described as signed, with a specific GPU type, a 60MW Childress footprint and an early-2027 ramp (claims 9180, 9287). The accompanying 5GW 'strategic partnership' is a collaboration framework with no stated economics (claims 11077, 9289). NVIDIA is also a $2.1bn investor whose rights vest only as NVIDIA infrastructure is deployed (claim 9179). That makes NVIDIA simultaneously customer, supplier ecosystem and capital provider, and readers should weigh that circularity. The Dell agreements (claims 20308, 4161, 8992, 1297) are purchase obligations guaranteed by the parent (claim 4164), not revenue. Bitmain, Kraken, BC Hydro and WEKA (claims 6374, 12183, 12198, 12005) are vendors or technology partners.

The unnamed layer is persistent. In May 2025 revenue came from white-labelled compute to unidentified 'leading US AI cloud providers' at a $28m run-rate (claims 29824, 28325). Horizon 1 customers at that time were only 'undergoing due diligence and contractual negotiations' (claims 29802, 29819). In February 2026, more than $0.5bn of British Columbia ARR was still in negotiation, against $0.4bn under contract (claims 16666, 16665). An unnamed 'frontier AI lab' and 'a new leading AI developer' appear in mid-2026 (claims 29143, 8101). The $2.8bn July tranche is attributed to companies 'like' Figure AI, Fluidstack, Fireworks AI and Hume AI, without a per-counterparty split (claims 360, 8117).

Concentration moved from effectively zero named customers to a single dominant one, with only modest dilution since. On the arithmetic, $9.7bn over five years implies roughly $1.94bn a year, which reconciles with the $1.9bn expected ARR (claims 20245, 8996). NVIDIA's $3.4bn over five years adds about $0.68bn. Prince George adds $0.4bn. Together these come to roughly $3.0bn, which is broadly consistent with the stated $3.1bn ARR under contract (claim 9183). On that base, Microsoft represents about 61% of contracted ARR, and Microsoft plus NVIDIA about 83%. The July $2.8bn at a roughly four-year weighted term (claim 8108) would add around $0.7bn, but no updated ARR total is on record to confirm it. Several dollar figures in the record are inconsistently scaled (for example $9,700 versus $9,700,000,000), so totals were recomputed from the verbatim quotes. Claim 8543 carries no quote and was not relied on.

IREN's signed-customer record is narrower than its partnership language suggests. The only fully documented customer agreement is the five-year, $9.7 billion Microsoft GPU services contract dated November 2, 2025 (claims 4157, 20307, 1266). It now has operating evidence, since Horizon 1 was delivered to and accepted by Microsoft in August 2026 (claims 8673, 35, 7454). The $3.4 billion five-year NVIDIA AI Cloud contract is also described as signed (claims 9180, 11076, 8995). The 5GW NVIDIA 'strategic partnership' attached to it, however, is framework language with no stated economics (claims 11077, 9289). The $2.1 billion NVIDIA investment vests only as GPU infrastructure is deployed (claim 9179), and 'NVIDIA Preferred Partner' (claim 20300) is a program tier, not a contract. Many named counterparties are suppliers or vendors rather than customers: Bitmain (claims 14392, 6374, 6152), Dell (claims 20308, 4161, 1297, 8992), BC Hydro (claims 12182, 12198), Kraken (claim 12183), WEKA (claim 12005) and Mirantis (claims 8768, 8769). Together AI, Fluidstack and Fireworks AI are named as multi-year customers (claim 20247), but no values are attached. Poolside appears only as a service commencement (claim 6285). Several counterparties are anonymous. The $28 million white-label run-rate comes from unnamed 'leading US AI cloud providers' (claim 28325). The roughly $0.4 billion Prince George ARR has no named customer (claim 16665). The July list includes 'a new leading AI developer' (claim 8101), and the August contract is with an unnamed frontier lab (claim 29143). The more than $0.5 billion in British Columbia ARR is explicitly still in negotiation (claim 16666) and should not be counted as contracted. Concentration has moved from a business with no customers, mining into the Bitcoin network and selling through Kraken, to one dominated by a single hyperscaler. The arithmetic largely reconciles. Microsoft's $9.7 billion over five years implies about $1.94 billion a year, consistent with the stated $1.9 billion ARR (claims 20245, 8996). NVIDIA's contract adds about $0.68 billion a year. Together these are roughly $2.6 billion of the $3.1 billion ARR under contract reported in May (claim 9183), which leaves about $0.5 billion, a gap Prince George largely fills. On that base, Microsoft is about 61% of contracted ARR and NVIDIA about 22%. The July $2.8 billion in contracts (claims 360, 8117), spread over a weighted term of about four years (claim 8108), implies roughly $0.7 billion of additional ARR, which would dilute Microsoft toward about half. No updated ARR total confirms that figure, and the supplied record is cut off at claim 8594. Prepayment terms improved from 20% under Microsoft (claim 20245) to about 45% of GPU capex on recent deals (claim 8595). That still leaves Dell commitments of $5.8 billion and $1.6 billion largely funded by IREN (claims 4161, 1297).

IREN's signed-customer record is narrower than its partnership vocabulary suggests. Microsoft is the only counterparty backed by an executed agreement, a disclosed value and verified performance. The $9.7bn, five-year GPU services contract dated November 2, 2025 is described in filing language (claims 20307, 4157, 1266), carries a 20% prepayment (20245), and passed its first delivery test when Microsoft accepted Horizon 1 (8673, 35). NVIDIA's $3.4bn five-year AI Cloud contract is also described as signed (9180, 11076). The surrounding '5GW strategic partnership' is framework language for 'collaboration' across a pipeline (9289, 11077), not a revenue commitment. NVIDIA is also an investor whose $2.1bn rights vest only as its GPUs are deployed (9179), so its customer, supplier-ecosystem and shareholder roles overlap. Several other named counterparties are suppliers or vendors rather than customers: Bitmain (14392, 6374, 6152), Dell (20308, 4161, 8992), WEKA (12005), BC Hydro (12182) and Kraken (12183).

Naming is uneven below the top two. Together AI, Fluidstack and Fireworks AI are named as multi-year contracts but carry no values (20247). The July $2.8bn tranche (360) is attributed to Figure AI, Fluidstack, Fireworks AI and Hume AI only through a third-party summary (8117). IREN's own roster includes 'a new leading AI developer' (8101), and a 'frontier AI lab it couldn't name' appears in August (29143). The $28m white-label run-rate comes from unidentified 'leading US AI cloud providers' (28325). The Prince George $0.4bn ARR names no counterparty (16665), and the further '>$0.5bn' is explicitly still in negotiation (16666), so it should not be counted as contracted.

Concentration has moved from no real customer in the mining era to heavy single-name exposure. Microsoft's $1.9bn expected ARR (20245, 8996) was roughly 61% of the $3.1bn ARR under contract reported in May 2026 (9183). That May figure broadly reconciles. Microsoft's $1.9bn, plus about $0.68bn a year from NVIDIA's $3.4bn over five years, plus Prince George's $0.4bn, gives about $3.0bn, leaving a small residual for smaller AI cloud customers. Two caveats apply. NVIDIA revenue does not ramp until early 2027 (9287). The July $2.8bn, at the roughly four-year weighted term (8108), implies about $0.7bn a year that no updated ARR figure on record yet confirms. Several records also show unit inconsistencies ($9,700 versus $9.7bn), and the claim list was truncated at claim 361.

IREN's signed-customer record is narrower and more concentrated than its roster language suggests. The Microsoft agreement is the clearest signed contract on file. It is dated November 2, 2025, it is described as signed, and it was executed through the IE US Hardware 3 subsidiary (claims 20307, 4157, 1266). It has since progressed to delivery and acceptance of Horizon 1 (claims 35, 8673), so it has moved from paper to performance. The NVIDIA $3.4bn five-year contract is also described as signed (claims 9180, 11103). It sits alongside a 5GW "strategic partnership" that is explicitly a collaboration framework, not a commitment (claims 11077, 9289), and alongside a $2.1bn NVIDIA investment whose rights vest only as NVIDIA infrastructure is deployed (claim 9179). That makes NVIDIA simultaneously customer, investor and supplier-ecosystem partner, a circularity diligence should price. The Dell agreements of $5.8bn and $1.6bn (claims 4161, 1297) are binding, but they are procurement obligations guaranteed by the parent (claim 4164), not revenue.

Named counterparties have widened, but disclosure depth has not kept pace. Together AI, Fluidstack and Fireworks AI were cited as "multi-year contracts" without values (claim 20247). The July $2.8bn tranche names Figure AI, Fluidstack, Fireworks AI and Hume AI but folds in "a new leading AI developer" (claims 360, 8101, 8117). A further multiyear contract is with a frontier lab IREN "couldn't name" (claim 29143). The earlier $28m run-rate came from unidentified "leading US AI cloud providers" via white-label resale (claim 28325). In British Columbia, ~$0.4bn ARR is "under contract" with no counterparty named, and the additional >$0.5bn is explicitly still in "negotiations" (claims 16665, 16666). "NVIDIA Preferred Partner" (claim 20300) and the WEKA tie-up (claim 12005) are programme or vendor relationships, not customer contracts.

The counterparty base has shifted from a mining-era profile of Bitmain, BC Hydro and Kraken liquidation (claims 14392, 12198, 12183) to hyperscaler dependence. The arithmetic largely reconciles. Microsoft's $9.7bn over five years implies roughly $1.94bn a year, consistent with the stated $1.9bn (claims 20245, 8996). NVIDIA's $3.4bn implies about $0.68bn. Adding Prince George's $0.4bn gives roughly $3.0bn, close to the $3.1bn ARR under contract (claim 9183). On that base, Microsoft is about 60% and NVIDIA about 22%, so two counterparties carry over four-fifths of contracted ARR. The July $2.8bn at a roughly four-year weighted term (claim 8108) would add perhaps $0.7bn and dilute that share somewhat. However, no updated ARR figure on record confirms it, and part of that value sits with an unnamed counterparty.

IREN's signed-customer record is thinner than its partnership vocabulary suggests, but the core deals are real. The anchor is the Microsoft GPU services contract dated November 2, 2025, which is documented at the subsidiary level through IE US Hardware 3 (claims 20307, 4157, 1266). It became operational when Horizon 1, a 50MW tranche, was delivered and accepted (claims 35, 8673, 7454). The NVIDIA $3.4bn five-year AI Cloud contract is described as signed (claims 9180, 11076) and has a matching Dell hardware purchase (claims 1297, 8992). The accompanying 5GW strategic partnership (claims 11077, 9289) is framework language, however, and NVIDIA's $2.1bn investment vests only as deployments occur (claim 9179). That makes NVIDIA customer, supplier ecosystem and contingent investor at once, a circularity diligence should weigh. Dell (claims 20308, 4161), Bitmain (claims 14392, 6374, 6152), BC Hydro (claims 12182, 12198) and Kraken (claim 12183) are vendors or service providers, not revenue counterparties. WEKA (claim 12005) is a technology partnership, and the Mirantis and NVIDIA program items (claims 8768, 8769, 8543) are ecosystem credentials, not contracts. Smaller counterparties are named but unquantified. These include Poolside (claim 6285), Together AI, Fluidstack and Fireworks AI (claim 20247), and Perplexity, Figure AI, Fal AI and Hume AI (claim 8101). Several counterparties are never identified: the 'leading US AI cloud providers' behind the white-labelled Hopper revenue (claim 28325), 'a new leading AI developer' (claim 8101), and a frontier AI lab IREN 'couldn't name' (claim 29143). The $2.8bn July package (claims 360, 8117) is not allocated by customer. In British Columbia, only about $0.4bn of ARR is under contract (claim 16665); the further >$0.5bn remains in negotiation (claim 16666) and should not be counted. Concentration has moved from no meaningful customer base in the mining era to heavy dependence on Microsoft. Microsoft's expected $1.9bn of annual revenue (claims 20245, 8996) was about 61% of the $3.1bn of contracted ARR reported in May (claim 9183). Together with NVIDIA, roughly 83% of that base sat with two counterparties. The July contracts dilute this somewhat, but no updated ARR total appears in the record. The arithmetic largely reconciles. $9.7bn over five years is about $1.94bn a year, consistent with $1.9bn. NVIDIA's $3.4bn over five years adds about $0.68bn. Prince George adds about $0.4bn and Hopper about $0.03bn. The total is roughly $3.0bn against the stated $3.1bn, leaving a small residual for the smaller named customers. At a weighted average term of about four years (claim 8108), the July package implies roughly $0.7bn of additional ARR. That figure is an inference, not a disclosed number.

IREN's disclosed counterparties fall into three tiers, and only the top tier is signed revenue. The anchor is the Microsoft GPU services contract dated November 2, 2025. It is executed, with a $9.7 billion total, a five-year term and a 20% prepayment (claims 4157, 20307, 20245, 1266), and it now has delivery evidence: Microsoft accepted Horizon 1 in August 2026 (claims 8673, 35). The $3.4 billion five-year NVIDIA AI Cloud contract is also described as signed (claims 11076, 9180). The two Dell agreements, $5.8 billion and $1.6 billion, are supplier obligations rather than customers (claims 20308, 1297), and IREN's parent guarantee makes them balance-sheet commitments (claim 4164).

The second tier is framework or status language that carries no stated revenue. This includes the 5GW NVIDIA strategic partnership (claims 11077, 9289), the Preferred Partner and Exemplar Cloud designations (claims 20300, 8543), the WEKA 'partnering' (claim 12005), and the $2.1 billion NVIDIA investment, which vests only as NVIDIA infrastructure is deployed (claim 9179). That vesting structure also makes NVIDIA customer, investor and ecosystem gatekeeper at once. The Mirantis items describe Mirantis's own credentials, not IREN's customers (claims 8768, 8769). The third tier is unidentified counterparties. These are the white-label 'leading US AI cloud providers' (claim 28325), the 'new leading AI developer' (claim 8101), the frontier lab IREN 'couldn't name' (claim 29143), and a generic 'broad range of customers' (claim 20306). The July $2.8 billion tranche is attributed only loosely to 'companies like' Figure AI, Fluidstack, Fireworks and Hume (claims 360, 8117), with no per-counterparty split.

Concentration has shifted sharply. Through 2024 the named relationships were suppliers and utilities, namely Bitmain, BC Hydro and Kraken (claims 14392, 12182, 12183, 6152), with Poolside as the first GPU customer (claim 6285). Since November 2025, Microsoft and NVIDIA together account for $13.1 billion of the roughly $15.9 billion in disclosed contract value. The July tranche is the first real diversification.

The arithmetic mostly reconciles. $9.7 billion over five years implies $1.94 billion a year, matching the stated $1.9 billion (claims 20245, 8996). Adding about $0.68 billion a year from NVIDIA and about $0.4 billion from Prince George (claim 16665) gives roughly $3.0 billion, against the $3.1 billion ARR under contract stated in May 2026 (claim 9183). The >$0.5 billion still 'in negotiation' (claim 16666) should not be counted. The July $2.8 billion has no disclosed ARR. At the roughly four-year weighted term (claim 8108) it would imply around $0.7 billion a year, but that figure is our inference, not company guidance. Hardware outlays equal roughly 47-60% of the related contract values, so the margin depends on actual utilization. Claim 7454, a '50MW' Microsoft reference, is truncated and conflicts with the Horizon framing, so it should be checked against the source.

IREN's customer record is anchored by one firmly documented agreement, the five-year, $9.7bn Microsoft GPU services contract dated November 2, 2025. It is described in filing language naming the contracting subsidiary and counterparty (claims 4157, 20307, 1266), and it has moved past paper: Microsoft has accepted delivery of Horizon 1 (claims 35, 8673). The $3.4bn five-year NVIDIA AI Cloud contract is also described as signed (claims 9180, 11076), with a dedicated Dell hardware purchase supporting it (claims 1297, 8992). Readers should keep it separate from the 5GW NVIDIA 'strategic partnership' (claims 11077, 9289, 362), which is collaboration language with no disclosed revenue terms. The $2.1bn NVIDIA investment, which vests only as NVIDIA infrastructure is deployed (claim 9179), puts NVIDIA in the positions of customer, investor and technology sponsor at once. That circularity deserves scrutiny. Several large figures on the record are cost-side commitments rather than customers: the Dell purchase agreements of $5.8bn and $1.6bn (claims 4161, 1297), the Bitmain hardware contracts (claims 14392, 6374, 6152) and BC Hydro power (claim 12198). Identification is uneven. Microsoft and NVIDIA are named in primary disclosures. Together AI, Fluidstack and Fireworks AI appear by name (claim 20247), but the July $2.8bn tranche is attributed in company language only to 'leading AI developers' (claim 360), and the names attached to it come from secondary coverage phrased as 'companies like' (claim 8117). The company has also described counterparties without naming them: white-labelled 'US AI cloud providers' (claim 28325), 'a new leading AI developer' (claim 8101), an unnamed frontier lab (claim 29143) and the Prince George customers behind roughly $0.4bn of contracted ARR (claim 16665). The more than $0.5bn of British Columbia ARR still in 'negotiations' (claim 16666) is not contracted revenue. Concentration has moved from no customer base at all, when mined bitcoin was liquidated via Kraken (claim 12183), to a single early AI client in Poolside (claim 6285), to heavy dependence on Microsoft. On the arithmetic, Microsoft's $1.9bn expected annual revenue (claims 20245, 8996) matches $9.7bn spread over five years. Adding roughly $0.68bn a year from NVIDIA and $0.4bn from Prince George gives about $2.98bn, which reconciles closely with the $3.1bn of ARR under contract reported in May 2026 (claim 9183). On that basis Microsoft accounted for about 61% of contracted ARR. If the July $2.8bn is spread over the roughly four-year weighted average term (claim 8108), it adds approximately $0.7bn and would dilute Microsoft's share toward half. No updated ARR total appears on the record to confirm that figure.

IREN's signed-customer record is far more concentrated than its expanding roster suggests. Before 2024 the company had no named customers in the commercial sense. Bitmain (claims 14392, 6374, 6152) and BC Hydro (12182, 12198) are suppliers, and mined Bitcoin was sold through Kraken (12183). The first customer-like disclosure is the commencement of GPU cloud service with Poolside AI (6285). The WEKA tie-up (12005) is a vendor partnership, not revenue. The decisive signed contract is Microsoft's five-year, $9.7bn GPU services agreement (4157, 20307, 1266). Delivery and acceptance of Horizon 1 (8673, 35) move it from paper to performance, with three further 50MW tranches still to come (8622). NVIDIA's $3.4bn five-year contract (11076, 9287, 9180) is also signed. Its companion '5GW strategic partnership' (11077, 9289), however, is framework language describing collaboration, not committed capacity. NVIDIA's $2.1bn investment rights vest only as its infrastructure is deployed (9179), so NVIDIA is simultaneously customer, investor and architectural sponsor, and that circularity deserves scrutiny. Several other names arrive mainly as lists rather than as individually priced contracts: Together AI, Fluidstack and Fireworks AI (20247), and later Perplexity, Figure AI, Fal AI and Hume AI (8101, 8117). Two counterparties remain unidentified, 'a new leading AI developer' (8101) and an unnamed 'frontier AI lab' (29143). The 'hyperscalers to AI-native enterprises' framing (20306) is descriptive rather than contractual. The arithmetic broadly reconciles. $9.7bn over five years implies roughly $1.94bn a year, matching the stated $1.9bn ARR (20245, 8996). NVIDIA's $3.4bn implies about $0.68bn. Together these give about $2.6bn of the $3.1bn ARR under contract (9183). The residual of about $0.5bn is consistent with Prince George's ~$0.4bn (16665) plus smaller contracts. The separate '>$0.5bn' in British Columbia was explicitly still under negotiation (16666) and should not be counted as signed. On that basis Microsoft is about 61% of contracted ARR, and Microsoft plus NVIDIA about 84%. The July $2.8bn tranche (360) would dilute this only modestly. At the roughly four-year weighted term (8108) it adds about $0.7bn of ARR, and no per-counterparty split was disclosed. The Dell purchases of $5.8bn and $1.6bn (4161, 1297) are capital commitments that mirror these contracts. They are partly funded by customer prepayments of 20% for Microsoft and roughly 45% on recent deals (20245, 8595). The claim record supplied ends mid-entry at 8543, so later disclosures may be missing.

IREN's signed customer paper is concentrated in two contracts. The rest of the record mixes supplier agreements, framework partnerships and unnamed counterparties. The Microsoft agreement is the clearest signed customer contract. It is a dated November 2, 2025 GPU services contract worth $9.7bn over five years, with a 20% prepayment (claims 4157, 20307, 20245, 1266), and it has moved to performance, since Horizon 1 has been delivered and accepted (claims 35, 8673). The NVIDIA $3.4bn five-year AI Cloud contract is also described as signed (claims 9180, 11076). The 5GW NVIDIA 'strategic partnership' alongside it is framework language with no stated economics (claims 11077, 9289, 362). The $2.1bn NVIDIA investment vests only as deployments occur (claim 9179). The Dell agreements of $5.8bn and $1.6bn (claims 4161, 1297) and the Bitmain contracts (claims 14392, 6374, 6152) are purchase obligations, not revenue. BC Hydro, Kraken and WEKA are utility, liquidity and software relationships rather than customers (claims 12198, 12183, 12005).

The named-versus-anonymous split matters. Together AI, Fluidstack and Fireworks AI were named as multi-year contract holders without values (claim 20247), and Poolside appears only as a service commencement (claim 6285). Several counterparties are not identified at all: the $28m run-rate white-label revenue from 'leading US AI cloud providers' (claim 28325), the 'new leading AI developer' (claim 8101) and the frontier lab IREN 'couldn't name' (claim 29143). The July $2.8bn tranche is attributed to Figure AI, Hume AI and others only by secondary reporting (claims 360, 8117), so allocation by counterparty is unverified.

Concentration has fallen but remains high. Microsoft's $1.9bn expected annual revenue (claim 8996) was roughly 61% of the $3.1bn ARR under contract reported in May (claim 9183), and Microsoft plus NVIDIA together were about 83%. The May arithmetic broadly reconciles. Microsoft at $1.9bn, NVIDIA at about $0.68bn a year and Prince George at $0.4bn (claim 16665) sum to about $3.0bn, with the residual plausibly covered by smaller contracts. The >$0.5bn British Columbia figure was still in negotiation and should not be counted (claim 16666). The July $2.8bn, spread over the roughly four-year weighted term (claim 8108), would add about $0.7bn of ARR and imply near $3.8bn, which would bring Microsoft's share toward 50%. No updated ARR figure appears in this record to confirm that. The claim list was also truncated at claim 7454, so later disclosures were not assessed.

IREN's record divides into three groups: signed revenue contracts with named counterparties, signed contracts whose counterparties are not identified, and partnership or negotiation language that should not be counted as backlog. The anchor of the first group is Microsoft. IREN described the $9.7bn five-year GPU services agreement as signed (claim 20307), filed it at the subsidiary level (claim 4157), and later reported that Horizon 1 was delivered and accepted (claims 35, 8673), which moves it from paper to performance. The $3.4bn five-year NVIDIA contract is also described as signed (claim 9180). The 'broader 5GW strategic partnership' attached to it (claims 11077, 9289) is collaboration language with no stated revenue. NVIDIA's $2.1bn investment vests only as GPU infrastructure is deployed (claim 9179), so it is contingent capital rather than a customer commitment. The Dell agreements (claims 4161, 20308, 1297) and the Bitmain contracts (claims 14392, 6374) are supplier obligations. WEKA (claim 12005) is a technology vendor. None of these are customers.

The second group is where identification thins out. The 2025 white-labelled revenue came from unnamed 'leading US AI cloud providers' (claim 28325). Prince George's ~$0.4bn ARR is 'under contract' without named parties (claim 16665). The adjacent >$0.5bn is explicitly still in 'negotiations' (claim 16666). The July $2.8bn was attributed by IREN only to 'leading AI developers' (claim 360). Names such as Figure AI and Hume AI come from a third-party characterization (claim 8117) and from a roster that also includes 'a new leading AI developer' (claim 8101). The August frontier-lab contract is undisclosed by name and by value (claim 29143).

Concentration has moved from a single Bitcoin liquidity venue (claim 12183) and one early GPU client, Poolside (claim 6285), to heavy dependence on Microsoft. At $1.9bn of expected ARR (claims 20245, 8996), Microsoft was roughly 61% of the $3.1bn under contract in May 2026 (claim 9183). That share is diluting as NVIDIA and the July cohort ramp.

The arithmetic broadly reconciles. Microsoft's $1.9bn, about $0.68bn from NVIDIA ($3.4bn over five years), and ~$0.4bn at Prince George sum to about $3.0bn. That is consistent with the $3.1bn figure once smaller named contracts (claim 20247) are included. Adding the July $2.8bn at the disclosed ~4-year weighted term (claim 8108) implies roughly $0.7bn more, or about $3.8bn. That total excludes the unpriced frontier-lab deal and the unsigned British Columbia negotiations, neither of which should be counted until terms are disclosed.

IREN's customer record splits cleanly into a small number of signed, quantified contracts and a wider layer of supplier, framework and loosely described relationships. The signed anchor is Microsoft: a five-year, $9.7bn GPU services contract dated November 2, 2025 (claims 4157, 20307, 1266), with a 20% prepayment and $1.9bn expected ARR (20245). Execution is now evidenced, since Horizon 1 has been delivered to and accepted by Microsoft (35, 8673) as the first of four 50MW deployments (8622). The second signed contract is the five-year, $3.4bn NVIDIA AI Cloud contract (11076, 9180), backed by a $1.6bn Dell purchase agreement (8992, 1297). By contrast, the 5GW NVIDIA strategic partnership (11077, 9289) is framework language with no committed revenue. The $2.1bn NVIDIA investment vests only as NVIDIA infrastructure is deployed (9179), which makes NVIDIA simultaneously customer, supplier ecosystem and contingent investor. That circularity deserves scrutiny. Bitmain, Dell, BC Hydro, Kraken and WEKA (14392, 4161, 12182, 12183, 12005) are vendor or service relationships, not customers, and the Mirantis items (8768, 8769) describe Mirantis's own customer base rather than IREN's. Unnamed counterparties persist throughout. Examples include white-labelled compute for leading US AI cloud providers (28325), a customer base spanning hyperscalers to AI-native enterprises (20306), and a new leading AI developer (8101). The $2.8bn July package (360, 8594) names Figure AI, Fluidstack, Fireworks AI and Hume AI only illustratively (8117), with no per-counterparty allocation. Concentration has moved from a single anchor toward a still top-heavy book. In May 2026, Microsoft's $1.9bn of the $3.1bn ARR under contract (9183) was roughly 61%. The July additions dilute that toward about half, assuming a roughly four-year term (8108). The arithmetic broadly reconciles. Microsoft at about $1.94bn, NVIDIA at about $0.68bn, Prince George at about $0.4bn (16665) and earlier contracts with Together AI, Fluidstack and Fireworks AI (20247) together approximate $3.1bn. However, the more than $0.5bn of British Columbia ARR was still described as under negotiation (16666) and should not be counted as signed. Prepayments rising from 20% to roughly 45% of GPU capex (8595) improve funding quality, but the data provided ends mid-record (8553), so later disclosures were not assessed.

IREN's signed-customer record is narrower and more concentrated than its partnership vocabulary suggests. The anchor is a single executed instrument: the five-year, $9.7bn Microsoft GPU services contract dated November 2, 2025, with the counterparty, site (Childress, Texas), tranche structure and 20% prepayment all disclosed (claims 4157, 20307, 20245, 1266). It is now partly performing, since Horizon 1, a 50MW tranche, has been delivered and accepted (claims 8673, 35, 7454). The second named, signed contract is the five-year, $3.4bn AI Cloud agreement with NVIDIA (claims 11076, 9180, 9287). It sits beside a '5GW strategic partnership' (claims 11077, 9289, 362) that is framework language with no stated revenue, and a $2.1bn NVIDIA investment whose vesting is tied to IREN deploying NVIDIA infrastructure (claim 9179). That makes NVIDIA simultaneously customer, supplier ecosystem and prospective shareholder. Dell appears only as a vendor, at $5.8bn and $1.6bn (claims 4161, 1297), as do Bitmain, WEKA and Kraken (claims 14392, 6152, 12005, 12183). Before 2025, the only named AI customer was Poolside (claim 6285). The $28m white-label run rate was attributed only to unnamed 'leading US AI cloud providers' (claim 28325), and the 'broad range of customers' framing named nobody (claim 20306). Later rosters list Perplexity, Figure AI, Fal AI and Hume AI (claims 8101, 363), but none of them is tied to a disclosed contract value. The July $2.8bn tranche includes 'a new leading AI developer' who is left unnamed (claims 360, 8101), and the attribution of that deal to specific firms comes from secondary 'companies like' phrasing (claim 8117). Mirantis's 1,500 enterprise customers belong to Mirantis, not IREN (claim 8768). The British Columbia figure of more than $0.5bn ARR is still in negotiation and is not signed (claim 16666). The arithmetic broadly reconciles. Microsoft's $1.9bn of expected annual revenue (claims 20245, 8996) plus about $0.68bn from NVIDIA's $3.4bn over five years, plus roughly $0.4bn at Prince George (claim 16665), comes to about $3.0bn. That is consistent with the stated $3.1bn of ARR under contract (claim 9183) if small contracts fill the gap. On that base, Microsoft represents roughly 60% of contracted ARR, and Microsoft plus NVIDIA about 83%. The July $2.8bn, at the roughly four-year weighted term IREN cites (claim 8108), would add about $0.7bn. That would dilute Microsoft toward half, but it would not change the two-counterparty dependence.

IREN's customer record splits cleanly into a few documented, signed contracts and a wider layer of partnership and roster language. The anchor is the Microsoft GPU services agreement dated November 2, 2025, which is described in filing-grade detail as a five-year, $9.7 billion contract with a named contracting subsidiary, tranche structure and Childress location (claims 4157, 20307, 1266). It has since moved past paper: Horizon 1, the first of four facilities, was delivered to and accepted by Microsoft (claims 35, 8673, 7454). The five-year, $3.4 billion NVIDIA AI Cloud contract is also described as signed (claims 9180, 11076). The 5GW NVIDIA 'strategic partnership' is a different instrument. It is framed as collaboration across a pipeline (claims 9289, 362) and carries no disclosed value or term. NVIDIA's $2.1 billion investment rights vest only as NVIDIA infrastructure is deployed (claim 9179), which makes NVIDIA customer, supplier-aligned partner and prospective shareholder at the same time. Earlier ties are softer still: the WEKA 'partnering' (claim 12005), the Poolside service 'commencement' (claim 6285) and the 'NVIDIA Preferred Partner' label (claim 20300) carry no economics. The Dell agreements of $5.8 billion and $1.6 billion (claims 4161, 1297) are supplier obligations guaranteed by the parent (claim 4164), not revenue. The Mirantis figures describe Mirantis's customers, not IREN's (claim 8768). Naming is uneven. Microsoft and NVIDIA are fully identified. The July $2.8 billion tranche is attributed by the company only to 'leading AI developers' (claim 360); the allocation to Figure AI, Fluidstack, Fireworks AI and Hume AI comes from a secondary 'companies like' framing (claim 8117), and the roster still includes an unnamed 'new leading AI developer' (claim 8101). Hyperscaler breadth is asserted without names (claim 20306). Concentration has fallen but remains heavy: of roughly $15.9 billion in disclosed contract value, Microsoft is about 61% and Microsoft plus NVIDIA about 82%. The revenue arithmetic broadly reconciles. $9.7 billion over five years implies about $1.94 billion a year, matching the $1.9 billion ARR guidance (claims 20245, 8996). NVIDIA adds about $0.68 billion and Prince George about $0.4 billion (claim 16665), which sums near the $3.1 billion ARR under contract (claim 9183). The British Columbia figure above $0.5 billion is still under negotiation (claim 16666) and should not be counted. At a roughly four-year weighted term (claim 8108), the July $2.8 billion implies about $0.7 billion of incremental ARR, unverified against named counterparties.

IREN's customer record falls into three tiers. The first is signed contracts with named counterparties. The second is named counterparties with no disclosed contract value. The third is framework or unidentified language. Only the Microsoft agreement sits fully in the first tier. It is a dated November 2, 2025 GPU services contract (claims 4157, 1266) valued at $9.7 billion with a 20% prepayment (20245), and delivery and acceptance of Horizon 1 has been confirmed (35, 8673). Horizon 1 is the first of four facilities (8553). The NVIDIA $3.4 billion five-year contract (11076, 9180) is also described as signed. The 5GW 'strategic partnership' attached to it (11077, 9289, 362) is collaboration language with no revenue commitment, and it is entangled with a $2.1 billion NVIDIA investment whose vesting depends on deployment (9179). That makes NVIDIA simultaneously customer, supplier-aligned partner and prospective shareholder.

Naming is uneven. Together AI, Fluidstack and Fireworks AI were identified in November 2025 without values (20247). The July 2026 $2.8 billion tranche is attributed first-party only to 'leading AI developers' (360). The specific names come from a secondary source (8117) and a roster listing (8101, 363), and that roster also includes an unnamed 'new leading AI developer' (8101). The earlier $28 million white-label run-rate came from unidentified 'leading US AI cloud providers' (28325). The 'hyperscalers to AI-native enterprises' framing (20306) is generic. Dell (20308, 8992) is a $7.4 billion cumulative supplier, not a customer, and should not be read as demand. The Mirantis items (8768, 8769) describe Mirantis's own customers and partnerships, not IREN's.

Concentration has moved sharply. Pre-2024 counterparties were suppliers and utilities: Bitmain, BC Hydro and Kraken (14392, 12182, 12183). The roster then shifted to a single hyperscaler. Of roughly $15.9 billion in announced contract value, Microsoft is about 61%, and Microsoft plus NVIDIA is about 82%.

The arithmetic broadly reconciles. $9.7 billion over five years implies about $1.9 billion ARR, matching 8996. The NVIDIA contract implies about $0.68 billion ARR. Add the $0.4 billion contracted at Prince George (16665) and the total is about $3.0 billion, close to the $3.1 billion ARR under contract reported in May (9183). The small residual plausibly covers the November neocloud contracts. Two cautions apply. The '>$0.5bn ARR' tied to British Columbia was explicitly still under negotiation (16666) and should not be counted. The July $2.8 billion, at the roughly four-year weighted term (8108), would add about $0.7 billion ARR, but no updated ARR-under-contract figure appears on record to confirm that.

IREN's signed-customer record is shorter than its partnership vocabulary suggests. Before late 2023 the named counterparties were suppliers and service venues, not customers. These were Bitmain for miners (claims 14392, 6374, 6152), BC Hydro for power (12198, 12182) and Kraken for selling mined Bitcoin (12183). The WEKA 'partnering' language (12005) describes a software vendor. Poolside AI is the first named cloud customer, but it was disclosed only as a 'commencement of GPU cloud service' with no contract value (6285). The two firm, valued, named contracts are Microsoft and NVIDIA. Microsoft is a signed five-year, $9.7bn agreement (20307, 4157, 1266), and Horizon 1 has been delivered to and accepted by Microsoft (8673, 35). NVIDIA is a signed five-year, $3.4bn contract (9180, 11076). The surrounding NVIDIA '5GW strategic partnership' (11077, 9289) is framework language with no committed volume. NVIDIA's $2.1bn investment rights vest only as its GPUs are deployed (9179), so NVIDIA is at once customer, technology supplier and contingent equity holder. Readers should weigh that circularity. The Dell agreements (4161, 1297) are purchase commitments, meaning costs rather than revenue. Several counterparties are unnamed or not yet signed. The $28m run-rate comes from white-labelled compute for unnamed 'leading US AI cloud providers' (28325). The July 2026 roster includes 'a new leading AI developer' (8101). The British Columbia figure of more than $0.5bn ARR is described as 'remaining contract negotiations' (16666) and should not be counted as signed. Prince George's roughly $0.4bn is 'under contract' but carries no counterparty names (16665). The $2.8bn July tranche names Figure AI, Fluidstack, Fireworks AI and Hume AI only as 'companies like' (8117, 360), with no allocation by customer. Concentration has moved from no external customers to heavy dependence on Microsoft. Its $1.9bn expected annual revenue (20245, 8996) was about 61% of the $3.1bn ARR under contract as of May 2026 (9183). The arithmetic broadly reconciles. Microsoft's $9.7bn over five years is about $1.94bn a year. NVIDIA's $3.4bn over five years is about $0.68bn a year. Adding Prince George's $0.4bn gives roughly $2.98bn, close to the stated $3.1bn, though it is not disclosed whether Prince George overlaps with Together AI, Fluidstack or Fireworks AI (20247). The July $2.8bn, at the roughly four-year weighted average term (8108), implies about $0.7bn of added ARR. That would cut Microsoft's share to around half, but no updated ARR total appears in the record to confirm it. Claim 362 is truncated and could not be assessed.

IREN's record divides cleanly into a small set of signed, valued customer contracts and a wider set of framework, partnership and unnamed-counterparty language.

The anchor is the Microsoft agreement. It is a dated, signed GPU services contract (claims 20307, 4157, 1266) worth $9.7bn over five years. It has been operationally validated by Microsoft's acceptance of Horizon 1 (claims 8673, 35). The NVIDIA $3.4bn five-year AI Cloud contract is also described as signed (claim 9180) and is backed by a $1.6bn Dell hardware purchase (claims 8992, 1297). The 5GW NVIDIA "strategic partnership" is different. It is framework language about collaboration across a pipeline (claims 11077, 9289, 8642) with no committed revenue attached. The same applies to the earlier WEKA "partnering" (claim 12005), Exemplar Cloud status (claim 8543) and the Mirantis references (claims 8768, 8769), which describe Mirantis's own customers rather than IREN's. Dell and Bitmain (claims 4161, 14392, 6152) are supplier contracts, not revenue.

Naming is uneven. Together AI, Fluidstack and Fireworks AI were identified in November 2025 without contract values (claim 20247). The July $2.8bn tranche is attributed to "leading AI developers" (claim 360). The fuller name list comes from a third-party summary (claim 8117), and the company's own roster includes "a new leading AI developer" left unidentified (claim 8101). In British Columbia, about $0.4bn ARR is under contract (claim 16665), but the further >$0.5bn is explicitly still in negotiation (claim 16666) and should not be counted as signed.

Concentration has moved from Bitcoin-era dependencies (Kraken, BC Hydro; claims 12183, 12182) to a hyperscaler-dominated book. On the $3.1bn ARR under contract reported in May 2026 (claim 9183), Microsoft's $1.9bn (claims 20245, 8996) is roughly 61%. NVIDIA at about $0.68bn annualised is roughly 22%. NVIDIA is simultaneously a customer and a conditional $2.1bn investor whose rights vest with deployment (claim 9179), which warrants scrutiny of arm's-length economics.

The arithmetic largely reconciles. $9.7bn over five years supports the $1.9bn Microsoft figure. Microsoft plus NVIDIA gives about $2.6bn, leaving about $0.5bn for Prince George and the smaller AI-native contracts, which is consistent with claim 16665. The July $2.8bn, at the roughly four-year weighted average term (claim 8108), implies about $0.7bn of incremental ARR. That would dilute Microsoft toward about half of contracted ARR, but no updated company ARR figure appears on record to confirm it. Prepayments rising from 20% on Microsoft to about 45% on recent deals (claims 20245, 8595) modestly improve credit quality on the smaller counterparties.

IREN's contracted revenue rests on two named, signed hyperscale agreements. Much of the surrounding partnership language is framework, certification, or supplier material that should not be read as customer demand. The anchor is the five-year, $9.7 billion Microsoft GPU services contract dated November 2, 2025 (claims 4157, 20307, 1266). It has moved past paper: Horizon 1 was delivered to and accepted by Microsoft in August 2026 (claims 35, 8673, 7454). The second is the $3.4 billion five-year NVIDIA AI Cloud contract signed in May 2026 (claims 9180, 11076, 9287). The 5GW NVIDIA strategic partnership attached to it is described only as a collaboration across a pipeline (claims 11077, 9289), and the $2.1 billion NVIDIA investment vests only as deployments occur (claim 9179). Both are conditional frameworks, not revenue.

Several counterparties are suppliers or vendors, not customers. These include Bitmain (claims 14392, 6374, 6152), Dell's $5.8 billion and $1.6 billion GPU purchase agreements (claims 4161, 1297), BC Hydro (claims 12182, 12198), Kraken (claim 12183) and WEKA (claim 12005). NVIDIA Preferred Partner and Exemplar Cloud status (claims 20300, 8543) are certifications. The Mirantis references describe Mirantis's own 1,500 customers, not IREN's (claims 8768, 8769).

Named smaller customers do exist. They are Poolside AI from February 2024 (claim 6285), and Together AI, Fluidstack and Fireworks AI on undisclosed multi-year terms (claim 20247). The July 2026 $2.8 billion tranche is attributed to "leading AI developers" (claim 360), and the roster still includes "a new leading AI developer" left unnamed (claim 8101). The Prince George contracts behind ~$0.4 billion ARR are also unidentified (claim 16665). The further >$0.5 billion in British Columbia is explicitly still in negotiation (claim 16666).

Concentration moved from no contracted customers in the bitcoin-mining era to near-total Microsoft dependence in late 2025. NVIDIA and the July cohort have since diluted it modestly. The arithmetic reconciles at the headline level. $9.7 billion over five years gives roughly the $1.9 billion annual figure IREN cites (claims 20245, 8996). The NVIDIA contract adds about $0.68 billion, and Prince George adds $0.4 billion. Together that is roughly $3.0 billion, close to the $3.1 billion ARR under contract reported in May (claim 9183). The residual is plausibly the smaller named contracts. The July $2.8 billion has no disclosed ARR translation; at the stated ~4-year weighted term (claim 8108) it would add perhaps $0.7 billion. That figure is inference, not disclosure.

IREN's genuinely signed, named customer book rests on two counterparties. The Microsoft agreement is a dated, executed GPU services contract worth $9.7 billion over five years at Childress (claims 20307, 4157, 1266), and it has moved past paper: Horizon 1 was delivered to and accepted by Microsoft (claims 8673, 35). The NVIDIA deal is a signed five-year, $3.4 billion AI Cloud contract (claims 11076, 9180). Everything wrapped around that NVIDIA deal is softer. The '5GW strategic partnership' is collaboration language with no committed volume (claims 11077, 9289), and the $2.1 billion NVIDIA investment vests only as GPU infrastructure is deployed (claim 9179). Together AI, Fluidstack and Fireworks AI are named as multi-year contract holders, but no values are attached (claim 20247). The July $2.8 billion tranche is described only as contracts 'with leading AI developers' (claim 360). The customer roster that accompanied it includes an unidentified 'new leading AI developer' (claim 8101), and attributions of that tranche to Figure AI or Hume AI come from secondary coverage (claim 8117). In British Columbia, roughly $0.4 billion of ARR at Prince George is under contract with an unnamed customer (claim 16665), while the further >$0.5 billion is explicitly still in negotiation (claim 16666). Several earlier relationships were never customer revenue at all. Bitmain, Dell, BC Hydro and Kraken are suppliers or service providers (claims 14392, 4161, 1297, 12198, 12183), and WEKA was a vendor partnership (claim 12005). Concentration has inverted. Before 2023, IREN was a self-mining operation with no customers. It is now an AI cloud host whose contracted revenue is dominated by Microsoft. The arithmetic largely reconciles. Microsoft's $9.7 billion over five years implies about $1.94 billion a year, consistent with the stated $1.9 billion ARR (claims 20245, 8996). NVIDIA's contract implies about $0.68 billion a year, and Prince George adds about $0.4 billion. Together these come to roughly $3.0 billion against the $3.1 billion of ARR under contract disclosed in May (claim 9183), leaving about $0.1 billion for the smaller named AI customers. At that point Microsoft accounted for roughly 61% of contracted ARR, and Microsoft and NVIDIA together about 83%. If the July $2.8 billion carries the roughly four-year weighted term cited (claim 8108), it would add about $0.7 billion of ARR and pull Microsoft's share toward half. However, that term is a portfolio average, and the record supplied here contains no updated aggregate ARR figure to confirm it. Prepayment terms have also tightened in IREN's favour, from 20% on Microsoft to roughly 45% of GPU capex on recent deals (claims 20245, 8595).

IREN's disclosure record falls into two parts. Through 2024 the named counterparties are suppliers and service venues, not revenue customers. Bitmain supplies hardware (claims 14392, 6374, 6152), BC Hydro supplies power (12182, 12198), Kraken is a liquidation venue (12183) and WEKA is a software vendor (12005). Poolside AI (6285) is the first named cloud customer, but only a service commencement was disclosed, with no contract value. The material signed customer agreements start in November 2025. The Microsoft GPU services contract is fully identified and documented: $9.7bn, five years, Childress, a 20% prepayment, and a filing-level description of the counterparty and term (20307, 4157, 1266, 20245). Its first tranche has been delivered and accepted (8673, 35). The NVIDIA $3.4bn five-year contract is also described as signed (9180, 11076). The surrounding '5GW strategic partnership' is framework language with no stated commitment (11077, 9289), and NVIDIA's $2.1bn investment vests only as deployments occur (9179). These should not be counted as contracted revenue. The Dell agreements of $5.8bn and $1.6bn (4161, 8992) are purchase obligations that IREN guarantees (4164), not revenue. The weakest attribution concerns the July 2026 $2.8bn tranche. IREN's own wording names only 'leading AI developers' (360). The attribution to Figure AI, Fluidstack, Fireworks AI and Hume AI comes from a secondary source (8117), and the customer roster includes 'a new leading AI developer' who is not named (8101). Together AI, Fluidstack and Fireworks were named in November 2025 without values (20247). In British Columbia, about $0.4bn of ARR is under contract (16665), while a further >$0.5bn was still in negotiation (16666) and should be treated as pipeline. Concentration is high but easing. Microsoft's expected $1.9bn annual revenue (8996) is roughly 61% of the $3.1bn ARR under contract reported in May (9183). The figures reconcile. $9.7bn over five years gives $1.94bn. NVIDIA's $3.4bn over five years adds about $0.68bn. Prince George adds about $0.4bn. Together that is about $3.0bn, with a small residual for smaller contracts. The July $2.8bn at the roughly four-year weighted term (8108) implies about $0.7bn of additional ARR, which would bring Microsoft's share to around half. However, no updated consolidated ARR figure appears in the claims reviewed, which were truncated at claim 8563. Prepayment terms have also shifted, from 20% on Microsoft (20245) to about 45% on recent deals (8595). That improves funding quality, but it cannot be checked contract by contract while the July counterparties remain unnamed.

IREN's signed-customer record is narrower and more concentrated than its partner language suggests. The firmest deal is Microsoft. The company says it 'signed' a GPU cloud services contract (claim 20307), and the agreement is dated November 2, 2025 (claim 1266). Horizon 1 has since been delivered to and accepted by Microsoft (claims 8673, 35). NVIDIA is the second firm counterparty, with a $3.4bn five-year contract described as 'signed' (claim 9180) and backed by a $1.6bn Dell purchase agreement to service it (claims 8992, 1297). Delivery, however, does not ramp until early 2027 (claim 9287). Together AI, Fluidstack and Fireworks AI are named as holding multi-year contracts (claim 20247), and Poolside AI was an early named GPU cloud customer (claim 6285). Bitmain, Dell, BC Hydro, Kraken and WEKA are suppliers or vendors, not revenue counterparties (claims 14392, 4161, 12198, 12183, 12005).

Several items are framework or status language rather than customer contracts. The 5GW NVIDIA strategic partnership is a 'collaboration to support deployment' with no stated economics (claims 11077, 9289). 'NVIDIA Preferred Partner' is a program designation (claim 20300). Mirantis's 1,500 enterprise customers belong to Mirantis, not IREN (claim 8768). In British Columbia, only about $0.4bn of ARR is under contract (claim 16665). The more than $0.5bn beyond that was still 'negotiations' with unnamed counterparties (claim 16666). The July $2.8bn tranche was attributed by the company to unnamed 'leading AI developers' (claim 360). The names attached to it, including Figure AI and Hume AI, come from a secondary write-up (claim 8117), and the customer roster itself includes 'a new leading AI developer' left unidentified (claim 8101). NVIDIA is also both a customer and a $2.1bn investor whose rights vest as its GPUs are deployed on IREN campuses (claim 9179), so that relationship is not arm's-length.

The revenue arithmetic broadly reconciles. $9.7bn over five years is about $1.94bn a year, consistent with the $1.9bn ARR guidance (claims 20245, 8996). NVIDIA's $3.4bn over five years adds about $0.68bn, and Prince George adds about $0.4bn. Together that is roughly $3.0bn, close to the stated $3.1bn of ARR under contract (claim 9183). Microsoft therefore represented about 61% of contracted ARR in May 2026, and Microsoft plus NVIDIA over 80%. The July $2.8bn, at a roughly four-year weighted average term (claim 8108), adds about $0.7bn of ARR, which dilutes Microsoft toward half but leaves two counterparties dominant. Customer prepayments rose from 20% on the Microsoft deal (claim 20245) to about 45% on recent agreements (claim 8595). The claim record supplied ends mid-entry at claim 8563, so later disclosures were not assessed.

IREN's record divides into a small set of signed, named agreements and a larger layer of partnership and aggregate-contract language. The signed core is clear. Microsoft's five-year, $9.7bn GPU services contract is described as signed (20307), is tied to a dated November 2, 2025 agreement (1266), and has been partly performed, with Horizon 1 delivered to and accepted by Microsoft (8673, 35). The $3.4bn five-year NVIDIA AI Cloud contract is also described as signed (9180). On the supply side, Dell purchase agreements of about $5.8bn (4161, guaranteed by the parent under 4164) and $1.6bn (1297, 8992) are definitive, as were the earlier Bitmain hardware contracts (14392, 6152). Other items carry softer language. WEKA was a 'partnering' arrangement (12005). The NVIDIA 5GW 'strategic partnership' (11077, 8642) and 'Preferred Partner' status (20300) are frameworks with no stated value. British Columbia capacity backing more than $0.5bn of ARR was still in 'contract negotiations' (16666). Several counterparties are unidentified. The ~$0.4bn Prince George ARR is 'under contract' with no named customer (16665). The July $2.8bn is attributed to 'leading AI developers' (360) and to 'a new leading AI developer' (8101). The only named roster for that tranche comes from a secondary source using 'companies like' (8117), so the full list is not confirmed. Together AI, Fluidstack and Fireworks AI are named as customers (20247), but no individual contract values are given for them. Concentration has moved from a single AI customer, Poolside (6285), to heavy reliance on two counterparties. Of roughly $15.9bn in headline contract value, Microsoft accounts for about 61% and NVIDIA about 21%. NVIDIA is also a $2.1bn investor whose rights vest with GPU deployment (9179), which makes it customer, supplier ecosystem and shareholder at once. The arithmetic largely reconciles. $9.7bn over five years is about $1.94bn a year, consistent with the stated $1.9bn ARR (20245, 8996). NVIDIA's $3.4bn over five years adds about $0.68bn. With Prince George's ~$0.4bn, the total is about $2.98bn, close to the $3.1bn ARR under contract reported in May (9183). The July $2.8bn has no term disclosed, so its ARR contribution cannot be confirmed; at the ~4-year weighted average term (8108) it would imply roughly $0.7bn. Hardware commitments equal about 60% of Microsoft contract value and 47% of NVIDIA's. The claim record supplied ends mid-entry at 8641, so later disclosures are not assessed.

IREN's signed-contract record is stronger than most AI-infrastructure pivots, but the language thins out quickly beyond its two anchor counterparties. The legacy mining book rested on binding contracts with Bitmain (14392, 6374, 6152) and BC Hydro supply contracts with defined terms (12198). Kraken was a liquidation venue, not a customer (12183). The 2023 WEKA arrangement was 'partnering' language with no economics (12005). Poolside AI was the first named GPU cloud customer, though disclosed only as service 'commencement' (6285). The Microsoft agreement is unambiguously signed: it is a dated five-year, $9.7bn contract executed through IE US Hardware 3 (20307, 4157, 1266), and Horizon 1 has been delivered and accepted (35, 8673), which moves it past paper. The $3.4bn NVIDIA contract is also described as signed (9180, 9287). The accompanying '5GW strategic partnership' is a framework with no stated revenue commitment (11077, 9289, 8642), and 'NVIDIA Preferred Partner' is a program designation, not a customer contract (20300). The Dell agreements are supplier obligations backed by a parent guarantee and should not be read as demand (4161, 4164, 1297). Together AI, Fluidstack and Fireworks AI are named as multi-year contracts but carry no values (20247). The July $2.8bn tranche is attributed to 'leading AI developers' (360), including 'a new leading AI developer' that is still unidentified (8101). Coverage describes the counterparties only as 'companies like' Figure AI and Hume AI (8117). The Prince George ~$0.4bn ARR is unattributed (16665), and the >$0.5bn British Columbia figure is explicitly still in negotiation (16666). Concentration has eased from essentially a single Microsoft contract in November 2025 to roughly 61% of the ~$15.9bn disclosed contract value by August 2026. Still, Microsoft and NVIDIA together account for about 82%, and NVIDIA is simultaneously customer, architecture partner and a $2.1bn investor whose rights vest only as its GPUs are deployed (9179). That circularity deserves scrutiny. The arithmetic largely reconciles. $9.7bn over five years is about $1.94bn, consistent with the $1.9bn ARR cited (20245, 8996). NVIDIA's $3.4bn over five years implies about $0.68bn. Adding Prince George's $0.4bn gives roughly $3.0bn, leaving a modest residual against the $3.1bn ARR under contract stated in May (9183). The July $2.8bn at a ~4-year weighted term (8108) would add roughly $0.7bn, but no updated ARR figure appears on record, so that increment cannot yet be verified.

IREN's record divides cleanly into signed contracts and softer relationship language, and the split matters more than the headline totals. The firmest customer evidence is the Microsoft GPU services agreement dated November 2, 2025, which the company's own filing names as a contract with Microsoft Corporation (claims 20307, 4157, 1266). It carries a 20% prepayment (claim 20245), and Horizon 1 has since been delivered to and accepted by Microsoft (claims 35, 8673). The $3.4 billion NVIDIA contract is also described as signed (claim 9180), and Together AI, Fluidstack and Fireworks AI are named on multi-year contracts (claim 20247). By contrast, the 5GW NVIDIA 'strategic partnership' is a collaboration framework with no stated economics (claims 11077, 9289). The $2.1 billion NVIDIA investment vests only as deployments occur (claim 9179). 'NVIDIA Preferred Partner' is a program designation, not a contract (claim 20300). The British Columbia figure of more than $0.5 billion of ARR is explicitly still in 'contract negotiations' (claim 16666). Several relationships on the list are supply or vendor ties, not revenue: Bitmain (claims 14392, 6374, 6152), Dell (claims 4161, 1297), BC Hydro (claim 12182), Kraken (claim 12183) and WEKA (claim 12005). Identification gaps persist. The roughly $0.4 billion of Prince George ARR has no named counterparty (claim 16665), and the customer list includes 'a new leading AI developer' (claim 8101). The company itself attributes the July $2.8 billion only to 'leading AI developers' (claim 360). The named roster for that tranche comes from secondary press using 'companies like' (claim 8117). Concentration has moved from a Bitcoin model with no customers in the ordinary sense to heavy reliance on one hyperscaler. Microsoft's expected $1.9 billion of annual revenue (claims 20245, 8996) was about 61% of the $3.1 billion of ARR under contract reported in May 2026 (claim 9183). The May figure roughly reconciles: Microsoft's $1.9 billion, about $0.68 billion a year from NVIDIA's $3.4 billion over five years, and Prince George's $0.4 billion total about $2.98 billion, leaving roughly $0.1 billion for smaller named customers. This is our reconstruction, not a company bridge. No updated ARR figure is on record after the July $2.8 billion. If that tranche runs on the roughly four-year weighted average term (claim 8108), it would add about $0.7 billion a year and bring Microsoft's share closer to half. That conclusion depends on undisclosed tenor and on the negotiated British Columbia contracts actually being signed.

IREN's record divides cleanly into a small set of executed, named contracts and a larger layer of partnership and pipeline language. The anchor is the Microsoft GPU services agreement dated November 2, 2025, a five-year, $9.7 billion contract through subsidiary IE US Hardware 3, documented in filing language (4157, 1266, 20307). Its delivery milestone is also on record: Horizon 1 was delivered to and accepted by Microsoft (35, 8673, 8631). The second signed revenue contract is the five-year, $3.4 billion AI Cloud contract with NVIDIA (9180, 11076, 9287). The Dell agreements are signed too: $5.8 billion, carrying a parent guarantee (4161, 20308, 4164), and $1.6 billion (1297, 8992). So are the Bitmain contracts (14392, 6374, 6152). These are supply-side purchases, however, and not customers. Kraken and BC Hydro are a liquidation venue and a utility (12191, 12198), and WEKA is a vendor partnership (12005).

The softer language clusters around NVIDIA. The "5GW strategic partnership" is described as a collaboration on aligned architecture and carries no stated value (11077, 9289, 362, 8642). "Preferred Partner" is a program tier (20300). The $2.1 billion NVIDIA investment vests only as NVIDIA infrastructure is deployed (9179). That leaves NVIDIA as customer, investor, and upstream supplier at once, a circularity diligence should weigh. Unidentified counterparties persist alongside the named ones. These include "a new leading AI developer" (8101), Prince George's ~$0.4 billion ARR "under contract" with no named buyer (16665), and generic "hyperscalers" phrasing (20306). The >$0.5 billion British Columbia figure is explicitly still in negotiation (16666). Mirantis's 1,500 customers are Mirantis's, not IREN's (8768).

Concentration has moved from no customers at all (self-mined Bitcoin) to a single dominant counterparty. Microsoft is roughly 61% of the ~$15.9 billion in announced contract value, and Microsoft plus NVIDIA is about 82%. The July $2.8 billion tranche (360) diversifies this somewhat, but no per-customer values were disclosed. Its attribution is also muddy: the press account in 8117 names Fluidstack and Fireworks AI, both already announced in November 2025 (20247).

The revenue arithmetic largely holds. $9.7 billion over five years is about $1.94 billion a year, matching the $1.9 billion ARR guidance (20245, 8996). NVIDIA's $3.4 billion over five years implies about $0.68 billion. Together with Prince George's ~$0.4 billion, that comes to roughly $3.0 billion, consistent with the $3.1 billion ARR under contract (9183). At a ~4-year weighted term (8108), the July contracts would add about $0.7 billion. No updated ARR figure confirming this appears in the record supplied, which was truncated at claim 8642.

IREN's record contains two signed, named, dollar-quantified customer contracts. The first is the November 2025 Microsoft GPU services agreement at $9.7 billion over five years (claims 20307, 4157, 1266). It is backed by a 20% prepayment (20245), a matching $5.8 billion Dell hardware commitment that IREN guarantees (4161, 4164), and the delivery and acceptance of Horizon 1 (35, 8673). That acceptance is the strongest performance evidence in the file. The second is the five-year, $3.4 billion NVIDIA AI cloud contract, which IREN describes as signed (9180, 11076) and which is paired with a $1.6 billion Dell purchase of air-cooled Blackwell systems (8992, 1297). The NVIDIA contract should be read separately from the '5GW strategic partnership' (11077, 9289). That partnership is described only as a collaboration on infrastructure alignment, with no stated economics, and so is framework language. It also sits alongside a $2.1 billion NVIDIA investment whose rights vest as NVIDIA infrastructure is deployed (9179). NVIDIA is therefore customer, supplier ecosystem and prospective investor at once, which diligence should treat as related-party-adjacent. Smaller named cloud customers are Poolside (6285), Together AI, Fluidstack and Fireworks AI (20247), and later Perplexity, Figure AI, Fal AI and Hume AI (8101). None of these carries individual contract values. The July 2026 $2.8 billion tranche is attributed only to 'leading AI developers' (360, 8594) and to 'companies like' four names (8117). The same roster includes an unidentified 'new leading AI developer' (8101). Several counterparties labelled partners are vendors or service providers, not customers: Bitmain (14392, 6152), BC Hydro (12198), Kraken (12191), WEKA (12005) and Dell. In British Columbia, about $0.4 billion of ARR is under contract (16665), while more than $0.5 billion is still in negotiation (16666) and should not be counted. Customer concentration has shifted from Bitcoin pool revenue to hyperscaler revenue. Microsoft accounts for roughly 61% of the roughly $15.9 billion in announced contract value, and Microsoft plus NVIDIA for about 82%. The headline arithmetic broadly reconciles. $9.7 billion over five years implies about $1.94 billion a year, consistent with the $1.9 billion guided (8996). Adding about $0.68 billion a year from NVIDIA and $0.4 billion from Prince George gives about $3.0 billion, close to the $3.1 billion ARR cited in May (9183). However, no post-July ARR figure is on record. The $2.8 billion tranche therefore cannot be reconciled until its term mix, weighted at about four years (8108), and its counterparty split are disclosed.

IREN's signed-customer record is shorter than its partnership language suggests. Most pre-2024 counterparties are suppliers or service venues, not customers. Bitmain's binding hardware contracts (14392, 6374, 6152), Kraken as a liquidation venue (12191, 12183) and BC Hydro power supply (12198, 12182) are cost-side arrangements. The WEKA 'partnership' (12005) is a software-vendor relationship. The first named AI cloud customer is Poolside AI (6285), and no contract value was disclosed. The anchor deal is Microsoft, a signed $9.7bn, five-year GPU services agreement executed by IE US Hardware 3 (20307, 4157, 1266). Microsoft has since accepted Horizon 1 (35, 8673, 8631), which is the only delivery milestone on record. Its mirror is a $5.8bn Dell purchase commitment backed by a parent guarantee (4161, 4164). That is capex, not revenue. NVIDIA's $3.4bn five-year contract is described as signed (9180, 11103), with a $1.6bn Dell equipment purchase behind it (1297, 8992). The accompanying '5GW strategic partnership' is framework language, a 'collaboration to support deployment' with no quantified obligations (11077, 9289, 362, 8642). NVIDIA also sits on three sides of the relationship: it is a customer, a vendor ecosystem and an investor whose $2.1bn investment rights vest with deployment (9179). Diligence should treat that as a related-party-like circularity. Several counterparties are unidentified. The ~$0.4bn Prince George ARR is 'under contract' but unnamed, and a further >$0.5bn remains in 'negotiations' (16665, 16666). The July $2.8bn tranche is attributed only to 'leading AI developers' (360). The roster adds 'a new leading AI developer' that is never named (8101), and later rosters drop it (363). The mapping of Figure AI, Fluidstack, Fireworks AI and Hume AI to that $2.8bn comes from secondary coverage, not company disclosure (8117). The Mirantis 1,500-customer figure (8768) belongs to Mirantis, not IREN. The arithmetic largely reconciles. $9.7bn over five years is about $1.94bn a year, consistent with the $1.9bn guided (20245, 8996). NVIDIA adds about $0.68bn a year. Together with Prince George's ~$0.4bn, that totals roughly $3.0bn, close to the $3.1bn ARR under contract stated in May (9183). This holds only if the BC negotiations were excluded. On that basis Microsoft was about 61% of contracted ARR, and Microsoft plus NVIDIA about 83%. The July tranche, over the ~4-year weighted term (8108), implies roughly $0.7bn a year more and dilutes Microsoft toward half. No updated ARR figure appears in the record, which is truncated at claim 8642.

IREN's record divides into a small set of signed, named revenue contracts, a larger set of supplier and vendor agreements, and framework language that should not be counted as revenue. The anchor is the Microsoft GPU services contract signed November 2, 2025 (claims 20307, 4157, 1266). It is a five-year, $9.7bn contract with a 20% prepayment and an expected $1.9bn of annual revenue (20245, 8996). It is also the only customer deal with evidence of performance: Microsoft accepted delivery of Horizon 1 in August 2026 (35, 8673). The NVIDIA five-year, $3.4bn AI Cloud contract is described as signed (9180, 11076). The adjacent '5GW strategic partnership' (11077, 9289, 8642) is a collaboration framework with no stated revenue attached. NVIDIA's $2.1bn investment rights vest only as its GPUs are deployed (9179), which makes NVIDIA simultaneously a customer, a prospective equity holder and the ultimate chip supplier. That circularity deserves scrutiny. The Bitmain (14392, 6374, 6152) and Dell (4161, 1297) agreements are purchase commitments, not customers, and $7.4bn of Dell obligations sit against the two anchor contracts, with a parent guarantee (4164). BC Hydro, Kraken, WEKA and Mirantis are vendors or partners. Mirantis's '1,500 enterprise customers' belong to Mirantis, not IREN (8768). Named smaller customers have accumulated over time: Poolside in 2024 (6285), Together AI, Fluidstack and Fireworks AI in November 2025 (20247), and Perplexity, Figure AI, Fal AI and Hume AI by July 2026 (8101). The July $2.8bn tranche (360, 8641) still includes 'a new leading AI developer' who is not named. The 'broad range of customers' language (20306) and the Prince George figures are not tied to counterparties. The ~$0.4bn under contract (16665) and the >$0.5bn still 'in negotiation' (16666) should be excluded from contracted totals. On the arithmetic, Microsoft's $9.7bn over five years gives about $1.94bn a year, and NVIDIA's $3.4bn gives about $0.68bn. Together with Prince George that is roughly $3.0bn, close to the $3.1bn ARR under contract reported in May 2026 (9183), so the figures broadly reconcile. At that point Microsoft was about 63% of ARR. The July $2.8bn, at a roughly four-year weighted term (8108), implies about $0.7bn more ARR. That should lift the total toward $3.8bn and dilute Microsoft's share to around half, but no updated ARR figure appears on record to confirm it. Note that the claim record supplied was truncated at claim 7454.

IREN's record divides into two eras. Before late 2025, IREN was a self-miner with no customers in the commercial sense. Its named counterparties were suppliers and service providers: Bitmain for hardware (14392, 6374, 6152), BC Hydro for power (12182, 12198), Kraken for liquidation (12183) and WEKA for software (12005). The one early AI customer, Poolside AI, is described only as the "commencement of GPU cloud service" (6285), with no disclosed value. The signed-customer era begins with the $9.7bn, five-year Microsoft GPU services contract dated November 2, 2025 (20307, 4157, 1266). It is the best-evidenced agreement on file, with a 20% prepayment (20245) and delivery and acceptance of Horizon 1 confirmed (35, 8673). The $3.4bn five-year NVIDIA AI Cloud contract is described as signed (9180, 11103) and is backed by a $1.6bn Dell hardware purchase (8992, 1297). The Dell agreements (4161, 20308, 8992) are supply obligations, not revenue. The 5GW NVIDIA "strategic partnership" (11077, 9289, 8642) is framework language, and the $2.1bn NVIDIA investment only vests as NVIDIA infrastructure is deployed (9179). That makes NVIDIA simultaneously customer, investor and ecosystem gatekeeper, a circularity diligence should weigh. Named smaller customers include Together AI, Fluidstack and Fireworks AI (20247), later joined by Perplexity, Figure AI, Fal AI and Hume AI (8101), but no per-customer values are disclosed. Unidentified counterparties persist: "hyperscalers" (20306), "a new leading AI developer" (8101), and British Columbia contracts still in negotiation supporting ">$0.5bn ARR" (16666). That last figure should be excluded from contracted totals. The secondary attribution of the July $2.8bn to "companies like" Figure AI and others (8117) is not a disclosure of who signed what (360). Concentration is heavy. Of roughly $15.9bn in announced contract value, Microsoft accounts for about 61%, and Microsoft plus NVIDIA about 82%. The revenue arithmetic broadly reconciles. The Microsoft contract's $1.9bn ARR (20245, 8996) matches $9.7bn over five years. NVIDIA's $3.4bn implies about $0.68bn a year. Together with roughly $0.4bn contracted at Prince George (16665), that is about $3.0bn, close to the $3.1bn ARR under contract stated in May (9183). The July $2.8bn, at the roughly four-year weighted term (8108), would add about $0.7bn. No updated ARR figure on record confirms that step-up.

IREN's signed-customer disclosure rests on two contracts, and the rest of its partnership record sits in softer categories. The Microsoft agreement is fully identified: a dated contract between subsidiary IE US Hardware 3 and Microsoft Corporation for tranche-based GPU services at Childress (claims 4157, 1266), described as a five-year, $9.7bn contract with 20% prepayment (20245). It moved from signature to performance when Horizon 1 was delivered to and accepted by Microsoft (35, 8673). The $3.4bn five-year NVIDIA contract is also named and signed (11076, 9180), and it is backed by a $1.6bn Dell purchase agreement tied to it explicitly (8992). Dell and Bitmain are suppliers rather than customers (4161, 1297, 14392, 6152), and so are BC Hydro and Kraken (12182, 12183).

Several claims use partnership or negotiation language and should not be read as revenue. These include the 5GW NVIDIA strategic partnership, which is framed as a collaboration on architecture across the pipeline (11077, 9289, 362), the NVIDIA Preferred Partner status (20300), and the WEKA partnering (12005). The Poolside service commencement has no disclosed value (6285). The Mirantis claims describe Mirantis's own customers and not IREN's (8768, 8769). The British Columbia disclosures separate about $0.4bn ARR under contract (16665) from more than $0.5bn still in negotiation (16666), and neither names a counterparty. The $2.8bn July tranche is attributed only to 'leading AI developers' (360, 8641). Its counterparties appear only through roster lists, including one 'new leading AI developer' who is not identified (8101), and through a secondary source (8117). No per-customer values are given for Together AI, Fluidstack, Fireworks, Figure, Fal or Hume (20247, 363).

Concentration is high but falling. Microsoft's expected $1.9bn ARR (20245, 8996) is about 61% of the $3.1bn ARR under contract reported in May (9183). That total reconciles closely: Microsoft's $1.9bn, plus NVIDIA's $3.4bn over five years (about $0.68bn a year), plus Prince George's $0.4bn comes to roughly $3.0bn. Applying the roughly four-year weighted term (8108) to the July $2.8bn implies about $0.7bn of additional ARR, or around $3.8bn in total. On that basis Microsoft falls to roughly half and the top two customers to about two-thirds. This updated total is our inference, because the record we received is truncated at claim 8594. One caution: NVIDIA is simultaneously a customer, the GPU architecture partner, and a $2.1bn investor whose rights vest as NVIDIA infrastructure is deployed (9179). That counterparty should be analysed as one linked relationship, not three independent ones.

IREN's customer record separates fairly cleanly into signed commercial contracts, supplier agreements, and framework language, but the headlines tend to blur the three. The strongest item is the Microsoft GPU services contract. It was disclosed as signed (claim 20307), described at the subsidiary level with tranche mechanics at Childress (claims 4157, 1266), and supported by a 20% prepayment and $1.9bn expected ARR (claim 20245). Delivery and acceptance of Horizon 1 (claims 8673, 35) move it from backlog to performing contract. The NVIDIA $3.4bn five-year contract is also described as signed (claim 9180) and is backed by a matching Dell hardware order (claims 8992, 1297). The accompanying 5GW 'strategic partnership' (claims 11077, 9289, 8642), however, is collaboration language with no stated revenue. The $2.1bn NVIDIA investment only vests as NVIDIA infrastructure is deployed (claim 9179), which makes NVIDIA customer, supplier-adjacent and investor at once. Several claims describe vendors rather than customers: Bitmain (claims 14392, 6152), Dell (claims 20308, 4161), Kraken (claim 12183), BC Hydro (claim 12182) and WEKA (claim 12005). They should not be read as demand. The named AI roster has grown from Poolside (claim 6285), which is absent from later lists, to Together AI, Fluidstack and Fireworks AI (claim 20247), and then to the nine-name roster in claims 8101 and 363. Contract values for the smaller names are not broken out. The July $2.8bn tranche is attributed only to 'leading AI developers' (claim 360), and it includes an unnamed 'new leading AI developer' (claim 8101). The per-counterparty allocation in claim 8117 comes from secondary coverage, not from the company. In British Columbia, $0.4bn ARR is under contract (claim 16665), while more than $0.5bn is still in negotiation (claim 16666), with no counterparties named. The arithmetic broadly reconciles. $9.7bn over five years gives about $1.94bn a year, matching the $1.9bn guidance (claim 8996). $3.4bn over five years adds roughly $0.68bn. Against the $3.1bn ARR under contract (claim 9183), the two leave about $0.5bn, close to the Prince George figure. That implies Microsoft and NVIDIA together made up roughly 80-85% of contracted ARR in May 2026. The July $2.8bn, spread over the roughly four-year weighted term (claim 8108), would dilute that concentration somewhat, but no updated ARR figure appears among the claims shown. The record also truncates at claim 8640, so later reconciliation cannot be confirmed.

IREN's disclosed record separates cleanly into signed revenue contracts, supplier and utility agreements, and framework-style partnership language, and the three should not be read together. The signed, named customer contracts are Microsoft, a five-year, $9.7bn GPU services agreement dated November 2, 2025 (claims 4157, 20307, 1266), and NVIDIA, a five-year, $3.4bn managed AI cloud contract (claims 11103, 9180, 8995). Microsoft's acceptance of the 50MW Horizon 1 deployment (claims 35, 8673, 7454) is the strongest execution evidence on file. Named but unquantified are Together AI, Fluidstack and Fireworks AI (claim 20247), and later Perplexity, Figure AI, Fal AI and Hume AI (claims 8101, 363). The July 2026 $2.8bn tranche is attributed to "leading AI developers" (claim 360) and includes an unidentified "new leading AI developer" (claim 8101). Secondary coverage names counterparties only as "companies like" (claim 8117), so no per-customer split exists. Poolside AI (claim 6285) was an early 2024 customer and is absent from every 2026 roster. Several items are not customer revenue at all. Bitmain and Dell are equipment purchases (claims 14392, 6152, 20308, 1297), BC Hydro is power supply (claim 12182), and Kraken is a liquidation venue (claim 12183). The WEKA "partnership" (claim 12005), the NVIDIA "Preferred Partner" designation (claim 20300) and the 5GW NVIDIA "strategic partnership" (claims 9289, 362) are framework language with no disclosed economics. Mirantis's 1,500 enterprise customers (claim 8768) belong to Mirantis, not IREN. NVIDIA's $2.1bn investment, which vests as its GPUs are deployed (claim 9179), makes it simultaneously customer, supplier-aligned partner and investor, so that relationship is not arm's-length diversification. On arithmetic, $9.7bn over five years is about $1.94bn a year, consistent with the stated $1.9bn (claims 20245, 8996). NVIDIA's $3.4bn over five years adds about $0.68bn, and Prince George adds roughly $0.4bn (claim 16665). Together these total about $3.0bn, which reconciles with the $3.1bn ARR under contract (claim 9183) once the smaller named contracts are included. The additional >$0.5bn for British Columbia was described only as negotiations (claim 16666) and should not be counted as signed. Concentration is high. Microsoft represented roughly 61% of contracted ARR in May, and Microsoft plus NVIDIA roughly 83%. If the July $2.8bn runs near the disclosed four-year weighted term (claim 8108), it adds about $0.7bn of ARR and lowers Microsoft's share to roughly half. Even so, the book remains dominated by two counterparties.

IREN's disclosed book rests on two signed hyperscale-scale contracts. The Microsoft GPU services agreement, dated November 2, 2025, is documented at the contract level with a named counterparty, a named subsidiary obligor, a site, and a term (claims 4157, 20307, 1266). Its execution is further evidenced by Microsoft's acceptance of Horizon 1 (claims 35, 8673, 8631). The $3.4 billion NVIDIA AI Cloud contract is described as signed, with a stated term and deployment footprint (claims 9180, 9287, 11076). Everything else on record needs a closer look. The 5GW NVIDIA 'strategic partnership' is collaboration language with no dollar value or binding capacity commitment (claims 11077, 9289, 362). NVIDIA's $2.1 billion investment vests only as its GPUs are deployed (claim 9179), which makes NVIDIA simultaneously customer, supplier ecosystem and prospective shareholder. The 'Preferred Partner' designation (claim 20300) and the WEKA tie-up (claim 12005) are program and vendor relationships, not revenue. Dell (claims 4161, 1297), Bitmain (claims 14392, 6374), BC Hydro (claims 12182, 12198) and Kraken (claims 12183, 12191) are suppliers or service venues and should not be read as customers.

Naming is uneven below the top two. Together AI, Fluidstack and Fireworks AI are named as multi-year contracts (claim 20247). The July $2.8 billion tranche is attributed by the company only to 'leading AI developers' (claims 360, 8641). The specific names attached to it come from a secondary characterization (claim 8117), and the roster itself includes an unidentified 'new leading AI developer' (claim 8101). Prince George's roughly $0.4 billion ARR has no named counterparty (claim 16665), and the further >$0.5 billion is explicitly still in negotiation (claim 16666).

Concentration has moved from a single early GPU customer, Poolside (claim 6285), to heavy Microsoft dependence, and has since diluted somewhat. The arithmetic largely holds. $9.7 billion over five years supports the stated $1.9 billion average annual revenue (claims 20245, 8996). NVIDIA's $3.4 billion over five years implies about $0.68 billion. Together those account for about $2.58 billion of the $3.1 billion ARR under contract (claim 9183), leaving roughly $0.5 billion for the smaller AI Cloud customers and Prince George. That plausibly reconciles, but Microsoft is still around 60 percent of contracted ARR. The July $2.8 billion, at a roughly four-year weighted term (claim 8108), would add on the order of $0.7 billion. However, no updated aggregate ARR figure appears in the record supplied, which ends mid-entry at claim 8594. Until the NVIDIA ramp begins in early 2027 (claim 9287), realized revenue is effectively a Microsoft story.

IREN's disclosed agreements fall into three tiers: signed customer contracts, signed supply agreements that are not revenue, and partnership language that commits little. The Microsoft agreement is the anchor. It is a five-year, $9.7bn GPU services contract dated November 2, 2025 (claims 4157, 1266), with a 20% prepayment and about $1.9bn of expected ARR (claim 20245). Delivery has begun: Microsoft accepted Horizon 1, the first of four 50MW tranches at Childress (claims 8673, 8632). NVIDIA's $3.4bn five-year AI Cloud contract is described as signed (claims 9180, 9287). The 5GW 'strategic partnership' announced alongside it is framed only as a 'collaboration to support deployment' (claim 9289) and should be treated as framework language, not backlog. Dell's $5.8bn and $1.6bn purchase agreements (claims 4161, 1297) are cost commitments for GPUs, and the first carries a parent guarantee (claim 4164). The same applies to the earlier Bitmain, BC Hydro, Kraken and WEKA relationships (claims 14392, 12182, 12191, 12005). Mirantis's 1,500 enterprise customers (claim 8768) are Mirantis's customers, not IREN's.

Naming is uneven. Together AI, Fluidstack and Fireworks AI are named but carry no disclosed values (claim 20247). The July $2.8bn tranche is attributed only to 'leading AI developers' (claim 360). The allocation to Figure AI, Fluidstack, Fireworks AI and Hume AI comes from secondary coverage (claim 8117), not a company breakdown. The customer roster also includes 'a new leading AI developer' who is not identified (claim 8101). At Prince George, about $0.4bn of ARR is under contract, but a further >$0.5bn sits in 'remaining contract negotiations' (claims 16665, 16666). That second figure is pipeline.

Concentration has moved from zero hosted customers to heavy dependence on Microsoft, and is only now diluting. The May 2026 figure of $3.1bn ARR under contract (claim 9183) reconciles reasonably well. Microsoft contributes about $1.9bn, NVIDIA about $0.68bn ($3.4bn over five years), and Prince George about $0.4bn, which totals roughly $3.0bn and leaves about $0.1bn for the smaller named customers. On that basis Microsoft was about 60% of contracted ARR. Taking the ~4-year weighted average term at face value (claim 8108), the July $2.8bn would add roughly $0.7bn and lower that share toward half. No updated ARR total appears on record to confirm this. Readers should also note a circularity: NVIDIA is both a $3.4bn customer and a prospective $2.1bn investor whose rights vest with GPU deployment (claim 9179).

IREN's signed-customer record is shorter than its partnership vocabulary suggests. Before late 2025 the named counterparties were suppliers and service providers, not revenue customers: Bitmain for miners (14392, 6374, 6152), BC Hydro for power (12198), Kraken for liquidating mined Bitcoin (12191), and WEKA as a software partner (12005). The one early AI customer, Poolside AI, was disclosed only as the start of a GPU cloud service with no value attached (6285). The first large, named and dated customer agreement is the five-year, $9.7bn Microsoft contract of November 2, 2025 (4157, 1266). It has since moved from announcement to performance, with Horizon 1 delivered to and accepted by Microsoft (8673, 35) and three more 50MW tranches scheduled (8632). The $3.4bn NVIDIA AI Cloud contract is described as signed (9180, 11076). The accompanying 5GW 'strategic partnership' is collaboration language with no stated value (11077, 9289, 8642). It should be read alongside NVIDIA's $2.1bn investment, which vests only as NVIDIA infrastructure is deployed (9179), rather than as additional revenue. Identification weakens after that. The ~$0.4bn ARR contracted for Prince George names no counterparty (16665), and the further >$0.5bn in British Columbia is explicitly still in negotiation (16666). The July $2.8bn tranche is attributed to 'leading AI developers' (360, 8641). A secondary source names Figure AI, Fluidstack, Fireworks AI and Hume AI only as 'companies like' (8117), and the customer roster still includes an unnamed 'new leading AI developer' (8101). Perplexity, Together AI and Fal AI appear on that roster with no disclosed contract terms. Concentration has narrowed and then broadened. Microsoft was essentially the whole contracted book in late 2025. On announced totals of roughly $15.9bn it still represents about 61%, with NVIDIA at about 21% and the July cohort at about 18%. The arithmetic largely reconciles. $9.7bn over five years is about $1.94bn a year, which matches the $1.9bn average annual figure (8996). NVIDIA's $3.4bn over five years adds about $0.68bn, and Prince George adds about $0.4bn. Together these come to roughly $3.0bn against the $3.1bn ARR under contract (9183), which leaves a small residual that plausibly reflects legacy AI cloud customers. About 13% of that ARR, however, rests on an unidentified Prince George counterparty. The July $2.8bn has no disclosed term, and the ~4-year weighted average (8108) applies to the whole book, so its ARR cannot be derived precisely. The ~45% prepayment coverage (8595) is also not tied to named contracts. The claim record supplied was truncated at claim 8594, so later disclosures are not reflected here.

IREN's disclosed counterparty record falls into three groups: signed and named contracts, suppliers mistaken for customers, and framework language that has not yet become a contract. The signed, named revenue contracts are few and large. The Microsoft GPU services agreement dated November 2, 2025 is documented at the subsidiary level (claims 4157, 1266), and its first tranche has been delivered and accepted (claims 8673, 35). The $3.4bn five-year NVIDIA AI Cloud contract is described as signed (claims 9180, 11076). Earlier relationships are mostly with suppliers or service providers rather than customers. Bitmain (claims 14392, 6374, 6152), Dell (claims 4161, 1297), BC Hydro (claims 12182, 12198) and Kraken (claims 12191, 12183) are all in this category. WEKA is described only as a partner (claim 12005). Mirantis's 1,500 enterprise customers belong to Mirantis, not IREN (claim 8768). Poolside is the first named cloud customer, but the disclosure describes only the start of service, with no contract value (claim 6285). The softer language needs care. The 5GW NVIDIA 'strategic partnership' is a collaboration framework with no stated value (claims 11077, 9289, 8642). The $2.1bn NVIDIA investment vests only as deployments occur (claim 9179). This makes NVIDIA customer, supplier-aligned partner and prospective investor at the same time, a circularity worth watching. In British Columbia, only about $0.4bn of ARR is under contract (claim 16665). The more than $0.5bn tied to 'remaining contract negotiations' is unsigned (claim 16666). The July $2.8bn tranche was announced as contracts with 'leading AI developers' (claims 360, 8641). Counterparties are named only in aggregate lists: Figure AI, Fluidstack, Fireworks AI and Hume AI (claim 8117, a secondary source), plus one still-unnamed 'leading AI developer' (claim 8101). No per-customer values are allocated. Concentration has shifted from none, in the Bitcoin-mining era, to heavy dependence on Microsoft, which is about 61% of the roughly $15.9bn of announced contract value. NVIDIA and the July cohort have since diluted that share. The revenue arithmetic broadly reconciles. $9.7bn over five years matches the stated $1.9bn average annual Microsoft revenue (claim 8996). NVIDIA adds about $0.68bn a year. Together these are about $2.6bn against $3.1bn of ARR under contract (claim 9183), leaving about $0.5bn, consistent with Prince George and legacy contracts. The July contracts carry roughly a four-year weighted term (claim 8108) and 45% prepayments (claim 8595), which imply about $0.7bn of further ARR. No updated ARR figure is on record to confirm this.

IREN's customer record splits into a small set of signed, named, dollar-denominated contracts and a larger layer of framework and partnership language. The signed core is clear. Microsoft's five-year, $9.7bn GPU services contract dated November 2, 2025 is documented at the subsidiary level (4157, 1266), and delivery has begun: Horizon 1 was accepted by Microsoft (8673, 35), with three further 50MW tranches scheduled for 2026 (8632, 8622). NVIDIA's five-year, $3.4bn AI Cloud contract is described as signed (9180, 11076) and is backed by a $1.6bn Dell hardware order (8992, 1297). The 5GW NVIDIA 'strategic partnership' is a different kind of commitment. It is framed as 'collaboration to support deployment' (9289, 362) with no stated revenue, and the related $2.1bn NVIDIA investment vests only as infrastructure is deployed (9179). It should be read as a framework, not as backlog.

Naming is weaker below that tier. The July $2.8bn of contracts is attributed in company language only to 'leading AI developers' (360, 8641). The counterparties Figure AI, Fluidstack, Fireworks AI and Hume AI come from a secondary source (8117), and the customer roster itself includes an unidentified 'new leading AI developer' (8101). The Prince George $0.4bn ARR has no named counterparty (16665), and the further >$0.5bn in British Columbia is explicitly still in 'negotiations' (16666). Bitmain, Dell, BC Hydro, Kraken, WEKA and Mirantis are suppliers or service providers, not revenue customers (14392, 4161, 12198, 12191, 12005, 8768). Mirantis's 1,500 enterprise customers are not IREN's.

Concentration has changed sharply. Before 2024 IREN's revenue came from bitcoin mining pools and had no named customers. Poolside was a small 2024 entry (6285). From November 2025 Microsoft became dominant, and NVIDIA and the July cohort have diluted that position only partially.

The arithmetic broadly reconciles. Microsoft's $1.9bn average annual revenue (8996) is consistent with $9.7bn over five years. NVIDIA's $3.4bn over five years implies about $0.68bn a year. Adding Prince George's $0.4bn gives roughly $3.0bn, against the $3.1bn ARR under contract reported in May (9183). Applying the book-wide ~4-year average term (8108) to the July $2.8bn implies roughly $0.7bn of additional ARR, but the record contains no updated ARR figure to confirm this, and it does not show whether Prince George overlaps the July cohort. The Dell commitments of $5.8bn and $1.6bn (4161, 8992) are large relative to contract value. Customer prepayments of about 45% of GPU capex (8595) cover only the 'recent agreements'. The claim record was truncated at claim 8553, so later disclosures may change this read.

IREN's disclosed relationships split cleanly into signed commercial contracts and looser partnership language. Most of the pre-2025 record describes suppliers and service providers rather than customers. The Bitmain hardware contracts (14392, 6374, 6152), BC Hydro power supply (12182, 12198) and Kraken as a Bitcoin liquidation venue (12191, 12183) are operating inputs, not revenue counterparties. The WEKA 'partnering' (12005) is a vendor relationship, and the only early named AI customer is Poolside (6285). The first large signed customer agreement is the November 2, 2025 Microsoft GPU Services contract, $9.7 billion over five years (4157, 1266). It has since been partly validated by delivery and acceptance of Horizon 1 (35, 8673), with three further 50MW tranches scheduled for 2026 (8622, 8632). The $3.4 billion five-year NVIDIA AI Cloud contract (11103, 9180, 9287) is also described as signed, with dedicated Dell hardware purchased to service it (8992, 1297). By contrast, the '5GW strategic partnership' with NVIDIA (11077, 9289, 8642) is framework language, a 'collaboration to support deployment,' with no disclosed economics. NVIDIA's $2.1 billion investment vests only as its GPU infrastructure is deployed (9179), which ties customer, supplier and shareholder roles together. The 'Exemplar Cloud' status (8543) and the Mirantis claims (8768, 8769) are certifications and partner credentials, not contracts. Identification is weakest in the newer contracts. The Prince George ARR (16665) and the July $2.8 billion package (360, 8641, 8594) are attributed to unnamed 'leading AI developers.' The customer roster (8101) includes 'a new leading AI developer' without a name, and the Figure AI, Fluidstack, Fireworks and Hume attribution comes from secondary coverage phrased as 'companies like' (8117). A further >$0.5 billion of British Columbia ARR was described in February as still under negotiation (16666), and its conversion is not confirmed on this record. Concentration has moved from no customers, to a single named AI customer, to heavy dependence on two counterparties. Microsoft's expected $1.9 billion in average annual revenue (8996) reconciles with $9.7 billion over five years. NVIDIA's contract implies about $0.68 billion a year. Together with roughly $0.4 billion at Prince George, these total about $3.0 billion, which is broadly consistent with the $3.1 billion ARR under contract disclosed in May (9183). On that basis, Microsoft and NVIDIA account for roughly 85% of contracted ARR. The July contracts, at a roughly four-year weighted term (8108), would add about $0.7 billion a year and dilute that concentration modestly. No updated ARR total confirming this appears in the record, and claim 361 is truncated and was not relied on.

IREN's record divides into a small set of signed, named revenue contracts and a larger layer of supply agreements, frameworks and unattributed deals. The clearest signed customer agreement is the five-year, $9.7bn Microsoft GPU services contract dated November 2, 2025 (4157, 1266). It has moved past paper: Horizon 1 has been delivered to and accepted by Microsoft (35, 8673), and Horizons 2-4 are scheduled for 2026 (8632). The $3.4bn five-year AI Cloud contract is named as NVIDIA and described as signed (9180, 11103), though the Dell procurement filing refers to it only as a 'managed services' contract (1299, 8995), and ramp is not targeted until early 2027 (9287). The accompanying 5GW 'strategic partnership' (11077, 9289, 8642) is framework language with no disclosed value. NVIDIA's $2.1bn investment right vests only as its GPU infrastructure is deployed (9179), which makes NVIDIA customer, supplier-aligned partner and prospective shareholder at once. Dell (4161, 1297), Bitmain (14392, 6152), BC Hydro (12198), Kraken (12183) and WEKA (12005) are suppliers or service providers, not customers. Mirantis's 1,500 enterprise customers (8768) belong to Mirantis, not IREN. The unnamed exposure is material. The July $2.8bn tranche is attributed to 'leading AI developers' (360, 8641); names appear only in secondary coverage framed as 'companies like' Figure AI, Fluidstack, Fireworks AI and Hume AI (8117), with no per-counterparty values. The customer list itself includes 'a new leading AI developer' (8101). Prince George's ~$0.4bn ARR has no named counterparty (16665), and the >$0.5bn British Columbia figure is explicitly still in negotiation (16666). Concentration has shifted from a single named 2024 GPU customer, Poolside (6285), to a book of roughly $15.9bn in announced contract value. Microsoft accounts for about 61% of that and Microsoft plus NVIDIA about 82%, so the July additions broaden the tail without changing the core dependence. The arithmetic largely reconciles. $9.7bn over five years is $1.94bn a year, matching the $1.9bn average annual revenue guided (8996). NVIDIA adds about $0.68bn, for about $2.62bn combined. The remaining roughly $0.5bn of the $3.1bn ARR under contract (9183) is consistent with Prince George's ~$0.4bn plus smaller contracts. The July $2.8bn post-dates that ARR figure, and no updated total is on record. At the ~4-year weighted term (8108), it would imply roughly $0.7bn additional ARR, but that is an inference. Reported prepayments of about 45% of GPU capex (8595) support contract substance, but the record supplied ends mid-claim at 8594.

IREN's disclosed customer book is best read in three layers: executed contracts with named counterparties, executed contracts where the counterparty goes unnamed in primary disclosure, and framework or negotiation language that should not be counted as revenue.

The first layer is anchored by Microsoft. The five-year, $9.7bn GPU services contract is dated November 2, 2025 in filing language (4157, 1266), and delivery risk has begun to retire: Horizon 1 has been delivered to and accepted by Microsoft (35, 8673), with three further 50MW tranches scheduled for 2026 (8632, 8622). The $3.4bn five-year AI Cloud contract with NVIDIA is also described as signed (9180, 11076), backed by a matched Dell purchase agreement of about $1.6bn (8992, 1297). Poolside AI is the only named cloud customer from the earlier period (6285). Bitmain, Dell, BC Hydro, Kraken and WEKA are suppliers or vendors, not customers (14392, 6152, 4161, 12198, 12183, 12005).

The second layer is weaker. The Prince George contracts (~$0.4bn ARR, 16665) carry no counterparty names, and the July $2.8bn tranche is attributed in company language only to "leading AI developers" (360, 8641). Names such as Figure AI, Fluidstack, Fireworks AI and Hume AI appear through secondary reporting and a customer-base roster (8117, 8101), not deal-by-deal attribution, and that roster itself includes an unidentified "new leading AI developer." The third layer should be excluded from contracted figures: the 5GW NVIDIA "strategic partnership" (11077, 9289, 8642) is a framework, and the >$0.5bn BC ARR sits in "remaining contract negotiations" (16666). NVIDIA's $2.1bn investment vests only as NVIDIA infrastructure is deployed (9179), which makes NVIDIA simultaneously customer, supplier-adjacent and investor.

Concentration has moved from zero customers, when IREN was a self-miner liquidating through Kraken, to near-total dependence on Microsoft, which still accounts for roughly 61% of the roughly $15.9bn in announced contract value. The arithmetic largely reconciles. Microsoft's $1.9bn average annual revenue (8996) matches $9.7bn over five years. Adding NVIDIA's ~$0.68bn per year and Prince George's ~$0.4bn gives about $3.0bn, consistent with the $3.1bn ARR under contract (9183), which suggests negotiation-stage revenue was excluded. The July $2.8bn has no disclosed tenor. At the ~4-year weighted average term (8108), it would add roughly $0.7bn of ARR, but no updated ARR figure appears in the record provided, which ends at claim 8594. The ~45% customer prepayments (8595) are not attributed to specific contracts.

IREN's counterparty record falls into three tiers, and it should be read in that order. The first tier is signed, dollar-denominated customer contracts. The Microsoft GPU services agreement dated November 2, 2025 is documented at the subsidiary level (claims 4157, 1266) and valued at $9.7 billion over five years (claims 8112, 8674). It has also moved past paper: Horizon 1 has been delivered to and accepted by Microsoft (claims 35, 8673), with three further 50MW tranches scheduled for 2026 (claims 8622, 8632). The NVIDIA $3.4 billion five-year AI Cloud contract is described as signed (claims 9180, 11076, 9287) and is backed by a matching $1.6 billion Dell hardware purchase (claims 8992, 1297). However, the ramp does not begin until early 2027.

The second tier is signed but thinly identified. The $2.8 billion of July multi-year contracts is attributed to unnamed "leading AI developers" (claims 360, 8641, 8594). Only secondary coverage links it to Figure AI, Fluidstack, Fireworks AI and Hume AI, and only with "companies like" phrasing (claim 8117). The roster in claim 8101 includes "a new leading AI developer" who remains unnamed. The Prince George ~$0.4 billion ARR is under contract without a named counterparty (claim 16665). The further >$0.5 billion is explicitly still in negotiation (claim 16666).

The third tier is framework language that carries no revenue. This includes the 5GW NVIDIA "strategic partnership" (claims 11077, 9289, 362), the WEKA partnership (claim 12005), Mirantis, whose 1,500 enterprise customers are its own and not IREN's (claim 8768), and the Exemplar certification (claim 8543). The earlier Bitmain, BC Hydro and Kraken relationships (claims 14392, 12182, 12191) are supplier and service arrangements, not customers.

Concentration has shifted from zero-customer bitcoin mining, through a single small Poolside deployment (claim 6285), to heavy dependence on two names. Microsoft accounts for roughly 61% of the ~$15.9 billion of announced contract value, and Microsoft plus NVIDIA account for about 83%. NVIDIA also appears as supplier-aligned partner and as a $2.1 billion investor whose rights vest only as its GPUs are deployed (claim 9179). That circularity deserves scrutiny.

The arithmetic broadly reconciles. The Microsoft contract works out to about $1.94 billion a year, consistent with the $1.9 billion stated (claim 8996). NVIDIA adds about $0.68 billion and Prince George about $0.4 billion, which together roughly equal the $3.1 billion ARR claimed in May (claim 9183). The July $2.8 billion, at the ~4-year weighted term (claim 8108), would add roughly $0.7 billion. Customer prepayments of about 45% of GPU capex (claim 8595) partly offset the $7.4 billion of Dell commitments (claims 4161, 1297), but execution risk on those commitments remains substantial.

IREN's record divides into a small set of signed, named, dollar-denominated agreements and a larger layer of partnership and roster language. The anchor is the Microsoft GPU Services contract dated November 2, 2025, a five-year, $9.7 billion commitment executed through IE US Hardware 3 at Childress (claims 4157, 1266). Delivery and acceptance of Horizon 1, a 50MW tranche, was confirmed in August 2026 (claims 8673, 35, 7454). The second is a signed $3.4 billion, five-year AI Cloud contract with NVIDIA for air-cooled Blackwell capacity targeting an early-2027 ramp (claims 9180, 9287). Both sit alongside supply-side obligations that are also signed: Dell purchase agreements of about $5.8 billion, guaranteed by the parent (claims 4161, 4164), and $1.6 billion (claims 1297, 8992). Earlier signed items are vendor or utility arrangements with Bitmain, BC Hydro and Kraken (claims 14392, 6152, 12198, 12183), not revenue customers.

The softer layer needs separating out. The NVIDIA 5GW 'strategic partnership' is collaboration language without disclosed economics (claims 11077, 9289), and NVIDIA's $2.1 billion investment rights vest only as its infrastructure is deployed (claim 9179). WEKA and Mirantis are technology partners, and Mirantis's 1,500 enterprise customers are not IREN's (claims 12005, 8768). The Prince George contract (~$0.4 billion ARR) has no named counterparty (claim 16665). The '>$0.5bn ARR' is explicitly still in negotiation (claim 16666). The July $2.8 billion is attributed only to 'leading AI developers' in company language (claim 360); the counterparty list comes from a secondary source hedged with 'companies like' (claim 8117). The roster itself includes an unnamed 'new leading AI developer' (claim 8101), and names such as Perplexity and Together AI appear there without any disclosed contract value.

Concentration has moved from no customers at all, when revenue came from self-mined Bitcoin, to a single named GPU customer in Poolside (claim 6285), and then to heavy Microsoft dependence. The arithmetic largely holds. $9.7 billion over five years is $1.94 billion a year, matching the stated $1.9 billion average (claim 8996). Adding $0.68 billion a year from NVIDIA and $0.4 billion from Prince George gives about $3.0 billion, against the $3.1 billion ARR reported in May (claim 9183). That leaves Microsoft at roughly 60% of contracted ARR. If the July $2.8 billion runs near the stated ~4-year weighted term (claim 8108), it would add roughly $0.7 billion. That would dilute Microsoft toward half of contracted ARR, but no updated ARR figure appears in the record provided. The supplied record also ends mid-entry, so later disclosures may be missing.

IREN's disclosure record divides cleanly along a line that most of the press coverage blurs. The genuinely binding agreements are: the Bitmain hardware purchases, described as binding on their face (14392) and amended twice (6374, 6152); the BC Hydro supply contracts with stated initial terms (12198, 12182); the Microsoft GPU services agreement dated November 2, 2025 (4157, 1266), now partially performed given delivery and acceptance of Horizon 1 (8673, 35); the NVIDIA five-year AI Cloud contract described as signed (9180, 11076); and the two Dell purchase agreements at $5.8bn and $1.6bn (4161, 1297), the first backstopped by a parent guarantee (4164). Against these sit softer constructions. The 5GW NVIDIA arrangement is consistently labeled a strategic partnership and a collaboration across a pipeline (11077, 9289, 8642, 362), with no contract value, no term, and no committed capacity — it is framework language, not backlog. Prince George splits explicitly: ~$0.4bn ARR under contract versus >$0.5bn still in negotiation (16665, 16666), and only the former belongs in any forward number.

On identification, the record is better than typical. The July 2026 cohort names Perplexity, Figure AI, Together AI, Fluidstack, Fireworks AI, Fal AI and Hume AI (8101), with one counterparty left as a new leading AI developer — the sole unnamed party in a $2.8bn tranche (8641, 8117). The Mirantis material (8768, 8769) describes NVIDIA's ecosystem program and Mirantis's own customer count, not an IREN relationship, and should not be counted.

The arithmetic largely holds. Microsoft at $9.7bn over five years implies $1.94bn annually, matching the disclosed $1.9bn average (8996, 8674). NVIDIA at $3.4bn over five implies ~$0.68bn. Those two sum to ~$2.6bn against the $3.1bn ARR under contract stated May 10 (9183), with Prince George's $0.4bn plausibly bridging the gap — and the July $2.8bn at a roughly four-year weighted term (8108) adding ~$0.7bn thereafter. Two cautions: ARR here is contracted-at-full-ramp, with NVIDIA targeting early 2027 (9287); and Dell hardware consumes roughly 60% of the Microsoft contract value and 47% of NVIDIA's before power and operations. Concentration has genuinely broadened from a single hyperscaler in late 2025, though NVIDIA now sits as customer, indirect supplier and conditional $2.1bn investor (9179) simultaneously.

IREN's disclosure record separates cleanly into contracts that are executed and counterparties that are merely gestured at, and the distinction has widened as the dollar figures have grown. The definitively signed, named-counterparty agreements are Microsoft's five-year, $9.7bn GPU services contract (claims 4157, 1266, 8674), NVIDIA's five-year, $3.4bn AI Cloud contract (claims 11076, 9180, 9287), and the two Dell supply agreements at approximately $5.8bn and $1.6bn (claims 4161, 1297, 8992). These carry contract-formation language — 'entered into,' 'agreement dated,' 'unconditionally guarantee' (claim 4164) — and in Microsoft's case have progressed to performance, with Horizon 1 delivered and accepted (claims 8673, 35). The earlier record is similar in form but different in kind: Bitmain (claims 14392, 6374, 6152), BC Hydro (claims 12198, 12182) and Kraken (claims 12183, 12191) are suppliers, utilities and liquidation venues, not revenue counterparties, and should not be read as customer wins.

The softer tier deserves separate treatment. The '5GW strategic partnership' with NVIDIA (claims 11077, 9289, 362) is framework language with no consideration attached, and the associated $2.1bn NVIDIA investment vests only as GPU infrastructure is deployed (claim 9179) — a conditional commitment, not capital received. The Prince George disclosure splits itself honestly: ~$0.4bn ARR 'now under contract' (claim 16665) against '>$0.5bn ARR' still in 'remaining contract negotiations' (claim 16666), with no counterparty named on either side. Claim 8117's '$2.8 billion deal with companies like Figure AI, Fluidstack, Fireworks AI, and Hume AI' is illustrative rather than an allocation, and claim 8101's roster ends with 'a new leading AI developer' left unidentified.

Concentration has moved in the right direction. Through November 2025 the AI book was effectively one name; the NVIDIA contract added a second, and the July 2026 $2.8bn tranche (claims 360, 8594) spread the next increment across at least six named developers. The arithmetic broadly reconciles. Microsoft at $1.9bn average annual revenue (claim 8996) plus NVIDIA at roughly $0.68bn annualized plus Prince George's $0.4bn sums to about $3.0bn against the $3.1bn ARR figure (claim 9183) — close enough that the residual is plausibly Poolside-era and smaller cloud business. The $0.5bn under negotiation is correctly excluded. The July $2.8bn, spread over the stated four-year weighted average term (claim 8108), implies roughly $0.7bn of incremental ARR rather than a step to $5.9bn, a distinction some secondary coverage blurs.

The record separates cleanly into executed agreements and everything else, and the executed column is shorter than the announcement cadence implies. Genuinely binding and counterparty-named: the Bitmain hardware contracts (14392, 6374, 6152), the BC Hydro supply contracts (12182, 12198), the Kraken liquidation relationship (12183, 12191), the Microsoft GPU Services Agreement dated 2 November 2025 (4157, 1266), the Dell purchase agreements at $5.8bn and $1.6bn with a parent guarantee (4161, 4164, 1297), and the five-year $3.4bn NVIDIA AI Cloud contract (11076, 9287, 9180). Note that four of those seven are procurement or utility relationships, not revenue. Poolside AI (6285) is described only as a service commencement, with no contract value or term disclosed.

The softer language clusters where the numbers are largest. The NVIDIA 5GW arrangement is consistently worded as a strategic partnership and a collaboration to support deployment (11077, 9289, 8642), with no contracted capacity or price attached; the associated $2.1bn NVIDIA investment vests only as GPUs are deployed (9179), which is an option, not a commitment. Claim 16666 is explicit that >$0.5bn of ARR rests on remaining contract negotiations, while the adjacent $0.4bn at Prince George is under contract but the counterparty is unnamed (16665). The July 2026 $2.8bn cohort is attributed to Figure AI, Fluidstack, Fireworks AI and Hume AI only through hedged construction — companies like (8117) — and the roster in 8101 closes with an unidentified new leading AI developer. The Mirantis items (8768, 8769) describe Mirantis and NVIDIA, not IREN, and should be pulled from this file.

Concentration has broadened in name count while staying narrow in dollars. Against the $3.1bn ARR stated on 10 May 2026 (9183), Microsoft alone contributes $1.9bn (8996), consistent with $9.7bn over five years, or roughly 61 percent; NVIDIA at $0.68bn annualised and Prince George at $0.4bn account for nearly all the balance. That arithmetic reconciles. The July additions do not yet: $2.8bn across a roughly four-year weighted average term (8108) implies about $0.7bn of incremental ARR, and one should confirm the $0.5bn still in negotiation (16666) is not being counted alongside the $0.4bn already contracted. Separately, Dell hardware absorbs 60 percent of the Microsoft contract value and 47 percent of NVIDIA's, and the 45 percent prepayment offset (8595) is scoped only to the recent agreements.

IREN's disclosure record separates cleanly into signed obligations and softer framing, and the distinction matters because the two are often announced in the same sentence. On the signed side sit the Microsoft GPU Services Agreement dated November 2, 2025 (claim 4157, restated at claims 1266 and 8674), the five-year NVIDIA AI Cloud contract executed in early May 2026 (claim 9180, described as signed that week), and the July 20, 2026 tranche of multi-year cloud services contracts totaling $2.8 billion (claims 360, 8641). Alongside these run supplier commitments that are contracts but not revenue: the Dell purchase agreements at roughly $5.8 billion (claim 4161) and $1.6 billion (claims 1297, 8992), the latter explicitly procured to service the NVIDIA contract, plus the parent guarantee at claim 4164. The earlier Bitmain agreements (claims 14392, 6374, 6152), BC Hydro supply (claims 12198, 12183) and Kraken liquidation (claim 12191) belong to the same cost-side category and should not be read as demand validation.

The genuinely soft language clusters around NVIDIA. The 5GW strategic partnership (claims 11077, 9289, 8642) is collaboration framing across a pipeline, with no stated consideration, and the $2.1 billion NVIDIA investment vests only as GPU infrastructure is deployed (claim 9179), making it contingent rather than banked. The February 2026 British Columbia disclosure is the clearest instance of mixed framing: roughly $0.4 billion ARR is stated as under contract at Prince George (claim 16665) while a further figure above $0.5 billion ARR rests on remaining contract negotiations (claim 16666), and in neither case is the counterparty identified. The July 2026 roster is otherwise unusually specific by name, listing Perplexity, Figure AI, Together AI, Fluidstack, Fireworks AI, Fal AI and Hume AI (claims 8101, 363), though one participant remains only a new leading AI developer.

Concentration has therefore moved from near-total Microsoft dependence toward a broader book, but not as far as the roster suggests. The arithmetic mostly holds: Microsoft at $1.9 billion average annual revenue (claim 8996) plus NVIDIA at roughly $0.68 billion annualized plus the Prince George $0.4 billion lands near the $3.1 billion ARR under contract reported May 10 (claim 9183). Spreading July's $2.8 billion over the approximately four-year weighted average term (claim 8108) implies around $0.7 billion of incremental ARR, leaving Microsoft still above half.

The disclosure record splits cleanly along a line that IREN itself does not always draw: direction of cash flow. The earliest named counterparties are all suppliers or service venues, not customers. Bitmain appears under genuinely binding language (claim 14392, extended at claims 6374 and 6152), BC Hydro is a utility supply relationship (claims 12182, 12198), Kraken is a liquidation venue (claims 12183, 12191), and WEKA (claim 12005) and Mirantis (claims 8768, 8769) are software partners. Dell, despite $5.8bn and $1.6bn headline figures (claims 4161, 1297, 8992) and a parent guarantee (claim 4164), is a procurement obligation, not revenue. Read strictly, the first named revenue counterparty is Poolside AI in February 2024 (claim 6285), and it was never quantified.

Only two customer contracts are both named and sized: Microsoft at $9.7bn over five years, dated November 2, 2025 (claims 4157, 1266, 8674), and NVIDIA at $3.4bn over five years (claims 11076, 9180, 8995). The Microsoft deal has since reached performance milestones — Horizon 1 delivered and accepted (claims 8673, 8631) — which is the strongest evidence on the record. Against that, the 5GW NVIDIA arrangement is framework language (claims 11077, 9289, 8642) carrying no contracted revenue, the $2.1bn NVIDIA investment vests only on deployment (claim 9179), and Exemplar Cloud status (claim 8543) is a certification.

The unnamed tier matters. Prince George's ~$0.4bn ARR is described as under contract without a counterparty (claim 16665), while the adjacent >$0.5bn is explicitly still in negotiation (claim 16666) — correctly excluded, but easy to conflate. The July 2026 roster (claim 8101) names eight parties plus 'a new leading AI developer,' and the associated $2.8bn (claims 360, 8641) is never allocated among them.

The arithmetic reconciles. Microsoft's $9.7bn over five years implies $1.94bn annually, matching the disclosed $1.9bn (claim 8996); adding NVIDIA's $0.68bn and Prince George's $0.4bn yields roughly $3.0bn against the $3.1bn ARR claimed on May 10 (claim 9183). Concentration has moved from effectively total Microsoft dependence at end-2025 to roughly 40 percent post-July, but the diversification rests on counterparties whose individual contract values remain undisclosed.

The contracted record here is cleaner than most in the sector, but it requires separating vendors from customers. Bitmain (14392, 6374, 6152), BC Hydro (12182, 12198), Kraken (12183) and Dell (4161, 1297, 8992) are all named and all binding, yet none is a revenue counterparty; they are hardware supply, power transmission, liquidation venue and GPU procurement respectively. WEKA (12005) and Mirantis (8768, 8769) are described in partnership language and carry no disclosed revenue attached. Stripping those out, the genuine signed customer list is short: Poolside AI (6285), Microsoft (4157, 1266), NVIDIA (11076, 9180), and the July 2026 cohort of Perplexity, Figure AI, Together AI, Fluidstack, Fireworks AI, Fal AI and Hume AI (8101).

On identification, one counterparty in that cohort remains unnamed, disclosed only as a new leading AI developer (8101), and the $2.8bn July tranche is announced in aggregate against leading AI developers with no per-counterparty allocation (360, 8641). The attribution in 8117 naming Figure AI, Fluidstack, Fireworks AI and Hume AI is third-party phrasing — companies like — not a company-supplied breakdown, so the individual contract values are unverifiable. The one genuinely non-binding item is the 5GW NVIDIA strategic partnership (11077, 9289, 8642), framed as collaboration across a pipeline rather than a contract, and it should not be added to the $3.4bn. NVIDIA's $2.1bn investment (9179) vests only as GPU infrastructure is deployed, making it contingent rather than committed.

Concentration has moved from near-total to merely top-heavy. Of roughly $15.9bn in announced contract value, Microsoft is about 61 percent and Microsoft plus NVIDIA about 82 percent, but the July additions take the named customer count from two to nine or ten.

The arithmetic reconciles. Microsoft's $9.7bn over five years implies $1.94bn annually, consistent with the disclosed $1.9bn average (8996); NVIDIA's $3.4bn implies $0.68bn. Together that is $2.62bn against $3.1bn ARR under contract at 10 May (9183), leaving roughly $480m from legacy AI cloud and mining-adjacent business — plausible. July's $2.8bn at the stated four-year weighted average term (8108) adds about $700m. The caveat is timing, not accuracy: ARR under contract is not recognized revenue, NVIDIA ramps from early 2027 (9287), and only Horizon 1 of four Microsoft deployments has been accepted (8673, 8622), with $1.6bn of Dell capex (1297) offset by prepayments near 45 percent (8595).

Two counterparties carry the weight of IREN's contracted revenue, and both are named in primary documents rather than press paraphrase. The Microsoft GPU Services agreement dated November 2, 2025 is a signed, tranche-structured contract at the Childress "Horizon" facilities (claim 4157, restated at claim 1266), and the $3.4bn five-year NVIDIA AI Cloud contract is described as executed rather than contemplated, "this week signed" (claim 9180; also claims 11076, 9287). Everything else on the 2026 record carrying the word partnership is materially softer. The 5GW NVIDIA arrangement is collaboration language with no term, no committed capacity and no consideration attached (claims 11077, 9289); NVIDIA's $2.1bn investment vests only as GPU infrastructure is deployed (claim 9179), which is a contingent equity right, not demand. Exemplar Cloud status (claim 8543), the WEKA partnership (claim 12005) and Mirantis (claims 8768, 8769) are vendor and certification relationships; claim 8768 in fact enumerates Mirantis's 1,500 customers, not IREN's.

Concentration has inverted in character. Through 2023 every named counterparty sat on the supply or treasury side: Bitmain (claims 14392, 6374, 6152), BC Hydro (claims 12182, 12198) and Kraken for liquidation (claims 12183, 12191), with no named revenue customer until Poolside AI in February 2024 (claim 6285). Microsoft then took the book to effectively single-customer. The $2.8bn announced in July 2026 (claim 360) broadened it, but was disclosed against "leading AI developers" with no per-deal allocation; names surface later and loosely, as "companies like Figure AI, Fluidstack, Fireworks AI, and Hume AI" (claim 8117), and claim 8101's roster still closes with "a new leading AI developer," unidentified.

The arithmetic broadly reconciles. $9.7bn over five years implies $1.94bn annually, consistent with the disclosed $1.9bn average (claim 8996); NVIDIA adds roughly $680m. That sums to about $2.6bn against $3.1bn of ARR under contract (claim 9183), leaving some $0.5bn attributable to pre-existing AI Cloud business. Note the metric is a fully-deployed run rate, not realized revenue: Horizons 2 through 4 deliver across 2026 (claims 8632, 8622) and NVIDIA ramps from early 2027 (claim 9287). Cost is disclosed as firmly as revenue, with Dell at $5.8bn against Microsoft (claim 4161, parent-guaranteed at claim 4164) and $1.6bn against NVIDIA (claim 1297), 60% and 47% of gross contract value, partly offset by prepayments near 45% of GPU capex (claim 8595).

IREN's signed-customer record is short but substantive, and it should be read apart from the partnership vocabulary around it. The anchor is the five-year, $9.7 billion Microsoft GPU services contract dated November 2, 2025, which is disclosed at the level of the contracting subsidiary, the site, and the tranche structure (4157, 1266). It has since produced a verifiable milestone: Horizon 1, a 50MW deployment and the first of four facilities, was delivered to and accepted by Microsoft (8673, 35, 7454, 8553). The second signed agreement is the five-year, $3.4 billion NVIDIA AI Cloud contract for air-cooled Blackwell capacity at Childress, targeting a ramp from early 2027 (9287, 9180). Later filings call it a 'managed services' contract (8995, 1299), a change in wording that is worth monitoring.

Much of the rest is framework or supplier language. The 5GW NVIDIA 'strategic partnership' is described as collaboration on infrastructure alignment, not as committed revenue (9289, 362). NVIDIA's $2.1 billion investment vests only as its GPUs are deployed (9179), which makes NVIDIA customer, supplier-adjacent, and prospective shareholder at once. The Dell agreements of $5.8 billion and $1.6 billion (4161, 1297) are capital obligations that IREN guarantees (4164), not revenue. WEKA, Kraken, BC Hydro, Bitmain, and Mirantis are vendors or partners, and Mirantis's 1,500 enterprise customers (8768) are not IREN's. Poolside was announced as a service commencement with no disclosed value (6285).

Naming is where the record thins out. The July 2026 $2.8 billion tranche is attributed only to 'leading AI developers' (360), and the roster adds 'a new leading AI developer' without identifying it (8101). One secondary report assigns the deal to Figure AI, Fluidstack, Fireworks AI, and Hume AI (8117), but no company-sourced claim on record confirms that allocation. Perplexity, Together AI, and Fal AI appear on the roster (363) with no contract values attached.

Concentration has eased but remains high. Microsoft accounts for roughly 61% of the approximately $15.9 billion in announced contract value, down from 100% in late 2025.

The revenue arithmetic mostly reconciles. $9.7 billion over five years implies about $1.94 billion a year, consistent with the stated $1.9 billion (8996). NVIDIA adds about $0.68 billion a year. The $3.1 billion ARR figure from May (9183) therefore implies about $0.5 billion from smaller, unitemized customers. The July contracts, at a weighted average term of roughly four years (8108), could add around $0.7 billion. However, no updated ARR figure appears on record to confirm this.

IREN's signed customer book is heavily concentrated in two named counterparties, and the named customers account for only part of the headline contract value. The largest signed agreement is the five-year, $9.7 billion GPU services contract with Microsoft dated November 2, 2025 (claims 4157, 1266). It is backed by a parent-guaranteed $5.8 billion Dell hardware purchase (claims 4161, 4164). The first of four builds, the 50MW Horizon 1, was delivered to and accepted by Microsoft in August 2026 (claims 8673, 35, 8553, 7454), which moves this deal from signed paper to performing contract. The second is the five-year, $3.4 billion AI Cloud contract with NVIDIA, described as signed (claims 9180, 11103), with a $1.6 billion Dell purchase to service it (claims 8992, 1297). Its ramp is targeted only from early 2027 (claim 9287). The accompanying 5GW strategic partnership (claims 11077, 9289, 362) is framework language with no disclosed volume or price. NVIDIA's $2.1 billion investment vests only as NVIDIA infrastructure is deployed (claim 9179). That makes NVIDIA customer, investor and architecture partner at once, a relationship diligence should treat as partly circular. The July 2026 $2.8 billion of multi-year contracts (claim 360) is the weakest-identified tranche. It is attributed to companies like Figure AI, Fluidstack, Fireworks AI and Hume AI (claim 8117), and the roster adds an unnamed new leading AI developer (claim 8101). No per-counterparty split is given. Earlier relationships with WEKA, Bitmain, BC Hydro, Kraken and Dell (claims 12005, 6374, 12182, 12191) are supplier or service arrangements, not revenue customers. Mirantis's 1,500 enterprise customers (claim 8768) are not IREN's. Poolside (claim 6285) was the first disclosed GPU cloud customer, but no contract value was given. Concentration has moved from effectively zero named customers in the mining era to Microsoft dominance. On the arithmetic, $9.7 billion over five years reconciles with the stated $1.9 billion average annual revenue (claim 8996). NVIDIA adds roughly $0.68 billion a year. Together these are about $2.6 billion of the $3.1 billion ARR under contract (claim 9183), so Microsoft is roughly 60 percent. The residual of about $0.5 billion is attributed to no named customer. The July tranche, at the roughly 4-year weighted average term (claim 8108), implies about $0.7 billion more ARR. That arithmetic is inferred, not disclosed, and it would dilute Microsoft to about half.

IREN's signed-customer base is narrower than its announcement volume suggests, but the core contracts are real and executed. The Microsoft GPU services agreement dated November 2, 2025 is documented at the contracting-entity level (4157, 1266). It is backed by a guaranteed $5.8 billion Dell purchase agreement (4161, 4164), and Microsoft has accepted delivery of Horizon 1 (8673, 35). The NVIDIA $3.4 billion five-year AI Cloud contract is described as signed (9180) and paired with a $1.6 billion Dell hardware order (1297, 8992). Earlier supplier-side agreements with Bitmain were binding purchase contracts (14392, 6374, 6152), though they are not customer revenue. Softer language sits alongside these deals. WEKA was a 'partnering' arrangement with no stated economics (12005). Poolside was disclosed as a 'commencement of GPU cloud service' without value or term (6285). The 5GW NVIDIA 'strategic partnership' is a collaboration framework, not a revenue commitment (11077, 9289, 362). NVIDIA's $2.1 billion investment rights vest only as deployments occur (9179). Mirantis contributes capability, not customers (8768, 8769).

Naming is uneven. The July $2.8 billion of contracts was attributed to 'leading AI developers' (360), and the company never itemized counterparties by contract value. Figure AI, Fluidstack, Fireworks AI and Hume AI were linked to that tranche only through secondary press phrasing, 'companies like' (8117). The company's own roster adds 'a new leading AI developer' who remains unidentified (8101), and roster names such as Perplexity, Together AI and Fal AI carry no disclosed contract values (363).

Concentration has moved from a mining business dependent on BC Hydro power and Kraken liquidity (12182, 12191) to a heavy dependence on hyperscalers. Of roughly $15.9 billion in announced contract value, Microsoft accounts for about 61% (8674). Microsoft and NVIDIA together account for about 82%. The July tranche dilutes this somewhat but does not change the profile.

The revenue arithmetic broadly reconciles. $9.7 billion over five years is about $1.94 billion a year, which matches the stated $1.9 billion average (8996). NVIDIA adds about $0.68 billion a year. Together these leave roughly $0.5 billion of the $3.1 billion ARR under contract (9183) to the remaining, largely unnamed book. The July contracts, at a roughly four-year weighted term (8108), imply about $0.7 billion a year more, but no updated ARR figure appears on record to confirm this. Contracted ARR also still depends on delivery. Only the first of four Microsoft data centers has been accepted (8553), and the NVIDIA ramp targets early 2027 (9287).

IREN's signed-contract base is narrower and more concentrated than its customer roster suggests. Only two AI cloud agreements are documented with named counterparties, a stated term and a stated value. The first is the five-year, $9.7bn GPU services contract with Microsoft dated November 2, 2025 (claims 4157, 1266), for which Horizon 1 has been delivered and accepted (claims 8673, 35). The second is the five-year, $3.4bn AI Cloud contract with NVIDIA, described as signed (claims 9180, 11076). The Dell agreements of $5.8bn and $1.6bn (claims 4161, 1297) are signed, but they are supply commitments backing those two contracts, not revenue. Earlier Bitmain contracts (claims 14392, 6152) and the Kraken and BC Hydro relationships (claims 12191, 12182) are also supplier or service arrangements.

Much of the remaining language should be read as framework rather than contract. The 5GW NVIDIA "strategic partnership" (claims 11077, 9289, 362) is a collaboration across the pipeline with no stated value. NVIDIA's $2.1bn investment vests only as GPU infrastructure is deployed (claim 9179). The WEKA partnership (claim 12005), NVIDIA Exemplar Cloud status (claim 8543) and Mirantis's 1,500 enterprise customers (claim 8768) are vendor or certification items, and the Mirantis customers are not IREN customers. The July $2.8bn tranche is described only as contracts with "leading AI developers" (claim 360). One party appears solely as "a new leading AI developer" (claim 8101). The attribution to Figure AI, Fluidstack, Fireworks AI and Hume AI comes from secondary coverage using "companies like" (claim 8117), with no per-counterparty values disclosed. Poolside, announced in 2024 (claim 6285), is absent from the 2026 rosters (claims 363, 8101).

Concentration moved from no identified customers in the mining era to effectively one. Microsoft's roughly $1.9bn in average annual revenue (claim 8996) was about 61% of the $3.1bn ARR under contract reported in May (claim 9183). The July tranche dilutes that share, but the roster of nine or ten names mostly represents smaller, unquantified contracts.

The arithmetic broadly reconciles. $9.7bn over five years is about $1.94bn a year, consistent with claim 8996. NVIDIA's $3.4bn adds about $0.68bn, for a combined $2.6bn. That leaves roughly $0.5bn of the $3.1bn ARR attributable to undisclosed contracts. If the July $2.8bn runs near the four-year weighted average term (claim 8108), it adds about $0.7bn a year. No updated post-July ARR figure is on record to confirm this.

IREN's record holds two customer agreements that are both signed and valued. The first is the five-year, $9.7 billion Microsoft GPU services contract dated November 2, 2025 (claims 4157, 1266, 8674). Microsoft's acceptance of Horizon 1, the first of four 50MW deployments at Childress, moves it from paper to performance (claims 35, 8673, 8622). The second is the five-year, $3.4 billion NVIDIA AI Cloud contract, which management describes as signed (claims 9180, 11076, 9287). Much of the remaining record is not customer revenue at all. Bitmain and Dell are suppliers, and the Dell purchases of $5.8 billion and $1.6 billion are the capex side of those two contracts (claims 14392, 6152, 4161, 1297). BC Hydro and Kraken are a power provider and a liquidation venue (claims 12182, 12183). WEKA is a software partner (claim 12005).

The softer language clusters around NVIDIA. The '5GW strategic partnership' has no stated value or binding terms (claims 11077, 9289, 8642). NVIDIA is also a counterparty to a $2.1 billion investment whose rights vest only as NVIDIA infrastructure is deployed (claim 9179). That makes NVIDIA a customer, a vendor ecosystem and a prospective shareholder at once, and those roles should be read together. The July 20 package of $2.8 billion is attributed to unnamed 'leading AI developers' (claims 360, 8641). Only secondary coverage links it to Figure AI, Fluidstack, Fireworks AI and Hume AI (claim 8117). The company's own roster adds 'a new leading AI developer' without naming it (claim 8101). Poolside AI is named, but no contract value is given (claim 6285).

Concentration has moved from a model with essentially no customers, Bitcoin sold through Kraken, to heavy reliance on one hyperscaler. Of roughly $15.9 billion in announced contract value, Microsoft accounts for about 61%, NVIDIA about 21% and the July cohort about 18%. The spread is widening, but slowly.

The arithmetic partly reconciles. $9.7 billion over five years is $1.94 billion a year, which matches the guided $1.9 billion (claim 8996). NVIDIA adds about $0.68 billion a year. Together they come to about $2.6 billion, against the $3.1 billion of ARR under contract stated on May 10 (claim 9183). That leaves roughly $0.5 billion attributable to smaller, unvalued contracts. If the July $2.8 billion runs near the roughly four-year weighted term (claim 8108), it adds about $0.7 billion, implying ARR near $3.8 billion. No updated figure appears on record to confirm this. The 45% customer prepayment figure is not tied to specific agreements (claim 8595). The record supplied ends mid-claim at 363, so later disclosures may be missing.

IREN's record splits between a few executed, named contracts and a growing layer of framework and roster language. The Microsoft agreement is the firmest item. It is a GPU services contract with Microsoft Corporation dated November 2, 2025 (claims 4157, 1266), and it is now partly performed: Horizon 1 has been delivered and accepted (8673, 35), while Horizons 2-4 remain scheduled for 2026 (8632). Management describes the $3.4bn NVIDIA AI Cloud contract as signed (9180), with a stated term, GPU type and site (9287). The 5GW NVIDIA arrangement, by contrast, is only ever called a strategic partnership or collaboration (9289, 8642), with no value attached. NVIDIA's $2.1bn investment rights vest only as NVIDIA infrastructure is deployed (9179), which makes NVIDIA customer, investor and architecture sponsor at once. Dell (4161, 1297) and Bitmain (14392, 6374, 6152) are signed supply agreements rather than revenue counterparties, and Kraken, BC Hydro and WEKA (12183, 12182, 12005) are vendor or service relationships.

The July $2.8bn tranche is the least clearly attributed. The company's own wording is contracts with leading AI developers (360, 8641). The names Figure AI, Fluidstack, Fireworks AI and Hume AI come from a secondary source that prefaces them with companies like (8117). The customer roster also includes an unnamed new leading AI developer (8101) and gives no per-counterparty values. The Mirantis claims (8768, 8769) describe Mirantis's customers, not IREN's, and roster claim 363 is truncated in the record.

Concentration has moved from no customers at all, when IREN was a bitcoin miner selling through pools and Kraken, to a first cloud customer in Poolside (6285), and then to heavy dependence on Microsoft. Microsoft accounts for about $9.7bn of $15.9bn in announced contract value, roughly 61%. NVIDIA is near 21% and the July group near 18%.

The arithmetic largely reconciles. $9.7bn over five years is $1.94bn a year, which matches the $1.9bn annual figure (8996). NVIDIA adds about $0.68bn a year, for roughly $2.6bn in total. Reported ARR under contract that same week was $3.1bn (9183), which leaves about $0.5bn attributable to smaller, unquantified contracts. At the company-wide four-year weighted average term (8108), the July $2.8bn would add about $0.7bn a year, but the record contains no updated ARR figure. Dell hardware costs of $5.8bn and $1.6bn equal about 60% and 47% of the related contract values. That burden is partly offset by prepayments of about 45% of GPU capex on unspecified recent agreements (8595).

IREN's disclosure record splits into a few firmly signed, named contracts and a growing layer of partnership and aggregate-deal language. Before late 2025 the named counterparties were almost all suppliers or service providers rather than customers. These were binding Bitmain hardware contracts (claims 14392, 6374, 6152), BC Hydro power agreements (12182, 12198), Kraken for Bitcoin liquidation (12183) and a WEKA 'partnering' (12005). Poolside AI (6285) was the only named AI cloud customer, and no contract value was disclosed. The Microsoft agreement dated November 2, 2025 is the first clearly signed, named and priced customer contract: $9.7bn over five years (4157, 1266). Its delivery evidence is concrete, with Horizon 1 accepted by Microsoft (8673, 35) and three further 50MW tranches scheduled for 2026 (8632). The NVIDIA $3.4bn five-year contract is described as signed (9180) and is backed by a $1.6bn Dell purchase (8992, 1297). The accompanying 5GW 'strategic partnership' (11077, 9289, 8642) is framework language with no committed revenue. NVIDIA's $2.1bn investment vests only as infrastructure is deployed (9179), which makes it contingent rather than contractual. The July $2.8bn tranche is the weakest on identification. The primary phrasing is 'leading AI developers' (360, 8641). Named parties such as Figure AI, Fluidstack, Fireworks AI and Hume AI appear only in secondary 'companies like' framing (8117). A roster that adds Perplexity, Together AI and Fal AI still includes 'a new leading AI developer' left unnamed (8101), and no per-counterparty values are given. Mirantis's 1,500 enterprise customers (8768) are Mirantis's, not IREN's. Concentration is extreme but slowly easing. Of roughly $15.9bn in announced contract value, Microsoft accounts for about 61%, and Microsoft plus NVIDIA for about 82%. On the revenue arithmetic, $9.7bn over five years matches the stated $1.9bn average annual Microsoft revenue (8996). Adding NVIDIA's roughly $0.68bn gives about $2.6bn, which leaves roughly $0.5bn of the $3.1bn ARR under contract (9183) attributable to undisclosed smaller contracts. The July tranche cannot be converted to ARR without term data. The stated weighted average term of about four years (8108) is hard to reconcile by contract value: two five-year contracts make up over 80% of value, so four years implies either weighting by remaining term or ARR, or a meaningful book of short-dated contracts. The roughly 45% customer prepayment figure (8595) applies to unspecified 'recent agreements' and should not be generalized.

IREN's signed-contract base is narrower than its partnership vocabulary suggests. Only two customer agreements are documented as executed with named counterparties. The first is the November 2, 2025 GPU services contract with Microsoft Corporation, worth $9.7 billion over five years (claims 4157, 1266). The second is the $3.4 billion five-year AI Cloud contract with NVIDIA, described as signed (claims 9180, 11076). Both are backed by named supply agreements with Dell, at $5.8 billion under a parent guarantee (claims 4161, 4164) and at $1.6 billion (claims 1297, 8992). Microsoft's acceptance of Horizon 1 (claims 8673, 35) is the first evidence of performance, and it is only one of four 50MW deployments (claim 8622). Much of the rest is framework language. The 5GW NVIDIA strategic partnership is a collaboration on architecture across a pipeline, not a revenue commitment (claims 11077, 9289, 8642). The $2.1 billion NVIDIA investment vests only as GPU infrastructure is deployed (claim 9179). The 2023 WEKA 'partnering' (claim 12005) has no disclosed economics. Kraken, BC Hydro and Bitmain (claims 12183, 12198, 6374, 6152) are service providers and suppliers, not customers. The July 2026 $2.8 billion of contracts is attributed in company language only to unidentified 'leading AI developers' (claims 360, 8641). A secondary source links it to Figure AI, Fluidstack, Fireworks AI and Hume AI (claim 8117), but that appears to borrow from a general customer roster (claim 8101), which itself includes an unnamed 'new leading AI developer.' Counterparty concentration has shifted in stages. Before 2024 IREN was a bitcoin miner with no compute customers, and Poolside AI (claim 6285) was a small first step into GPU cloud. Microsoft alone then dominated contracted value from late 2025. NVIDIA was added in May 2026, and the July contracts broadened the base, though they still amount to less than a third of the Microsoft figure. The revenue arithmetic mostly reconciles. $9.7 billion over five years is about $1.94 billion a year, which matches the stated $1.9 billion (claim 8996). NVIDIA's contract adds about $0.68 billion a year, for roughly $2.6 billion combined. That leaves about $0.5 billion of the $3.1 billion ARR under contract (claim 9183) unattributed to any named deal. At the roughly four-year weighted term (claim 8108), the July contracts would add around $0.7 billion a year, but no updated ARR is on record. The 45% customer prepayment figure (claim 8595) is not tied to specific contracts. The claim record is truncated after claim 362, and some amount fields use inconsistent units, so later disclosures may change this picture.

IREN's disclosed record contains a small number of genuinely signed, dollar-denominated customer contracts and a larger layer of partnership language around them. Investors should keep the two apart. The anchor is the Microsoft GPU Services contract dated November 2, 2025. It is identified by counterparty, contracting subsidiary (IE US Hardware 3), site and five-year term (claims 4157, 1266), and is repeated at $9.7 billion throughout later coverage (8112, 8674, 8553). It is backed by a named $5.8 billion Dell supply agreement with a parent guarantee (4161, 4164). The May 2026 NVIDIA contract is similarly concrete. Management said it "signed" a five-year, $3.4 billion AI Cloud contract (9180, 11076), specified air-cooled Blackwell capacity within 60MW at Childress (9287), and matched it with a separately priced $1.6 billion Dell purchase (1297, 8992). By contrast, the "broader 5GW strategic partnership" with NVIDIA (11077, 9289, 362) is framework language. It describes collaboration across a pipeline with no stated value or obligations. The $2.1 billion NVIDIA investment vests only as NVIDIA infrastructure is deployed (9179), which makes NVIDIA simultaneously customer, prospective shareholder and architecture partner. The earlier WEKA "partnering" (12005) and Poolside AI service commencement (6285) are too thinly described to size.

Naming is weakest in the July 2026 tranche. The primary disclosure refers to $2.8 billion of contracts with "leading AI developers" (360, 8641). Attribution to Figure AI, Fluidstack, Fireworks AI and Hume AI comes from secondary coverage framed as "companies like" (8117). The roster itself includes an unidentified "new leading AI developer" (8101), and no per-counterparty values have been given. Concentration has therefore moved from a single supplier relationship with Bitmain (6374, 6152) to extreme customer concentration. On announced totals of roughly $15.9 billion, Microsoft represents about 61%, NVIDIA about 21%, and the diffuse July group about 18%.

The revenue arithmetic largely reconciles where data exist. $9.7 billion over five years is about $1.94 billion a year, consistent with the stated $1.9 billion average (8996). NVIDIA's contract implies roughly $0.68 billion a year. Together these give about $2.6 billion against $3.1 billion of ARR under contract in May (9183), leaving about $0.5 billion unitemized from smaller customers. No updated ARR has been disclosed after the July $2.8 billion, although a roughly four-year weighted term (8108) would suggest about $0.7 billion more. Realization is early. Only Horizon 1, the first of four 50MW Microsoft deployments, has been accepted (8673, 8622), and NVIDIA revenue is not targeted to ramp until early 2027 (9287). The 45% prepayment figure (8595) is not tied to named contracts.

IREN's disclosed customer base reveals a dramatic shift from foundational to hyperscaler concentration, with critical distinctions between signed service agreements and ancillary supply or investment instruments that merit careful parsing. The earliest announcements—Bitmain Technologies (claim 6374, January 2024) and Poolside AI (claim 6285, February 2024)—represent genuine customer agreements, the former explicitly stating entry into an agreement, though their revenue magnitude remains undisclosed and both have vanished from customer rosters by mid-2026. By November 2025, IREN pivoted entirely to named, substantial counterparties: Microsoft ($9.7 billion over five years, claim 8112), NVIDIA ($3.4 billion over five years, claim 11076), and Dell ($5.8 billion, claim 4161). Critical to this accounting is distinguishing customer from supplier agreements; Dell's engagement is a GPU supply pact underwriting IREN's fulfillment obligations to others, not a customer relationship, yet it is frequently cited alongside customer wins, inflating apparent contract value. NVIDIA's position is further complicated: the $3.4 billion AI Cloud contract (claims 11076, 9287) constitutes a genuine customer agreement for infrastructure services, whereas the "5GW strategic partnership" (claim 11077) and $2.1 billion investment (claim 9179) are distinct instruments—the investment's vesting contingent upon actual deployment (claim 9179), rendering its revenue recognition timing uncertain. Concentration is acute: Microsoft and NVIDIA account for $13.1 billion in five-year commitments, translating to approximately $1.9 billion and $680 million in average annual revenue respectively (claim 8996 confirms $1.9bn for Microsoft). Against this, IREN's ARR disclosure of $3.1 billion as of May 2026 (claim 9183) implies only $500 million-plus from remaining sources—a gap allegedly filled by subsequent July announcements of $2.8 billion in multi-year contracts (claim 360) with named AI developers (Perplexity, Figure AI, Together AI, Fluidstack, Fireworks AI, Fal AI, Hume AI). However, claim 8101 references "a new leading AI developer" without identification, suggesting incomplete disclosure. The customer roster underwent complete turnover between 2024 and 2026, replacing niche players with hyperscalers. Revenue arithmetic reconciles approximately when treating Microsoft and NVIDIA as the core base, contingent upon successful deployment—the Horizon 1 milestone (claims 35, 8673) confirms Microsoft execution commenced August 2026. The material risk remains counterparty concentration: two parties underpin 85% of ARR, with Dell as a critical but vulnerable intermediary whose own capacity and supply chain determine execution.

IREN's customer announcements reveal a progression from early-stage partnerships to signed material contracts, with critical distinctions obscured by terminology. Claims 6374 and 6285 document Bitmain and Poolside AI partnerships from early 2024, neither recurring in later disclosures, suggesting immateriality or discontinuation. The company's customer concentration shifted sharply in late 2025 toward named hyperscalers.

Two categories of signed agreements emerge clearly. Claims 4157 and 8112 document IREN's definitive $9.7 billion five-year cloud services contract with Microsoft, executed November 2025, with delivery and acceptance of Horizon 1 confirmed in claim 35, establishing this as fully operational. Revenue guidance of $1.9 billion annualized (claim 8996) approximates the contract's economic contribution. The May 2026 NVIDIA $3.4 billion five-year AI Cloud contract (claims 11076, 9180) is similarly explicit as a signed customer agreement.

However, NVIDIA's "5GW strategic partnership" language in claims 11077, 9289, and 8642 employs framework terminology without revenue quantification, distinct from the signed contract. Claim 9179 separately discloses a $2.1 billion NVIDIA investment vesting upon deployment—a capital commitment distinct from service revenue. This conflation obscures whether NVIDIA is one $3.4 billion customer or two distinct economic relationships.

Dell's characterization contains material variances. Claim 4161 states a $5.8 billion GPU purchase agreement; claim 1297 restates this as $1.6 billion, a 73 percent reduction. As IREN is the buyer (not seller) in this arrangement, it represents capital expenditure, not customer revenue. The July 2026 announcements (claims 360, 8117) name Figure AI, Fluidstack, Fireworks AI, and Hume AI in connection with $2.8 billion in multi-year contracts, narrowing concentration and establishing a visible secondary customer tier. Yet "multi-year" avoids specifying terms, and this cohort remains less formalized than Microsoft or the initial NVIDIA contract.

Concentration has evolved sharply: Microsoft and NVIDIA contracts total $13.1 billion across two counterparties; the July 2026 announcement adds four named entities. The claimed ARR under contract of $3.1 billion (claim 9183, May 2026) presents reconciliation challenges. Annualized against five-year agreements valued at approximately $15.9 billion, the gap suggests deferred recognition, partial contract utilization, or that announced commitments are not yet revenue-productive. Named counterparties have moved from undifferentiated "leading" language to specific identities, yet the distinction between signed contracts and strategic frameworks remains inconsistently disclosed.

IREN has disclosed three major signed customer contracts with identified counterparties: Microsoft Corporation for $9.7 billion over five years (announced November 2025, claim 8112), NVIDIA for $3.4 billion over five years (announced May 2026, claim 11076), and a $2.8 billion multiyear commitment from multiple AI developers announced July 2026 (claim 360). Each carries contractual language with delivery milestones—Microsoft's Horizon 1 facility was delivered and accepted in August 2026 (claim 8674), with three additional 50MW deployments scheduled throughout 2026 (claim 8622). These represent genuine signed agreements rather than MOUs or framework partnerships.

However, the announcement record reveals significant transparency gaps. Perplexity and Together AI appear in IREN's customer roster (claim 8101) yet lack any associated contract values or terms, leaving their status ambiguous between pilots and revenue commitments. Figure AI, Fluidstack, Fireworks AI, Fal AI, and Hume AI are aggregated into the $2.8 billion announcement without individual contract sizes disclosed (claim 8117). Early relationships with Bitmain (January 2024, claim 6374) and Poolside AI (February 2024, claim 6285) generated no quantified revenue and appear to have remained supply or pilot arrangements.

Counterparty concentration has compressed dramatically toward hyperscalers. Microsoft and NVIDIA together represent $13.1 billion of disclosed contract value—approximately 80 percent of announced commitments. NVIDIA's broader 5GW strategic partnership (claim 11077) and $2.1 billion investment (claim 9179) employ framework language (vests as deployed) more typical of LOI arrangements than locked revenue.

The Dell GPU supply component introduces reconciliation issues. November 2025 filings cited $5.8 billion (claim 4161); May 2026 documents reduced this to $1.6 billion (claim 1297)—a 72 percent downward revision suggesting either prior overstatement or contingency clauses. IREN disclosed $3.1 billion in ARR under contract by May 10, 2026 (claim 9183), before the July $2.8 billion was added. Applying the stated 4-year weighted average contract term (claim 8108) yields annualized revenue of roughly $4.2 billion, materially below the headline five-year contract totals, indicating either uneven deployment schedules or weighted-average terms shorter than top-line figures suggest.

IREN's customer announcement trajectory reveals a clear evolution from vaguely specified partnerships to named, executed contracts with major hyperscalers, followed by diversified customer tiers. Claims 6374 and 6285 from early 2024 reference specific counterparties (Bitmain, Poolside AI) but provide minimal financial substance; both use language—"agreement" and "commencement"—consistent with executed deals rather than frameworks. The disclosure landscape shifts decisively in November 2025. Claims 8112 and 4157 introduce the first material, quantified anchor customer: a $9.7 billion five-year Microsoft GPU services contract, signed and named.

May 2026 materializes a second blue-chip customer around NVIDIA. Claims 11076 and 9180 confirm a $3.4 billion five-year AI Cloud contract with specified counterparty. However, claims 11077 and 9289 layer in a parallel "5GW strategic partnership" using language more consistent with pipeline framework than discrete revenue contract. Claim 9179's $2.1 billion NVIDIA investment further blurs customer-from-partner boundaries; vesting hinges on actual GPU deployment rather than contractual certainty.

July 2026's announcement (claims 360, 8641) introduces IREN's first diversified customer roster, naming seven specific AI developers—Perplexity, Figure AI, Together AI, Fluidstack, Fireworks AI, Fal AI, Hume AI—aggregated into $2.8 billion of multi-year commitments. Claim 8101 includes one unnamed "leading AI developer," indicating at least one material counterparty remains undisclosed. This cohort likely comprises shorter-term, smaller-denominator contracts relative to the hyperscaler tier.

Counterparty concentration has been acute. Microsoft and NVIDIA alone represent $13.1 billion of disclosed contract value, roughly 82% of the combined $15.9 billion in announced commitments. The July AI developer tranche modestly dilutes this concentration, though preponderance remains with two counterparties.

Revenue arithmetic appears reconcilable. Claim 9183 cites $3.1 billion ARR under contract as of May 2026. Dividing disclosed totals by stated contract terms—$9.7 billion Microsoft over five years ($1.94 billion annually), $3.4 billion NVIDIA over five years ($0.68 billion annually), $2.8 billion developer contracts at roughly four-year weighted average per claim 8108 ($0.7 billion annually)—yields approximately $3.32 billion, consistent with the disclosed $3.1 billion when adjusting for deployment stagger. The Dell GPU supply agreement (claims 4161, 1297) notably scaled from $5.8 billion to $1.6 billion between November 2025 and May 2026, while claim 8595 discloses customer prepayments approximating 45% of GPU capital expenditure, providing working capital insulation against infrastructure intensity.

IREN's announced customer commitments fall into two distinct categories: confirmed customer contracts with identified counterparties and strategic partnerships characterized by non-binding language. The company has disclosed three major signed customer agreements. Microsoft signed a five-year, $9.7 billion cloud services contract (claims 8112, 4157), with delivery and acceptance of the first data center ("Horizon 1") confirmed in August 2026 (claim 8673). The revenue arithmetic reconciles: IREN disclosed expected annual revenue of $1.9 billion under the Microsoft contract (claim 8996), consistent with the stated contract size (claims 8627, 8674, 8576). NVIDIA signed a distinct five-year, $3.4 billion AI Cloud contract in May 2026 (claims 11103, 9180), generating approximately $680 million in annual revenue. However, IREN simultaneously announced a "broader 5GW strategic partnership" with NVIDIA (claims 11077, 8642), indicating a relationship that extends beyond the defined customer contract and includes a $2.1 billion strategic investment (claim 9179)—capital deployment rather than customer revenue. The July 2026 announcement disclosed $2.8 billion in multi-year contracts (claims 360, 8641) with named AI developers including Figure AI, Fluidstack, Fireworks AI, and Hume AI (claims 8101, 8117), representing IREN's initial diversification beyond hyperscaler customers.

IREN's counterparty concentration has shifted materially. Early-stage partnerships with Bitmain Technologies (claims 6374, 6152) and Poolside AI SAS (claim 6285) announced in early 2024 have vanished from the August 2026 customer roster (claim 363), replaced by Microsoft, NVIDIA, and multiple AI-developer customers. These early partners either did not scale to named-contract status or were displaced. The August 2026 customer list names nine counterparties, but only three—Microsoft, NVIDIA, and an aggregated July cohort—have disclosed contract economics. Undisclosed customers remain identified only by category ("leading AI developer," claim 8101).

A disclosure gap exists regarding Dell's role. Dell's $5.8 billion GPU supply agreement (claims 4161, 8577) and the subsequently announced $1.6 billion purchase commitment (claim 1297) are supply contracts, not customer revenue. IREN also disclosed that recent customer agreements included prepayments covering roughly 45% of associated GPU capital expenditure (claim 8595), improving working capital and reducing execution risk on hardware procurement. Revenue under contract reached $3.1 billion ARR in May 2026 (claim 9183), suggesting material revenue contributions from undisclosed or incompletely disclosed customer relationships beyond the named deals.

IREN has disclosed substantive signed customer contracts with four named counterparties commanding material revenue commitments, alongside a diversifying roster of smaller AI developers. Microsoft and NVIDIA represent the core concentration: Microsoft committed to a five-year, $9.7 billion cloud services agreement announced November 2025 (claims 8112, 8674), with Horizon 1 delivery accepted as of August 2026 (claims 8673, 8631), establishing this as a binding, performing contract rather than framework language. NVIDIA signed a five-year, $3.4 billion AI Cloud contract (claims 11076, 9180, 8640), with infrastructure deployment targeted from early 2027 within Childress data centers. Dell engaged as a supply partner under purchase agreement for approximately $5.8 billion in GPUs (claim 4161), later amended to $1.6 billion (claim 1297), with scheduled delivery beginning March 2026—framed as CAPEX supplier rather than end-customer revenue. Earlier partnerships with Bitmain (claims 6374, 6152) and Poolside AI (claim 6285) lacked quantified commitments, appearing more exploratory. Concentration has shifted notably. Through mid-2024, IREN's named partnerships remained nascent and modest in scale. By November 2025, the Microsoft and Dell agreements crystallized as binding, large-denominated commitments. The May 2026 NVIDIA contract, combined with NVIDIA's $2.1 billion minority investment (claim 9179), signaled strategic depth beyond pure customer relationship. July 2026 represented a deliberate diversification into nine named AI developers plus an undisclosed counterparty within a $2.8 billion multiyear pool (claims 360, 8117, 8101), though individual deal sizes within this cohort remain unspecified. This shift reduces single-customer concentration risk but obscures revenue visibility. Revenue arithmetic shows claimed ARR of $3.1 billion as of May 2026 (claim 9183). Microsoft's $9.7 billion over five years yields approximately $1.94 billion annually; NVIDIA's $3.4 billion yields $0.68 billion annually—totaling $2.62 billion from these two anchors alone. Early partnerships and the July $2.8 billion pool would account for the remainder, though specific revenue recognition timing and ramp profiles remain opaque. The August 2026 disclosure of a weighted average contract term of 4 years (claim 8108) provides limited additional clarity without itemized customer contribution.

IREN's announced customer portfolio reveals a critical distinction between fully executed agreements and strategic partnerships, with material revenue discrepancies requiring clarification. The company's largest contracts are formally signed: Microsoft's five-year, $9.7 billion cloud services agreement (claim 8112, November 2025) and NVIDIA's concurrent $3.4 billion AI Cloud contract (claim 9287, May 2026) represent binding purchase commitments. Dell's GPU supply agreement presents a notable inconsistency—initially disclosed as $5.8 billion (claim 4161, November 2025) but subsequently revised to $1.6 billion (claim 1297, May 2026), suggesting either material contract amendment or potential misstatement in the original announcement.

In contrast, IREN's NVIDIA relationship conflates contract and strategic investment components. The $2.1 billion NVIDIA investment (claim 9179) is explicitly performance-gated, vesting only upon actual GPU infrastructure deployment—this represents contingent capital, not revenue. The 5GW "strategic partnership" framing (claim 9289, claim 362) lacks binding purchase commitments and functions as a framework rather than a signed customer agreement.

Earlier counterparties lack precision. Bitmain (claims 6374, 6152) receives only vague language about "amended and new agreements" without quantified value. Poolside AI (claim 6285) merits a single mention of "commencement of GPU cloud service" with no contractual detail. The July 2026 announcement (claim 360) names six counterparties—Perplexity, Figure AI, Together AI, Fluidstack, Fireworks AI, Fal AI, Hume AI—collectively representing $2.8 billion, but individual contract terms remain undisclosed.

Concentration has shifted dramatically. Early 2024 featured distributed partnerships; by November 2025, Microsoft and Dell represented over $15 billion in announced commitments. The May 2026 peak ARR guidance of $3.1 billion (claim 9183) reconciles approximately with Microsoft's ~$1.9 billion annual run rate (claim 8996) plus NVIDIA's proportional contribution, though the subsequent $2.8 billion announcement (claim 360) raises timing questions regarding recognition relative to May guidance. The unnamed "leading AI developer" referenced in claim 8101 adds opacity regarding customer concentration risk. IREN's customer roster has shifted from unidentified strategic partners to named hyperscalers and AI firms, yet several announced agreements lack binding specificity or remain performance-contingent.

IREN's customer announcements from 2024 through August 2026 predominantly feature signed, binding contracts rather than non-binding frameworks. The Microsoft ($9.7 billion over five years, claim 8112) and NVIDIA ($3.4 billion over five years, claim 9287) engagements are consistently described using binding language: "entered into an agreement" and "signed," not exploratory or pilot constructions. Early counterparties including Bitmain (miner supply partnership, claim 6374) and Poolside AI GPU services (claim 6285) similarly reference executed agreements from 2024. The July 2026 announcement of $2.8 billion in "multi-year cloud services contracts" (claim 360) initially used descriptive language—"leading AI developers"—but subsequent disclosures identified specific counterparties: Figure AI, Fluidstack, Fireworks AI, Fal AI, and Hume AI (claim 8117). The broader customer roster as of August 2026 (claim 363) names Microsoft, NVIDIA, Perplexity, and the aforementioned developers without hedging; no "strategic partner" designations obscure actual customer identity.

Counterparty concentration has intensified markedly. In early 2024, IREN announced mining and GPU cloud partnerships with Bitmain and Poolside AI. By late 2025, Microsoft and Dell emerged as dominant counterparties: Microsoft's $9.7 billion contract and Dell's GPU supply agreement (initially $5.8 billion, claim 4161) represented the overwhelming disclosed value. By mid-2026, NVIDIA's $3.4 billion AI cloud contract (claim 9287) plus a $2.1 billion strategic investment (claim 9179) and a $2.8 billion tranche of smaller AI developer contracts positioned Microsoft and NVIDIA as anchors, together representing roughly 82 percent of named contract value. Dell remains a supplier (GPU procurement), not a revenue customer.

The disclosed revenue arithmetic aligns with stated metrics. IREN reported $3.1 billion in annual recurring revenue under contract as of May 2026 (claim 9183). Microsoft's $9.7 billion five-year commitment yields approximately $1.9 billion annually (claim 8996, explicitly stated). NVIDIA's $3.4 billion contract scales to roughly $680 million per year. The smaller developer contracts totaling $2.8 billion, weighted at approximately four-year average terms (claim 8108), contribute approximately $700 million annually. These components sum to approximately $3.28 billion, consistent with the $3.1 billion ARR disclosure and accounting for ramp timing. No material discrepancy appears between announced deal value and revenue guidance.

IREN's announced partnership portfolio shifted decisively from early-stage named partnerships to hyperscaler concentration beginning November 2025. The 2024 announcements identified specific counterparties—Bitmain Technologies and Poolside AI—with explicit 'agreement' or 'commencement' language, establishing binding commitments. By contrast, the bulk of disclosed value arrived via three major contractual relationships: a $9.7 billion five-year Microsoft GPU services agreement (claim 4157, signed November 2, 2025), a $3.4 billion NVIDIA AI Cloud contract (claim 9180, signed May 2026), and a $2.8 billion multi-year arrangement announced July 20, 2026. The Microsoft and NVIDIA agreements use precise contract language with specified delivery tranches and timelines. However, NVIDIA also carries a parallel '5GW strategic partnership' designation (claims 9289, 8642) that, while attributed to NVIDIA, describes a broader framework rather than a fixed-dollar commitment, creating some ambiguity between contractual commitment and strategic intent. The July $2.8 billion announcement initially employed generic language—'leading AI developers'—before claim 8101 disclosed named counterparties including Figure AI, Fluidstack, Fireworks AI, Fal AI, and Hume AI, though one participant remained unidentified as 'a new leading AI developer.' Counterparty concentration is extreme: Microsoft and NVIDIA represent $13.1 billion of the disclosed $15.5 billion contract value through mid-2026. Dell appears as a GPU supplier rather than an end-customer, complicating the customer narrative. Revenue reconciliation is largely consistent: the Microsoft contract yields $1.9 billion average annual revenue (claim 8996), reconciling against the $9.7 billion five-year term; the stated $3.1 billion ARR under contract (claim 9183) aligns approximately with pro-rata annual run-rates given the contracts' weighted average four-year term (claim 8108). Notably, execution has begun—Microsoft accepted Horizon 1 delivery in August 2026 (claim 8673)—yet Horizons 2-4 remain scheduled for future delivery. The portfolio thus comprises binding, named, dollar-specific contracts with Microsoft and NVIDIA, a strategic partnership with aspirational scope around NVIDIA's 5GW pipeline, and a cohort of smaller AI developers announced with delayed naming transparency.

IREN's disclosure record reveals a material distinction between its foundational hyperscaler contracts and its broader AI developer base. The two anchor agreements are both signed and substantially executed: Microsoft's five-year, $9.7 billion cloud-services contract (claim 8112) with delivered infrastructure—Horizon 1 accepted August 2026 (claims 8673, 8631)—and NVIDIA's $3.4 billion five-year AI cloud contract signed May 2026 (claim 9287). Both counterparties are explicitly named and quantified. However, NVIDIA's disclosures conflate two distinct relationships. A separate "strategic partnership" covering IREN's 5GW global data center pipeline (claim 9289) sits alongside the customer contract, a framing that obscures whether the partnership carries binding commercial commitment or represents a framework pending deployment-linked vesting mechanics (claim 9179 discloses vesting tied to GPU deployment across IREN campuses).

Dell's agreements similarly require unpacking. The company disclosed a $5.8 billion GPU supply agreement in November 2025 (claim 4161), later amended to $1.6 billion in May 2026 (claim 1297, 8992), suggesting partial scope, timing changes, or a distinct second tranche. The vendor is named, but the commercial footing—purchase order versus long-term supply framework—remains imprecise.

IREN's counterparty concentration has shifted sharply. Early announcements featured Bitmain (claims 6374, 6152) and Poolside AI (claim 6285) without revenue disclosure. By mid-2025, Microsoft represented near-total concentration at $9.7 billion. The July 2026 announcement diversified the base, adding eight named AI developers—Figure AI, Together AI, Fluidstack, Fireworks AI, Fal AI, Hume AI, Perplexity, and one undisclosed "leading AI developer" (claims 8101, 360)—totaling $2.8 billion (claim 8117), though individual contract sizes remain unnamed.

Revenue arithmetic exhibits internal strain. IREN disclosed $3.1 billion ARR under contract in May 2026 (claim 9183). Microsoft's $9.7 billion over five years yields $1.94 billion annualized (confirmed by claim 8996: $1.9 billion expected annual revenue). NVIDIA's $3.4 billion over five years yields approximately $680 million annual. The $2.8 billion July pool distributed across eight counterparties (claim 8101) yields less than $350 million per customer on average. The disclosed $3.1 billion ARR cannot reconcile with these named contracts without either undisclosed customers, multi-year front-loading assumptions, or measurement methodology opacity around "under contract" versus cash revenue recognition.

IREN's customer portfolio contains a hierarchy between formally signed service agreements and strategic partnerships with contingent economics. Three named counterparties have executed binding service contracts: Microsoft (five-year, $9.7 billion cloud services contract, claims 4157, 8674, with delivery completed on Horizon 1, claim 8631), NVIDIA (five-year, $3.4 billion AI Cloud contract, claims 9287, 9180), and Dell (approximately $5.8 billion GPU supply agreement, claim 4161; supplemented by a $1.6 billion air-cooled Blackwell agreement in May, claim 1297). Earlier arrangements with Bitmain (claim 6374) and Poolside AI (claim 6285) employed "agreement" language without specifying contract value or term structure. NVIDIA's relationship includes a separately described "5GW strategic partnership" (claims 9289, 8642) where investment rights vest conditionally on infrastructure deployment—a structure materially different from binding purchase orders.

A material transparency gap separates named from contractually quantified counterparties. IREN identifies Perplexity, Figure AI, Together AI, Fluidstack, Fireworks AI, Fal AI, and Hume AI as customers (claims 8101, 363), yet only four receive attribution within a $2.8 billion collective figure for "multi-year cloud services contracts" (claim 8117). Perplexity and Together AI lack any disclosed contract values. This incomplete disclosure creates ambiguity about pipeline maturity and counterparty risk concentration.

Counterparty concentration has shifted from early-stage diversification to hyperscaler dominance. The 2024 portfolio (Bitmain, Poolside AI) was infrastructure-focused; by November 2025, IREN secured its two largest contracts (Microsoft, Dell) sequentially. By mid-2026, concentration was acute: Microsoft represents 59% of disclosed annualized revenue ($1.9 billion, claim 8996 specifying "expected $1.9bn average annual revenue under the Microsoft contract"), NVIDIA supplies roughly 22% ($0.68 billion on a five-year basis), and named AI developers contribute roughly 9% across $2.8 billion with unspecified term assumptions. Microsoft's weighting creates singular customer risk.

Revenue arithmetic exhibits unresolved discrepancies. IREN reported $3.1 billion in "ARR under contract" as of May 10, 2026 (claim 9183), but disclosed contracts (Microsoft $1.9B, NVIDIA $0.68B) sum to only $2.58 billion, leaving a $0.52 billion gap unexplained by public disclosure. This delta, predating the July $2.8 billion AI developer announcement, suggests either undisclosed customer wins or differing contract term assumptions materially affecting revenue models. Additionally, the company disclosed customer prepayments equal to roughly 45% of associated GPU capital expenditure (claim 8595), a structurally important cash-flow feature, yet IREN provided no granular breakdown by customer.

IREN's customer announcements distinguish sharply between binding revenue contracts and strategic partnerships. Three major signed cloud services agreements anchor the disclosed pipeline: Microsoft's November 2025 five-year contract for $9.7 billion (claims 8674, 8627, 8590), NVIDIA's May 2026 five-year $3.4 billion AI Cloud contract (claims 8277, 9180, 8640), and a July 2026 $2.8 billion multi-year cohort (claim 360). These are structured as definitive purchase agreements with specified terms, payment schedules, and—in Microsoft's case—physical delivery milestones tracked via Horizon facility deployments (claims 8631, 7454, 35). Earlier announcements of Bitmain (claim 6374) and Poolside AI (claim 6285) use similar signed-contract language.

Language diverges in other cases. NVIDIA's "strategic partnership" covering IREN's global 5GW data center pipeline (claims 9289, 8642, 362) emphasizes collaboration and architectural alignment rather than committed revenue. The Mirantis references (claims 8768, 8769) describe Mirantis' own 1,500 enterprise customers and its status as an NVIDIA AI Cloud Ready Initiative partner—not IREN customer commitments. The unnamed "leading AI developers" in claim 360 gained clarity when subsequently identified as Perplexity, Figure AI, Together AI, Fluidstack, Fireworks AI, Fal AI, and Hume AI (claims 8101, 363), though disaggregated contract sizes remain undisclosed.

Counterparty concentration has intensified since late 2025. Early 2024 announcements involving Bitmain and Poolside AI cede dominance to hyperscaler and AI platform players by mid-2026. Microsoft and NVIDIA alone represent approximately 78% of disclosed ARR under standard five-year amortization: $1.9 billion annually from Microsoft (claim 8996) plus roughly $0.68 billion from NVIDIA ($3.4B ÷ 5 years). The July 2026 $2.8 billion cohort appears designed to offset this concentration, though individual counterparty sizes within that group remain opaque.

Revenue arithmetic largely reconciles. The May 10 claim of $3.1 billion ARR under contract (claim 9183) aligns with Microsoft's $1.9 billion annual component plus NVIDIA's $0.68 billion plus smaller contributors. The July announcement of $2.8 billion "further boosting" ARR (claim 8117) suggests cumulative announced contracts exceed $5 billion in annual run-rate by August 2026. Dell agreements totaling $7.4 billion (claims 4161, 1297) represent capital purchases supporting customer fulfillment, not IREN's own revenue contracts.

IREN's disclosed customer base separates cleanly into signed service agreements and strategic arrangements, with meaningful concentration risk obscured by partnership nomenclature. Three named customers account for the overwhelming majority of announced contract value: Microsoft ($9.7 billion, five-year cloud services agreement signed November 2025, claim 8112), NVIDIA ($3.4 billion, five-year AI Cloud contract signed May 2026, claim 9180), and a partially-disclosed cohort of AI developers ($2.8 billion announced July 2026, claim 360). The Microsoft engagement is substantiated by physical delivery—Horizon 1 accepted in August 2026 (claim 8631)—and is corroborated by expected annual revenue of $1.9 billion (claim 8996). NVIDIA's contract includes a conditional $2.1 billion equity investment that vests only upon infrastructure deployment across IREN campuses (claim 9179), structurally distinct from service revenue. The July 2026 AI developer cohort is partially named: Figure AI, Fluidstack, Fireworks AI, and Hume AI are identified in claim 8117, but the full composition of the $2.8 billion remains undisclosed, creating opacity around customer concentration. Dell appears across three claims (4161, 8992, 1297) as a GPU supply vendor, not an end customer—a critical distinction often conflated in investor communications. Earlier partnerships with Bitmain Technologies (claims 6374, 6152) and Poolside AI (claim 6285) from 2024 have been commercially superseded and are no longer mentioned in current customer rosters. The disclosed ARR of $3.1 billion (claim 9183) reconciles approximately with announced contracts: Microsoft's $1.9 billion annual run rate plus NVIDIA's $680 million annualized ($3.4B ÷ 5) plus the AI developer cohort at roughly $560 million annualized yields approximately $3.14 billion, validating revenue arithmetic. Concentration has tightened markedly. Microsoft and NVIDIA together represent 84 percent of named contract value, with Microsoft alone representing 62 percent. The customer base statement (claim 8101) mentions ten entities, yet only three—Microsoft, NVIDIA, and the four partially-named AI developers—have substantiated, publicly-valued contracts. Perplexity, Together AI, and Fal AI are listed as customers without disclosed deal values, creating asymmetric visibility. The strategic partnership language surrounding NVIDIA (claims 8642, 9289) describes a 5GW data center pipeline collaboration rather than a bounded service contract, introducing ambiguity about expansion optionality versus commitment certainty.

IREN's customer and partnership disclosure matrix shows clear distinction between signed customer service contracts and ancillary supply arrangements, though nomenclature occasionally obscures which is which. The early 2024 partnerships with Bitmain Technologies (claims 6374, 6152) and Poolside AI (claim 6285) carry explicit "agreement" language but disclose no associated revenue figures, making their materiality difficult to assess. By contrast, the November 2025 Microsoft contract is unambiguously a signed, five-year $9.7 billion cloud services agreement (claims 8112, 4157, 1266) with explicit delivery milestones verified in August 2026 (claims 35, 8673). The May 2026 NVIDIA AI Cloud contract is similarly definitive at $3.4 billion over five years (claims 9180, 1299). The July 2026 announcement of $2.8 billion in multi-year contracts names four counterparties—Figure AI, Fluidstack, Fireworks AI, Hume AI (claims 8117, 8101)—yet claim 360 describes them generically as "leading AI developers," and claim 8101 references "a new leading AI developer" that remains unidentified.

Counterparty concentration has shifted markedly. Early 2024 relied on two named partners without disclosed valuations; by mid-2026, concentration consolidated around Microsoft ($9.7B) and NVIDIA ($3.4B), representing $13.1 billion of announced contractual value. The July addition of $2.8 billion across a widening roster—Perplexity, Together AI, and others—suggests portfolio diversification, though many relationships lack individual deal transparency and remain embedded in aggregate announcements.

Revenue arithmetic requires carefully parsing supply agreements from service contracts. The Dell purchase agreements (claims 4161, 1297) totaling approximately $7.4 billion ($5.8B November 2025, $1.6B May 2026) represent customer GPU prepayments, not IREN service revenue. The May 2026 annualized revenue expectation of $1.9 billion under the Microsoft contract (claim 8996) aligns with a five-year $9.7 billion deal, yet the disclosed ARR under contract of $3.1 billion (claim 9183) does not clearly reconcile against the $13.1 billion five-year aggregate when weighted by customer concentration. The February 2026 expanded ARR disclosure suggests revenue acceleration beyond the core Microsoft-NVIDIA relationships, but the unnamed counterparty and aggregate treatment of the July contracts limit visibility into which specific customers drive incremental value.

IREN's customer announcements from 2024 through August 2026 generally distinguish signed agreements from softer partnership frameworks, though disclosure precision varies across counterparties. Early partnerships with Bitmain (claims 6374, 6152) and Poolside AI (claim 6285) employed explicit "agreement" language. Material revenue acceleration arrived in late 2025: Microsoft's five-year, $9.7 billion cloud services contract (claims 8112, 4157) and Dell's $5.8 billion GPU supply agreement (claims 4161, 8577) both used purchase-agreement terminology, substantiated by Horizon 1 delivery and acceptance in August 2026 (claims 35, 8673). In May 2026, IREN disclosed a signed $3.4 billion five-year AI Cloud contract with NVIDIA (claims 9180, 8277), though separately announced a $2.1 billion NVIDIA investment with vesting tied to deployment milestones (claim 9179)—equity participation, not service revenue. A "broader strategic partnership" with NVIDIA covering 5 gigawatts of data center capacity (claims 362, 8642) uses non-contractual language obscuring binding commitments.

Counterparty concentration evolved sharply. Bitmain dominated 2024 but vanished after May. Microsoft became the largest single counterparty at announcement in November 2025. The May NVIDIA contract ($3.4 billion) and July $2.8 billion tranche with Figure AI, Fluidstack, Fireworks AI, and Hume AI (claims 8117, 8641) diversified the customer base materially. However, claim 8101 references one unnamed "new leading AI developer" contributing to the July volume, leaving that customer opaque. Perplexity and Together AI also appear in the customer roster (claims 8101, 363) without corresponding deal disclosures.

Disclosed revenue largely reconciles. IREN claimed $3.1 billion annualized recurring revenue as of May 2026 (claim 9183). Microsoft's stated $1.9 billion average annual revenue (claim 8996) matches $9.7 billion over five years. The NVIDIA contract yields roughly $680 million annually. The July tranche ($2.8 billion) divided by the disclosed four-year weighted average contract term (claim 8108) contributes approximately $700 million, totaling roughly $3.3 billion—consistent with the $3.1 billion ARR. An ambiguity persists regarding Dell. Two distinct agreements appear: $5.8 billion from November 2025 (claims 4161, 8577) and $1.6 billion from May 2026 (claims 1297, 8992). No disclosure clarifies whether these are tranches, revisions, or separate lines, leaving the consolidated Dell commitment ambiguous.

IREN's customer disclosures divide clearly into signed contracts and aspirational language, with material concentration shifts across the company's history. From 2024-2025, the company announced modest early partnerships—Bitmain Technologies (claims 6374, 6152) and Poolside AI (claim 6285)—characterized by straightforward "agreement" or "commencement" language, implying binding execution. By November 2025, IREN pivoted to hyperscaler-dominated deals: Microsoft's $9.7 billion five-year contract (claims 8112, 4157, 8674) and a companion Dell GPU supply agreement for $5.8 billion (claims 4161, 1297), the latter secured by parent-company guarantee (claim 4164). These bore explicit pricing and delivery schedules (tranches through 2026+).

In May-August 2026, IREN announced three additional large contracts: a $3.4 billion managed services deal (claims 1299, 8995), a $3.4 billion NVIDIA five-year agreement (claims 8277, 351), and $2.8 billion across multiple named AI developers (claim 8117—Figure AI, Fluidstack, Fireworks AI, Hume AI, et al.). The language shifts subtly across announcements. The $3.4 billion managed services contract appears labeled "previously announced" (claim 8995), suggesting prior disclosure in forms not captured here. The NVIDIA deal begins as a "strategic partnership" covering a "5 GW secured data center pipeline" (claim 8642, 362)—framework language—before crystallizing into a signed, priced $3.4 billion contract (claims 8277, 351).

Counterparty concentration shows dramatic evolution. Early 2024 reflects fragmentation across emerging partners. The November 2025 announcements create extreme concentration: Microsoft dominates as a single customer, with Dell as the sole supply counterparty. By mid-2026, the roster diversifies materially—the addition of NVIDIA plus eight named AI developers (claim 8101) reduces Microsoft's relative weight, though it remains the largest contract vehicle. One unnamed customer appears in claim 8101 ("a new leading AI developer"), and the third $3.4 billion contract's counterparty identity remains unspecified.

Revenue arithmetic largely reconciles. Microsoft's disclosed $1.9 billion annual revenue (claim 8996) aligns with its $9.7 billion five-year contract divided by five ($1.94 billion annually). Total announced customer contract value—Microsoft ($9.7B), NVIDIA ($3.4B), managed services ($3.4B), and AI developers ($2.8B)—sums to $19.3 billion, implying ~$3.86 billion annualized run-rate against the disclosed weighted-average contract term of ~4 years (claim 8108). Dell's $5.8 billion represents supply costs, properly classified as cost-of-revenue, not customer contracts. Claim 8595 discloses customer prepayments at ~45% of associated GPU capex, indicating upfront cash collection offsetting capital intensity.

IREN's customer announcements distinguish sharply between named signed agreements and generic partnership language. The company has disclosed four clearly executed contracts: Microsoft ($9.7 billion over five years, signed November 2025; claims 8112, 1266), Dell ($5.8 billion GPU supply, signed November 2025; claim 4161), NVIDIA ($3.4 billion over five years, signed August 2026; claim 8277), and a cohort of AI developers ($2.8 billion in multi-year contracts, announced July 2026; claim 8641). Horizon 1 delivery to Microsoft in August 2026 confirms contractual execution and payment (claims 8673, 8631). The Dell partnership includes a parent company guarantee of obligations, evidencing binding commitment (claim 4164). By contrast, early partnerships with Bitmain Technologies (2024) and Poolside AI (2024) used identical "entered into an agreement" language but received minimal subsequent disclosure, suggesting smaller or concluded arrangements. The July 2026 AI developer cohort includes partially named customers—Figure AI, Fluidstack, Fireworks AI, Hume AI, and Together AI are identified (claim 8117)—but the announcement also references generic language like "leading AI developers" and "a new leading AI developer" (claim 8101), masking the identity and contract terms of additional counterparties. Counterparty concentration has evolved dramatically from 2024's distributed supplier model (Bitmain, Poolside AI) to 2025–2026's hyperscaler dominance. Microsoft alone represents approximately 61% of disclosed contract value. The addition of NVIDIA as a $3.4 billion customer in August 2026 signals diversification, but the company remains exposure-concentrated among three counterparties (Microsoft, NVIDIA, Dell). Revenue arithmetic presents inconsistencies requiring clarification. Dell's GPU supply contract is disclosed as $5.8 billion in November 2025 announcements (claim 4161) but appears revised to $1.6 billion by May 2026 (claim 1297). Claim 8595 notes recent agreements included customer prepayments of roughly 45% of GPU capital expenditure, suggesting the May figure may represent prepayment obligations rather than total purchase price. Additionally, the $3.4 billion "managed services AI cloud contract" (claims 1299, 8640) is later identified as the NVIDIA agreement itself (claim 8277), not a separate customer. Taken together, IREN has signed approximately $15.9 billion in identified, multi-year capacity commitments across named customers, with Microsoft and infrastructure partnerships driving near-term revenue concentration.

IREN's customer announcements distinguish clearly between named, signed agreements with specific dollar commitments and partnerships of undisclosed material value. The company has disclosed three material signed customer arrangements with dollar figures: a five-year, $9.7 billion cloud services agreement with Microsoft (signed November 2025, Horizon 1 delivered and accepted August 2026, claims 8631, 8312); a five-year, $3.4 billion AI cloud agreement with NVIDIA, accompanied by a "strategic partnership" covering its 5 GW data center pipeline (claims 351, 362); and a multi-year $2.8 billion arrangement with a consortium of seven named AI developers—Perplexity, Figure AI, Together AI, Fluidstack, Fireworks AI, Fal AI, and Hume AI—announced July 2026 (claim 8641). Bitmain Technologies and Poolside AI, announced as partners in early 2024 (claims 6374, 6285), have not appeared in subsequent customer roster communications and carried no disclosed revenue figures. The company references unnamed additional customers contributing to its weighted-average 4-year contract term (claim 8108), suggesting material customers remain undisclosed.

Microsoft and NVIDIA concentration is pronounced, representing $13.1 billion of the $15.9 billion in explicitly quantified customer commitments, or 82 percent. The seven AI developers named in the July aggregate announcement lack individual contract disclosure, obscuring whether Perplexity, Figure AI, or others constitute material standalone counterparties or equivalent contract values. Revenue arithmetic reconciles: IREN disclosed $1.9 billion expected average annual Microsoft revenue (claim 8996), which matches $9.7 billion divided by five years. Assuming the AI developer segment spans the weighted-average 4-year term, annualized customer revenue across disclosed commitments approximates $3.2 billion to $3.5 billion. Contract language is uniformly transactional—"five-year agreement," "delivered and accepted," "purchase agreement"—except for NVIDIA's "strategic partnership" descriptor (claim 362), which appears supplementary to the quantified $3.4 billion term. No pilot, framework, or MOU language appears in material customer claims.

IREN's disclosed customer agreements distinguish sharply between signed, named contracts and strategic positioning language. The three largest deals—Microsoft ($9.7 billion), NVIDIA ($3.4 billion), and Dell hardware supply ($5.8 billion, later revised to $1.6 billion in May 2026)—are fully named and signed as of November 2025 (claims 4157, 8112, 1297). All three carry specific delivery terms; notably, only Microsoft's Horizon 1 has achieved delivery and acceptance as of August 12, 2026 (claim 8673), with Horizons 2–4 still pending in 2026 (claim 8622). The remaining $2.8 billion in July 2026 multiyear contracts (claims 8641, 8275) is attributed to "leading AI developers" generically in some filings (claim 360) but specifically named in others: Figure AI, Fluidstack, Fireworks AI, and Hume AI (claim 8117). Perplexity and Together AI appear in the customer roster (claim 8101) but lack individual contract disclosure. One entry—"a new leading AI developer" (claim 8101)—remains wholly unidentified.

Customer concentration has shifted dramatically. Early 2024 announcements named Bitmain Technologies and Poolside AI (claims 6374, 6285) without disclosed deal sizes; both vanish from August 2026 customer rosters (claim 8101), suggesting either terminated arrangements or immaterial ongoing revenue. By late 2025, concentration spiked around Microsoft, NVIDIA, and Dell. August 2026 disclosures introduce diversification into mid-tier AI developers, though Microsoft and NVIDIA remain anchor relationships. NVIDIA occupies an unusual dual classification: named as both customer (claim 8541) and "strategic partner" (claim 8642), potentially obscuring whether the $3.4 billion represents a customer contract or a framework commitment.

The revenue arithmetic raises questions. Disclosed signed contracts total approximately $15.9 billion (Microsoft $9.7B + NVIDIA $3.4B + AI developers $2.8B), plus Dell's GPU supply obligation. The Dell figure creates ambiguity: a $5.8 billion purchase agreement dated November 2, 2025 (claim 4161) is later superseded by a $1.6 billion agreement (claim 1297, May 2026), suggesting either scope reduction or separation of tranches. Claim 8595 notes customer prepayments covering roughly 45% of GPU capital expenditure, providing revenue timing benefits but not reconciling total deployment capacity. The Golden State Warriors sponsorship ($50 million annually, claim 7895) appears orthogonal to core cloud services revenue but inflates headline deal volume. Critical gap: no disclosure of IREN's total capacity, blended contract economics, or whether announced contracts exhaust buildout plans, making concentration risk and revenue sustainability difficult to assess.

IREN's announced customer portfolio reveals a portfolio heavily weighted toward explicitly signed contracts with major hyperscalers and cloud-native AI developers, though with notable gaps in early partnership disclosure. The company has named Microsoft, NVIDIA, Dell, and a consortium of AI developers—Figure AI, Fluidstack, Fireworks AI, Hume AI, and Perplexity—as signed counterparties under service agreements totaling at least $19.3 billion across three primary tranches: Microsoft's five-year, $9.7 billion contract (claims 8112, 8627), NVIDIA's parallel $3.4 billion engagement (claim 8277), and a $2.8 billion cohort of smaller AI developer contracts announced July 2026 (claim 8641). In contrast, IREN's early partnership narrative—Bitmain Technologies (claim 6374) and Poolside AI (claim 6285), both announced in early 2024—vanishes entirely from customer rosters by mid-2026, suggesting either pilot status or immaterial scale relative to hyperscaler engagements.

The concentration is acute and skewed. Microsoft alone likely represents 50 percent or more of disclosed AI cloud revenue, with Horizon 1's August 2026 delivery and acceptance (claim 8673) representing the first of four planned 50MW deployments under that single contract. NVIDIA occupies an unusual dual role: $3.4 billion customer contract holder (claim 8277) *and* strategic partner under a separate framework covering IREN's 5 GW secured pipeline (claim 8642). Dell functions as capex supplier rather than cloud customer, with $7.4 billion in GPU purchase agreements (claims 4161, 1297) financing the infrastructure underlying other contracts. The July 2026 announcement's $2.8 billion comprises smaller, individually named counterparties, suggesting diversification marketing rather than counterparty balance.

The disclosed revenue arithmetic requires scrutiny. Three distinct AI cloud contract announcements surface: May 2025's unnamed $3.4 billion engagement (claim 1299), July 2026's $2.8 billion consortium (claim 8641), and the continuous Microsoft contract. If these are non-overlapping, total signed cloud services revenue is approximately $19.3 billion, dominated by Microsoft's contract and presumed to flow across the stated weighted average 4-year term (claim 8108). However, claim 8595 notes customer prepayments represent only approximately 45 percent of GPU capex, suggesting revenue recognition trails expenditure. The August 2026 addition of a Golden State Warriors jersey-patch sponsorship ($50 million-plus annually, claim 7895) reflects brand extension rather than core cloud services concentration, and NVIDIA's Exemplar Cloud certification (claim 8543) appears a validation milestone rather than incremental revenue. The absence of Bitmain and Poolside from post-2024 disclosures and the sudden materialization of $19.3 billion in named contracts within a 12-month window warrant investigation into whether early partnerships pilot-tested capacity before hyperscaler deployment.

IREN's announced customer base exhibits a sharp bifurcation between explicitly signed contracts and framework-level partnerships, with material revenue concentration concentrated in three mega-deals. The named counterparties with signed customer agreements are Microsoft ($9.7 billion over five years, claim 8112), NVIDIA ($3.4 billion over five years, claim 8277), a cohort of identified AI developers totaling $2.8 billion multi-year (Figure AI, Fluidstack, Fireworks AI, Hume AI; claim 8117), and legacy partnerships with Bitmain and Poolside AI dating to early 2024. However, NVIDIA's characterization diverges materially: while a five-year $3.4 billion customer contract is named (claim 8277), the company separately disclosed entry into a broader "strategic partnership with NVIDIA covering its global 5 GW secured data center pipeline" (claim 8642)—language suggesting framework-level commitment rather than spot-verified revenue. Dell, prominently cited in six claims with $5.8 billion and $1.6 billion tranches, is a supplier of GPUs and equipment, not a customer, and should not be conflated with revenue-generating agreements. The customer roster has evolved substantially: early 2024 announcements centered on Bitmain and Poolside (single equipment/service relationships); by November 2025, IREN announced the Microsoft and Dell arrangements; July-August 2026 disclosures expanded to named AI developers and certified NVIDIA's Exemplar Cloud status. One material gap exists: claim 8101 identifies "a new leading AI developer" as part of IREN's customer base without naming the entity, suggesting not all customers are disclosed. The announced revenue base totals $15.9 billion across named multi-year contracts (Microsoft $9.7B plus NVIDIA $3.4B plus AI developers $2.8B), yet the AI developer figure lacks specified term length beyond "multi-year," creating arithmetic opacity. Microsoft appears to dominate (~61% of named customer value), followed by NVIDIA (~21%), with distributed AI developer relationships comprising the remainder. The Warriors sponsorship ($50 million annually, claim 7895) is categorically distinct—non-core, marketing-focused infrastructure partnership. Counterparty concentration has shifted from fragmented smaller deals to hyperscaler-dominated revenue, a structurally material evolution warranting scrutiny of customer diversification risk.

IREN's customer and partnership announcements show progression from unnamed pilots in 2024 to explicitly named, multi-billion-dollar contracts with hyperscalers by 2025-2026, though some counterparties remain partially unquantified. Early announcements included Bitmain Technologies (claims 6374, 6152) and Poolside AI (claim 6285) with agreement language but no disclosed monetary terms. Microsoft crystallized as anchor customer with a November 2025 five-year, $9.7 billion contract (claims 8112, 4157), explicitly specifying delivery tranches and confirmed acceptance of Horizon 1 (claim 35, 7454). NVIDIA operates in dual capacity: claims 351, 8277, and 8542 establish it as a five-year, $3.4 billion customer purchasing AI cloud compute services; separately, claim 362 discloses a "broader strategic partnership" covering NVIDIA's 5 GW secured data center pipeline, suggesting framework-level commitments layered atop the customer contract. Dell Technologies functions as supply counterparty, not customer, with $7.4 billion in obligated GPU deliveries (claims 4161, 1297, 8577), backed by IREN's parent company guarantee (claim 4164). By July-August 2026, IREN named four AI developers—Figure AI, Fluidstack, Fireworks AI, Hume AI—as parties to the $2.8 billion multi-year cloud contract (claims 360, 8117). However, claim 8101 lists Perplexity and Together AI as customers without disclosed contract values, and references "a new leading AI developer" without naming. Counterparty concentration accelerated dramatically: Microsoft and NVIDIA represent $13.1 billion of $15.9 billion in announced customer contracts, or 82 percent (claims 8590, 8277). Revenue arithmetic reveals friction between customer contracts and capital deployment. The $15.9 billion customer agreements, amortized over the disclosed four-year weighted average contract term (claim 8108), imply roughly $3.975 billion in annualized revenue. Yet Dell supply obligations ($7.4 billion) are capital expenditure, not customer revenue. Claim 8595 indicates customer prepayments covered approximately 45 percent of associated GPU capital expenditure, suggesting revenue lags capital deployment. This timing mismatch creates potential cash flow pressure, though Microsoft's August 2026 Horizon 1 acceptance (claim 35) validates execution of the largest contract; the portfolio remains critically dependent on two customers representing over 80 percent of disclosed backlog.

IREN's announced customer and partnership portfolio reveals significant concentration among three counterparties spanning roughly $19.7 billion in disclosed five-year commitments, though the arithmetic warrants scrutiny. Microsoft constitutes the largest named customer, formalized through a November 2025 five-year, $9.7 billion cloud services agreement (claims 8112, 4157) for dedicated GPU infrastructure across four 50MW "Horizon" deployments, with Horizon 1 accepted as of August 2026 (claim 35). NVIDIA enters as a second anchor customer via an August 2026 announcement of a five-year, $3.4 billion managed services AI cloud contract (claim 1299, later confirmed as NVIDIA per claim 8277), though early disclosure language called this "previously announced" without counterparty identification until late August. Dell operates as a GPU procurement partner rather than end customer, with two supply agreements totaling $7.4 billion ($5.8 billion per claim 4161; $1.6 billion per claim 1297), structurally representing cost of goods sold against Microsoft and NVIDIA revenue. The third-tier customer base emerged in July 2026 as a $2.8 billion aggregate announced against "leading AI developers" (claim 360), later named selectively as Figure AI, Fluidstack, Fireworks AI, and Hume AI (claim 8117, 363), alongside unnamed mentions of Perplexity, Together AI, and Fal AI. This naming pattern—vague descriptor followed by partial enumeration—obscures whether all listed entities have signed binding contracts or represent discussions. The portfolio exhibits acute concentration risk: Microsoft and NVIDIA represent $13.1 billion, or roughly 66 percent of named customer commitments, against $2.8 billion distributed across approximately eight smaller counterparties. Early-stage partnerships with Bitmain Technologies (claim 6374, amended per claim 6152) and Poolside AI (claim 6285) carried no disclosed revenue figures and have not reappeared in recent customer rosters. An August 2026 Golden State Warriors sponsorship agreement (claim 7895) for $50+ million annually sits outside the AI infrastructure segment. Concentration deepened materially between May 2026, when only NVIDIA and an unnamed $3.4 billion customer were disclosed, and August 2026, when Microsoft's $9.7 billion anchor role crystallized. The five-year revenue arithmetic—$19.7 billion in announced commitments across five years implies roughly $3.94 billion annualized contractual revenue, contingent on full tranche delivery and acceptance schedules remaining on track. Customer prepayments covering approximately 45 percent of GPU capital expenditure (claim 8595) suggest some upfront cash realization, though this represents customer financing of IREN's infrastructure build rather than incremental revenue recognition.

IREN's disclosed customer and partnership portfolio exhibits a shift from signed commercial agreements to vague categorization of major relationships. The early partnerships—Bitmain Technologies (claims 6374, 6152) and Poolside AI (claim 6285)—employed explicit contractual language ('entered into an agreement,' 'commencement of GPU cloud service'), establishing legal binding. The November 2025 announcements cemented this pattern: Microsoft ($9.7 billion, five-year contract; claims 8112, 4157, 8563) and the Dell GPU supply agreement ($5.8 billion; claim 4161) both deployed signed-contract terminology. However, subsequent disclosures muddied this clarity. By July-August 2026, IREN began describing relationships as 'customer base' or 'partnerships' without consistently specifying execution status. Claim 8101 lists Microsoft, NVIDIA, Perplexity, Figure AI, Together AI, Fluidstack, Fireworks AI, Fal AI, Hume AI, and notably 'a new leading AI developer'—an unnamed counterparty—without distinguishing which represent signed agreements versus prospective engagements. Claim 362 describes the NVIDIA relationship as a 'broader strategic partnership' rather than a quantified contract, despite the earlier claim 8277 specifying '$3.4 billion five-year AI cloud agreement.' The concentration narrative evolved sharply: 2024 showed dispersed partnerships; by late 2025, Microsoft and NVIDIA dominated disclosed value ($13.1 billion of $15.9 billion in named commitments). July 2026 introduced diversification via the $2.8 billion AI developer cohort (claim 8117), yet the cohort remained partially unnamed. Regarding revenue arithmetic, the disclosed $15.9 billion across Microsoft ($9.7B), NVIDIA ($3.4B), and AI developers ($2.8B) is nominally reconcilable, but deployment uncertainty obscures realization timing. Claims reference 'tranches' (4157, 4161) and a weighted average contract term of approximately four years (claim 8108), implying phased revenue recognition. The Dell agreements represent capital expenditure, not revenue. The Golden State Warriors sponsorship ($50+ million annually; claim 7895) operates on a separate economics track. Notably absent: explicit MOU or pilot language, suggesting most agreements carry binding legal status—yet the shift toward indefinite descriptors ('customer base,' 'strategic partnership') and the introduction of unnamed counterparties represent a material softening of disclosure precision compared to 2025's contract-specific nomenclature.

IREN's partnership disclosures conflate signed customer agreements with framework partnerships and strategic descriptions. The core cloud revenue contracts appear definitive: Microsoft's $9.7 billion five-year cloud services agreement (claims 8112, 4157, 686) shows clear execution with acceptance milestones documented (Horizon 1 accepted August 2026, claims 35, 722), and NVIDIA's parallel $3.4 billion five-year arrangement (claims 8277, 351) similarly reads contracted rather than exploratory. A $2.8 billion multi-year AI developer pool (claims 360, 8117, 8275) names specific counterparties—Figure AI, Fluidstack, Fireworks AI, Hume AI—though individual allocations remain undisclosed. Conversely, IREN describes NVIDIA using both customer and "strategic partnership" language (claim 362), creating ambiguity about relationship formality, and lists "a new leading AI developer" without identification (claim 8101), suggesting unnamed counterparties remain on record.

Customer concentration has shifted dramatically from early diversification to hyperscaler dependency. Initial announcements featured Bitmain (miner supply, claim 6374) and Poolside AI GPU services (claim 6285). By late 2025, two counterparties dominated: Microsoft and Dell accounted for $15.5 billion in stated commitments. Dell's deal, however, revised from $5.8 billion (claim 4161) to $1.6 billion (claim 1297) between May and June 2026—a 72 percent reduction signaling financial pressure or renegotiation. More concerning, Dell's arrangement functions as a supply contract for IREN's capital expenditure, not a revenue-generating customer agreement.

Revenue arithmetic reveals concentration risk and temporal misalignment. Across five-year contracts, Microsoft ($9.7 billion) and NVIDIA ($3.4 billion) represent approximately 77 percent of named cloud services commitments. The $2.8 billion AI developer bucket distributes across multiple counterparties without itemized transparency. IREN discloses a weighted average contract term of four years (claim 8108), which conflicts with the stated five-year terms on flagship deals and materially compresses implied annual revenue run-rate calculations. The August 2026 Golden State Warriors jersey sponsorship commitment (claims 7895, 1033) at $50 million annually represents non-core capital allocation, diluting focus from cloud infrastructure revenue concentration.

IREN's announced customer base reveals material concentration, limited transparency on deal timing, and potential double-counting across disclosure cycles. Between November 2025 and August 2026, the company articulated three explicitly named, signed cloud services agreements: a five-year, $9.7 billion contract with Microsoft (claim 8112), a five-year, $3.4 billion agreement with NVIDIA (claim 8277), and a $2.8 billion multi-year arrangement with AI developers including Figure AI, Fluidstack, Fireworks AI, and Hume AI (claim 8117). These three agreements represent $15.9 billion in disclosed gross contract value. Earlier announcements involving Bitmain (claim 6374) and Poolside AI (claim 6285) deployed non-standardized language—"entered into an agreement," "commencement of service"—without specifying contract size, suggesting either smaller early pilots or incomplete disclosure.

Customer concentration remains acute. Microsoft and NVIDIA together represent approximately 82 percent of disclosed named contract value—$13.1 billion of $15.9 billion. The August 2026 customer roster (claim 8101) adds Perplexity, Together AI, Figure AI, Fluidstack, Fireworks AI, Fal AI, and Hume AI; however, only Figure, Fluidstack, Fireworks, and Hume are explicitly tied to the $2.8 billion contract (claim 8117). Perplexity and Together AI remain unattached to specific contract values, complicating whether they represent signed deals or strategic positioning language.

Reconciliation issues cloud the narrative. A $3.4 billion "managed services AI cloud contract" was announced in May 2026 without customer attribution (claim 1299), then characterized as a NVIDIA agreement by August (claim 8277), suggesting either delayed customer identification or portfolio reframing. Dell's GPU supply agreements—$5.8 billion plus $1.6 billion subsequently (claims 4161, 1297)—are capital expenditures, not customer revenue, yet frequently cited alongside revenue contracts in earnings discussions. The August 2026 Golden State Warriors sponsorship ($50+ million annually; claim 1033) introduces a non-recurring revenue stream not integrated into core customer concentration analysis.

IREN's language uniformly uses "agreement" and "contract" terminology (claims 8112, 8277, 8117)—stopping short of MOU, LOI, or framework language—suggesting executed instruments. One exception: NVIDIA is separately characterized as a "broader strategic partnership" (claim 362) alongside the signed $3.4 billion cloud agreement, conflating distinct business relationships or implying scope beyond the disclosed contract value.

IREN's $28.6 billion in announced customer commitments exhibit stark concentration risk, with Microsoft and NVIDIA representing 46 percent of the total through signed five-year agreements valued at $9.7 billion and $3.4 billion respectively (claims 686, 8277). Nearly all material counterparties are identified by name: Microsoft, NVIDIA, Dell (GPU supply, $7.4 billion combined), and four discrete AI developers on a $2.8 billion arrangement—Figure AI, Fluidstack, Fireworks AI, Hume AI (claim 8117). The exception is material: claim 8101 references a "new leading AI developer" without identification, obscuring the identity of a potentially significant customer. Earlier partnerships with Bitmain Technologies and Poolside AI received formal agreement language (claims 6374, 6285), though both predate the hyperscaler concentration shift of 2025–2026.

IREN employs strategic-partnership language around NVIDIA despite having disclosed a priced, five-year service agreement (claim 362), blurring the distinction between true partnership and binding commercial contract. The customer roster clusters into three tiers: hyperscalers (Microsoft, NVIDIA), mid-market AI software developers ($2.8 billion), and legacy infrastructure partners (Bitmain, Poolside).

Revenue recognition mechanics present a material reconciliation gap. IREN discloses $2.3 billion in annualized recurring revenue, predominantly from the Microsoft contract, against a $9.7 billion five-year total—implying 24 percent annual run-rate revenue (claim 1023). The company has collected $1.9 billion in prepayments (claim 1021), with first billing triggered by Horizon 1 delivery and acceptance in August 2026 (claim 722). This phased revenue structure, combined with the four-tranche Microsoft deployment schedule, suggests earnings will exhibit material lumpiness through 2030.

Dell supply contracts ($7.4 billion) represent capital expenditure, not revenue. The absence of gross-margin disclosure on these pass-through GPU and infrastructure costs obscures the true economics of IREN's hyperscaler business model. Without trailing revenue or cash-receipt data, the stated ARR targets remain unverified against execution risk.

IREN's major disclosed customer agreements fall into clearly named, signed arrangements and a smaller category of partially identified or framework deals. The primary signed contracts are with Microsoft ($9.7 billion over five years, claim 8112, with formal acceptance of Horizon 1 triggering billing in August 2026 per claim 722) and NVIDIA ($3.4 billion over five years, claim 351). These represent genuine executed agreements, evidenced by Microsoft's documented acceptance of the first facility (claim 35) and tangible delivery milestones. Dell's agreements (claims 4161, 1297, totaling $7.4 billion) are supply purchase commitments, not revenue-generating customer contracts—IREN is the buyer, with the parent company guaranteeing obligations (claim 4164). The $2.8 billion announced in July 2026 (claims 360, 8117) names five counterparties (Figure AI, Fluidstack, Fireworks AI, Fal AI, Hume AI), but lacks granular disclosure of individual contract terms or durations, creating ambiguity about whether these are bundled or separate commitments. Earlier partnerships with Bitmain (2024, claims 6374, 6152) and Poolside AI (2024, claim 6285) disappear from recent customer rosters, suggesting maturation or termination without public disclosure. Concentration is acute: Microsoft and NVIDIA alone represent $13.1 billion across approximately four-year terms, while perimeter names like Perplexity and Together AI appear only in roster listings (claim 8101) without disclosed contract values. The revenue arithmetic reveals timing friction. IREN claimed $2.3 billion in annualized recurring revenue as of August 22, 2026 (claim 1023), with $1.9 billion in Microsoft prepayments (claim 1021). Yet pro-rata division of the $9.7 billion Microsoft and $3.4 billion NVIDIA commitments by their four-year weighted average term (claim 8108) yields approximately $3.3 billion expected ARR, suggesting either unrealized revenue from contracts not yet billing, or the disclosed ARR encompasses only executed tranches. The Golden State Warriors sponsorship ($50 million annually, claim 7895) represents media rights, not infrastructure services, and should be classified separately. Overall, IREN discloses two major signed customer agreements with name-brand hyperscalers, a supply relationship it finances, and a cluster of smaller AI developer commitments with limited contract visibility—a portfolio heavily dependent on Microsoft's acceptance milestones and NVIDIA's fulfillment across a multi-year ramp.

IREN's partnership disclosures reveal a sharp distinction between formally executed service contracts and strategically positioned relationships. Named, signed customer agreements include Microsoft ($9.7B five-year GPU services contract, partially delivered with Horizon 1 acceptance as of August 2026), NVIDIA ($3.4B five-year agreement, though labeled 'strategic partnership'), and Dell (supply agreements totaling $7.4B, though Dell is a vendor rather than cloud customer). Early Bitmain and Poolside AI partnerships from 2024 appear superseded by hyperscaler focus. IREN disclosed a July 2026 aggregate $2.8B multi-year contract with named AI developers—Figure AI, Fluidstack, Fireworks AI, Hume AI, Together AI—but without individual amounts (claims 360, 8117). Perplexity appears in customer rosters (claims 363, 8101, 350) without stated deal value. The vocabulary around NVIDIA merits scrutiny: claim 362 characterizes it as a 'broader strategic partnership' covering 'global 5 GW secured data center pipeline,' yet claim 351 cites a 'five-year, $3.4 billion AI cloud agreement,' suggesting contract substance beneath partnership nomenclature. IREN's counterparty concentration is extreme. Microsoft and NVIDIA account for roughly 78% of disclosed customer contract value ($13.1B of $16.7B). This concentration creates revenue dependency risk despite roster breadth. Regarding arithmetic: claim 1023 reports $2.3 billion in ARR 'linked primarily to its $9.7 billion multi-year contract with Microsoft.' If annualized evenly, Microsoft should yield $1.94B annually; NVIDIA another $680M. The gap between expected $2.62B run-rate and reported $2.3B ARR could reflect: (1) ramp schedules with delayed revenue recognition, (2) accounting conservatism, or (3) incomplete contract recognition at disclosure date. Claim 1021 notes $1.9B in Microsoft prepayments, suggesting front-loaded cash flow masking actual service delivery timing. The Golden State Warriors sponsorship (claims 1033, 7895, $50M+ annually) represents non-core revenue concentration. Fundamentally, IREN has transitioned from exploratory partnerships to signed hyperscaler commitments, but revenue visibility depends critically on Microsoft execution risk and the bundled $2.8B AI developer cohort remaining undefined at granular level.

IREN's customer announcements present a heavily concentrated portfolio with clear distinctions between signed and non-binding language, though one key counterparty remains partially unnamed. All major cloud-service agreements—Microsoft's $9.7 billion five-year contract (claims 8112, 4157, 686, 7454), NVIDIA's $3.4 billion cloud agreement (claims 351, 726, 781), and the Dell supply commitment of $5.8 billion (claim 4161)—employ executed agreement language ("entered into an agreement," "announced," "accepted delivery"). Early-stage partnerships with Bitmain (claim 6374) and Poolside AI (claim 6285) similarly reference actual agreements rather than preliminary frameworks. NVIDIA partnership language appears in claim 362 as "entered into a broader strategic partnership," but this describes the same signed $3.4 billion cloud agreement, not a separate MOU.

Counterparty concentration reflects dramatic consolidation around two hyperscalers. In 2024, IREN served diverse smaller partners (Bitmain, Poolside AI). By November 2025, Microsoft and the Dell supply chain dominated announced value. July-August 2026 announcements introduced material diversification: a $2.8 billion multiyear cloud contract with Figure AI, Fluidstack, Fireworks AI, and Hume AI (claims 360, 8117), alongside confirmed relationships with Perplexity and Together AI (claims 363, 8101). However, Microsoft and NVIDIA remain primary counterparties, representing approximately $13.1 billion of disclosed customer revenue over contract terms.

One material disclosure gap concerns an unidentified "new leading AI developer" in claim 8101's customer roster; no follow-up claim names this entity. Revenue reconciliation reveals internal inconsistencies. Claimed annualized recurring revenue of $2.3 billion (claim 1023) tied "primarily" to Microsoft aligns imperfectly with contractual terms: Microsoft's $9.7 billion over five years implies approximately $1.94 billion annually; NVIDIA's $3.4 billion similarly implies approximately $680 million annually, totaling $2.62 billion before the July $2.8 billion deal. This suggests either non-linear revenue recognition, near-term acceleration, or that claimed ARR excludes certain customer contracts. Dell's $7.4 billion combined supply agreements (claims 4161, 1297) represent infrastructure costs, not customer revenue—correctly excluded from revenue metrics but material as capital obligations. The arrangement's debt financing ($3.65 billion tied to Microsoft contracts per claim 1652) underscores leverage dependency on two primary customers.

IREN's public customer announcements reveal sharp distinctions between signed supply agreements and strategic partnership language that obscure actual revenue recognition. The cornerstone contracts—Microsoft ($9.7 billion, five-year agreement) and NVIDIA ($3.4 billion, five-year agreement)—are named and substantively documented (claims 686, 7454, 351). Microsoft's delivery and acceptance of Horizon 1 on 2026-08-21 (claims 722, 35) provides the only concrete evidence of execution; billing has commenced, and IREN has recognized $1.9 billion in prepayments and $2.3 billion in annualized recurring revenue (claims 1021, 1023). However, the disclosed revenue arithmetic exhibits material inconsistencies. The Dell GPU supply agreement was initially announced at $5.8 billion (claim 4161, November 2025), but resurfaced at $1.6 billion four months later (claim 1297, May 2026)—a 72-percent reduction that suggests either deal restructuring or disclosure of a separate tranch. The announced $3.4 billion "managed services AI cloud contract" (claim 1299, May 2026) and the August disclosure of a parallel $3.4 billion NVIDIA cloud agreement (claim 351) invite reconciliation questions: are these the same deal or separate counterparties? Only Microsoft and NVIDIA carry defined contract values and durations; a roster of eight additional counterparties—Perplexity, Figure AI, Together AI, Fluidstack, Fireworks AI, Fal AI, Hume AI, and earlier Bitmain and Poolside AI—are named without disclosed values (claims 6374, 6285, 363), while NVIDIA is simultaneously described with "strategic partnership" language alongside the quantified cloud agreement (claim 362). Concentration risk is acute: Microsoft and NVIDIA represent 52 percent of disclosed contract value on a two-party basis. The Golden State Warriors sponsorship (claim 7895) operates outside the core infrastructure business. Until the Dell and NVIDIA agreements show acceptance events and prepayment evidence matching Microsoft's cadence, IREN's revenue visibility remains anchored to a single, though substantial, counterparty.

IREN's disclosed customer relationships reflect a portfolio of signed agreements concentrated among a narrow set of counterparties, with material gaps in transparency around smaller customers. The contractual foundation rests on three core signed deals with named counterparties: Microsoft's five-year $9.7 billion GPU services agreement (signed November 2025, with Horizon 1 accepted by August 2026, claims 1266, 686, 7454), NVIDIA's five-year $3.4 billion cloud services agreement (claims 351, 726, 781), and Dell's two supply agreements totaling $7.4 billion spanning 2025–2026 (claims 4161, 1297). Earlier partnerships with Bitmain (claims 6374, 6152) and Poolside AI (claim 6285) involved signed agreements but disclosed no material terms, suggesting framework arrangements below anchor-customer scale.

The remaining customer roster—Perplexity, Figure AI, Together AI, Fluidstack, Fireworks AI, Fal AI, and Hume AI (claim 363)—are named without accompanying contract specificity, financial terms, or performance milestones. These are listed as members of a collective "customer base" rather than disclosed as signed agreements or material partnerships, creating a visibility gap: eight AI developers are cited for credibility without clarification of whether relationships constitute executed service contracts, pilots, or strategic framework discussions.

Counterparty concentration has consolidated sharply. Early-stage announcements (2024) emphasized diversification across Bitmain and Poolside AI. By late 2025, Dell and Microsoft emerged as capex and revenue anchors, respectively. As of August 2026, Microsoft and NVIDIA represent approximately $13.1 billion of the disclosed $20.3 billion in named-deal value—64% of announced contract size. Microsoft alone dominates: the $9.7 billion contract triggered billing upon Horizon 1 acceptance (claims 35, 722), and disclosures of $1.9 billion in prepayments and $2.3 billion in ARR (claims 1021, 1023) link these figures directly to Microsoft performance.

Revenue arithmetic appears reconcilable but dependent on prepayment treatment. The $9.7 billion Microsoft contract amortized across five years yields approximately $1.94 billion annualized run-rate, consistent with reported $2.3 billion ARR if supplemented by the $2.8 billion in multi-year contracts announced July 2026 (claim 360). However, the $1.9 billion prepayment raises questions about revenue recognition methodology and ARR sustainability absent diversification. The Golden State Warriors sponsorship ($50 million-plus annually, claims 7895, 1033) is material to brand presence but immaterial to revenue models.

In summary: core counterparties are named and signed; smaller customers lack disclosed terms; revenue concentration in Microsoft and NVIDIA is pronounced; arithmetic reconciles only if prepayments are properly accounted. Near-term revenue stability depends on Horizon deployment execution and NVIDIA infrastructure absorption.

IREN's customer disclosure evolved dramatically from early-stage partnerships to major equipment and cloud deals. The company's 2024 announcements (claims 6374, 6285) identified Bitmain and Poolside AI as counterparties to unnamed agreements without disclosed value. These early deals remain unquantified and their status—whether active, expanded, or concluded—is unclear.

The material shift began in November 2025, when IREN disclosed two definitive, signed agreements: Dell supplied $5.8 billion in GPUs scheduled for delivery starting March 2026 (claim 4161), and Microsoft committed to a five-year, $9.7 billion GPU services contract at Texas data centers (claim 4157). A second Dell agreement for $1.6 billion followed in May 2026 (claim 1297). That same month, claim 1299 referenced a "previously announced $3.4 billion managed services AI cloud contract," later confirmed as NVIDIA (claims 781, 1758, 2026-08-21). These three counterparties—Microsoft, Dell, and NVIDIA—account for the entirety of publicly disclosed contract values: roughly $20.5 billion across five-year terms.

From August 2026 forward, IREN expanded its named customer roster to include Perplexity, Figure AI, Together AI, Fluidstack, Fireworks AI, Fal AI, and Hume AI (claim 363). Critically, none of these seven names are attached to contract values. In contrast, NVIDIA was called a "strategic partner" (claim 362) despite the $3.4 billion deal being separately quantified, suggesting the partnership language encompasses both the cloud agreement and broader commercial cooperation. A $2.8 billion "multi-year cloud services contract" announced July 20, 2026 (claim 360) names no counterparties; their identities remain undisclosed.

Revenue arithmetic partially reconciles. Claim 1023 (2026-08-22) states $2.3 billion in annualized recurring revenue, attributed primarily to Microsoft's $9.7 billion five-year contract (implying ~$1.94 billion per year). Adding NVIDIA's $3.4 billion over five years (~$0.68 billion annually) yields ~$2.6 billion, roughly consistent with the $2.3 billion ARR, suggesting Dell revenue may be recognized differently or excluded. Counterparty concentration is extreme: Microsoft and Dell represent approximately 82 percent of the $20.5 billion disclosed value, creating significant revenue dependency on two customers. The unnamed Perplexity, Figure AI, and others may be small pilots or early-stage deployments lacking formal contract announcements, but this remains speculative.

IREN's announced customer and partnership portfolio shows a clear bifurcation between named, binding agreements and opaque framework arrangements. The company has disclosed signed contracts with three material counterparties: Microsoft ($9.7 billion over five years for dedicated GPU infrastructure delivered in tranches, with Horizon 1 accepted as of August 2026 per claims 35, 722, 686), NVIDIA ($3.4 billion over five years for managed cloud services, with Exemplar Cloud certification awarded in August per claims 347, 781, 1758), and Dell (GPU supply agreements totaling $7.4 billion in infrastructure purchases per claims 1297, 4161). These represent binding commitments with explicit counterparty identification and financial terms. Earlier partnerships with Bitmain Technologies and Poolside AI (claims 6374, 6285, 6152) similarly reference executed agreements. However, IREN announced $2.8 billion in multi-year AI cloud contracts on July 20 and August 20 (claims 360, 7575) described only as contracts "with leading AI developers"—a phrase that deliberately obscures counterparty identity. This same vague language appears in the August 13 customer roster claim (363), which names Perplexity, Figure AI, Together AI, Fluidstack, Fireworks AI, Fal AI and Hume AI without quantifying any deal value. Counterparty concentration has shifted dramatically from the fragmented 2024 portfolio (Bitmain, Poolside AI) to a two-customer model by mid-2026: Microsoft and NVIDIA account for $13.1 billion of the $15.9 billion in announced cloud services contracts, or 82 percent. The unnamed $2.8 billion contract represents material undisclosed exposure. Revenue arithmetic presents timing complications but no clear reconciliation failures. The company claims $2.3 billion in annualized recurring revenue as of August 22 (claim 1023), which aligns directionally with Microsoft contributing ~$1.94 billion annually ($9.7B/5 years) plus partial contribution from the NVIDIA deal and unnamed contracts. However, $1.9 billion in Microsoft prepayments (claim 1021) front-load recognized revenue, and the unnamed $2.8 billion contract—announced as "secured" but never sourced—introduces material uncertainty about whether the revenue mix disclosed in pricing actually matches counterparties disclosed elsewhere in the public record. The company has mixed infrastructure capex (Dell GPU purchases, $7.4B) with service contracts in its headline announcements, potentially inflating apparent customer wins.

IREN's disclosed customer base shows singular concentration on a named counterparty with evidence of contract performance. Claim 35 (2026-08-12) references Microsoft as the exclusive identified customer, describing "delivery to, and acceptance by Microsoft of Horizon 1"—language that indicates actual fulfillment rather than non-binding discussion or preliminary framework. The phrasing "acceptance by" suggests active counterparty acknowledgment, lifting this beyond letter-of-intent or memorandum-of-understanding territory into completed delivery.

The granularity available is limited. IREN has named only one customer across the record: Microsoft. No other strategic partners, anchor customers, or pilot arrangements appear to be publicly identified. This creates a significant concentration risk profile, where observable commercial traction reduces to a single counterparty relationship. The absence of disclosed discussions with other potential customers, named or described generically as "leading technology companies" or similar hedging language, leaves the market with incomplete visibility into the sales pipeline or customer diversification trajectory.

On contract type, the "delivery and acceptance" formulation differs sharply from typical MOU or pilot language. Memoranda of understanding usually contain qualifiers like "anticipated to," "subject to," or "under discussion." Pilot arrangements typically carry language around limited scope, evaluation periods, or contingency gates. The acceptance language here points toward binding purchase or development agreement completion. However, one claim does not establish a pattern of recurring signed revenue or customer retention.

Revenue reconciliation is unaddressed in the available claims. There is no disclosure of contract value, delivery timeline, expected revenue recognition, or order volume for the Horizon 1 delivery. Without corresponding financial statements or supplementary disclosure, it is impossible to verify whether IREN's disclosed revenue or backlog figures align with a Microsoft delivery. If IREN trades publicly or files financial reports, cross-reference of claim 35's delivery date against quarterly revenue or deferred revenue line items would be necessary to establish arithmetic coherence.

In summary, IREN's announced customer base consists of one named counterparty with evidence of actual contract performance. Concentration risk is acute. Independent financial verification is needed to link this known deal to company revenue.

IREN's announced footprint has expanded by roughly an order of magnitude, and the expansion has come in steps rather than through any single restatement. Childress entered the record as a 100 MW first phase with a $197.5M capital estimate (31139, 31140). By late 2023 it had become a 600 MW site (12176, 12186), and the estimated improvement cost had reached $1.4B by August 2024 (31389). The West Texas pipeline, now Sweetwater, was framed at 1,400 MW (6523). Hashrate followed a less linear path. The November 2021 target of 15.2 EH/s by September 2023 (14393) was effectively rebased: operating capacity sat at 5.5 EH/s through spring 2023 (12689, 12712), and later targets of 9.1, 9.4 and 10 EH/s (12465, 12138, 6341) were presented as fresh expansions rather than as recovery toward a missed goal. The verified tag on 14393 deserves scrutiny against that gap. The 2021 Bitmain contract behind it (14392) remains unverifiable.

Near-term operational dates have largely held. The 80 MW Childress expansion due mid-2024 (6495, 6410), the 50 MW of 2024 construction (6153), the H100 commissioning (6262) and the expanded GPU fleet of 1,896 units (5712) all carry verified status. The longer-dated abatement milestones look different. The original agreement set parallel deadlines of 300 MW by end-2024 and 200 MW by end-2025 (31399, 31400, 31409, 31411). The October 2024 filings restate the 200 MW tranche as Phase 3 due by end-2026 (31332, 31334), and Phase 2 is restated at 300 MW (31361). This reads as a one-year deferral made through a formal amendment, not through investor communications.

Independent support is strongest at Childress. County reinvestment-zone filings and land additions (31386, 31390) document the build, and the abatement deadlines themselves come from public agreements. Sweetwater has no corresponding permit, ERCOT interconnection or site-reporting evidence in this record, despite its under-construction status. The record supplied ends mid-entry in October 2024, so any 2025–2026 restatements are not assessed here.

The next 90 days include the 2026-12-31 Phase 3 deadline (31334). Watch for Childress County certifying 200 MW of qualifying capacity or another amendment pushing the date out. Also watch for ERCOT energization records or interconnection-agreement milestones at Sweetwater 1. Any capacity figure for Sweetwater that differs from the 1,400 MW framing would also matter.

IREN's record shows a company whose unit of ambition changed partway through. Through 2023 the headline metric was hashrate, and it climbed steadily: from above 2.3 EH/s (13824) to 3.7 (13604), a long plateau at 5.5 (12784, 12689), then targets of 9.1 and 9.4 (12443, 12138), 10 EH/s operating by May 2024 (6150), 20 installed by September 2024 (5689), and 31 targeted for Q4 2024 (5691). The more consequential scaling is in power. Childress began as a 100 MW Phase 1 with a $197.5 million capital estimate (31139, 31140). By late 2023 it was a 600 MW site (12176, 12186) with another 500 MW of headroom (6412), and the 2024 Childress filings estimate $1.4 billion in improvements (31389). Sweetwater adds a 1,400 MW West Texas footprint (6523). Alongside this, the GPU line grew from 568 H100s (6261) to 1,896 units (5712). Childress has turned from a mining site into the anchor of an AI-compute pitch.

On dates, the near-term commitments mostly held. The 80 MW Childress expansion (6495, 6410), the T21 shipments (6408) and the GPU commissioning (6262) are all marked verified by their mid-2024 deadlines. The longer Childress milestones are a different matter. The 2022 agreement set 300 MW by end-2024 (31399). The October 2024 filings set a Phase 2 300 MW block by end-2025 (31375), plus an optional extension of commercial operations to end-2026 (31364). In effect, the 300 MW milestone moved back a year and gained more slack, and the record shows no public acknowledgement of the shift. One more claim needs reconciling: 15.2 EH/s by September 2023 is marked verified (14393), yet the company's own November 2023 figure was 8.3 EH/s (6493).

The independent evidence is thin and comes from a single type of source. The Childress reinvestment-zone and abatement filings (31145, 31390, 31386) confirm land control and tax incentives, but they do not show energization. The record contains no ERCOT interconnection-queue entries, no Sweetwater building permits and no on-the-ground reporting. The announced Sweetwater pace therefore rests mainly on company statements. The record also ends mid-entry at claim 31337, so nothing after October 2024 was reviewed.

Over the next 90 days, the key test is the extended end-2026 commercial-operations date for Childress (31364). Watch whether IREN certifies the Phase 2 capacity or invokes further relief under the abatement terms. For Sweetwater 1, watch for an ERCOT energization or substation milestone that would back up its under-construction status. Finally, check whether the new GPU and Horizon disclosures report megawatts actually energized rather than capacity secured.

The extract supplied ends mid-entry at claim 31343, dated October 2024, so this read does not cover anything IREN has said in the two years since. That gap matters most for Sweetwater and the newer announced sites. Within that window, IREN has scaled its ambitions in two separate units. In hashrate, operating capacity climbed from 2.3 EH/s (13824) to a 5.5 EH/s plateau held across seven statements through April 2023 (12827, 12689). Targets then reset to 9.1 (12443), 9.4 (12138) and 10 EH/s (6150), before jumping to 20 installed (5689) and 31 EH/s for Q4 2024 (5691). The early 15.2 EH/s target for September 2023 (14393) is marked verified, yet it sits awkwardly beside a 5.5 EH/s plateau and a 9.1 EH/s target set only months before that deadline. The underlying Bitmain contract (14392) remains unverifiable, which suggests the original pathway was restructured rather than delivered as written. In megawatts, Childress has expanded steadily. The 2022 abatement filings set a 100 MW first phase at roughly $197.5M (31139, 31140). Company messaging later cited a 600 MW site (12176) plus 500 MW of additional power (6412), and county filings now reference $1.4B of improvements (31389) and a second reinvestment zone (31390). Sweetwater's 1,400 MW (6523) appears once, without a date. Near-term dates have mostly held: the 80 MW Childress block by mid-2024 (6410), the 50 MW 2024 build (6153), and GPU deliveries (6262, 5712) are all verified. The larger milestones behave differently. The October 2024 Phase 2 agreement keeps a 300 MW by end-2025 obligation (31375) but adds an optional extension of commercial operations to end-2026 (31364). That is a quiet softening of the deadline, built into the contract rather than announced as a slip. The strongest independent evidence here is the Childress County reinvestment zone record, including its 20-year retention clause (31145), which shows real land and capital commitments. Nothing in this record reflects ERCOT large-load or interconnection filings for Sweetwater, and Kiowa, Oklahoma, Bundey and Redbank have no third-party trace at all. Over the next 90 days, the key test is whether IREN met the Childress commitment by 31 December 2026 or exercised the extension, which county compliance filings should reveal. Next are ERCOT queue or energization evidence for Sweetwater 1, and whether any of the announced-only sites gain permits or utility agreements rather than further press mentions.

IREN's capacity story has grown on two separate tracks. On the mining side, operating hashrate climbed in steps from about 2.3 EH/s (13824) to 5.5 EH/s (12784), then 9.1 to 9.4 (12448, 12138), 10 (6150), 20 installed (5689) and a 31 EH/s target for Q4 2024 (5691). The early 15.2 EH/s target for September 2023 (14393) carries a verified tag. That tag sits uneasily beside the company's own statements through mid-2023 that installed capacity was 5.5 EH/s (12689, 12712), so the 2021 target looks met well after its date rather than on schedule. On the power side, Childress grew from a 100 MW Phase 1 floor (31139) to 600 MW of site potential (12176), plus another 500 MW (6412). The October 2024 filings re-tiered it as a 300 MW Phase 2 and a 200 MW Phase 3 (31375, 31343). Over the same period the estimated investment rose from $197.5M for Phase 1 (31140) to $1.4B (31389). Sweetwater entered at 1,400 MW (6523) with no phasing or dates attached.

Near-term dates have mostly held. The 80 MW Childress expansion (6495, 6410), the 50 MW build in 2024 (6153) and the H100 and H200 GPU deliveries (6262, 5711) are all marked verified. The longer abatement milestones are different. The 2022 filings set commercial-operations deadlines of end-2024 and end-2025 (31400, 31411). In 2024 an optional extension to end-2026 appeared (31364). The company has not presented this as a delay, but it moves the outside deadline back by a year or more.

The independent evidence is uneven. The Childress claims seem to come from county reinvestment-zone and tax-abatement filings, which tie the capacity floors to public obligations. The record contains no comparable permit, ERCOT interconnection-queue or site-reporting evidence for Sweetwater 1 and 2, the Oklahoma campus, Kiowa, Bundey or Redbank. Those sites list no capacity or cost on record. The supplied claims also stop in October 2024, so any later commitments are not assessed here.

Over the next 90 days, the key question is whether Childress meets or formally invokes the end-2026 operations deadline (31364), and whether the 300 MW Phase 2 floor (31375) is met in full or only in part. The other markers are Sweetwater 1 energization and any ERCOT queue filing that backs the 1,400 MW figure (6523). The last is whether any of the announced-only sites gets a permit or interconnection filing that would move it beyond a press release.

IREN's capacity story has expanded in steps, and each step has been large. Childress first appeared in 2022 county abatement filings as a 100 MW Phase 1 (31139) inside a 200 to 300 MW minimum commitment (31399, 31409). By September 2023 it was described as a 600 MW site (12176, 12186). By December 2023 another 500 MW had been added (6412). The 2024 filings brought a second reinvestment zone, adjacent land and a $1.4 billion improvement estimate (31389, 31390, 31360). Sweetwater then put a 1.4 GW West Texas figure on top of all that (6523). The headline numbers have only grown, and no announced figure has been publicly walked back. The hashrate guidance follows the same pattern. It went from 9.1 EH/s (12465) to 10 EH/s (6341) to a 20 EH/s pathway (6342), then 20 EH/s installed (5689) and 31 EH/s by year-end 2024 (5691).

Dates are more uneven. The near-term build commitments mostly held. The 80 MW Childress expansion due mid-2024 (6495, 6410), the 50 MW 2024 tranche (6153), the 568-GPU commissioning (6262) and the H200 fleet expansion to 1,896 units (5712) all carry verified marks. The early 15.2 EH/s target (14393) is harder to accept at face value. Operating capacity was repeatedly stated at 5.5 EH/s through April 2023 (12827, 12712, 12689) and 9.4 EH/s that October (12138). Meanwhile the underlying Bitmain contracts (14392) remain unverifiable. The original target appears to have been reset, not met on schedule, and the reset was never explicitly framed as a slip. Childress also carries two different commercial operations deadlines, end-2024 (31400) and end-2025 (31411). Each sits against a different MW floor, so it is unclear which obligation actually governs.

The independent evidence is strongest at Childress. Abatement agreements, reinvestment zone designations and land purchases are public record and match the stated pace. Sweetwater has a street address but no permit or interconnection queue evidence in this record. Sweetwater 2, Kiowa, the Oklahoma campus, Bundey and Redbank rest on announcements alone.

Over the next 90 days, watch whether the Childress Phase 2 floor (31375) and its December 2025 deadline show up in county compliance reporting as actually met. Watch for ERCOT large-load or interconnection filings that would anchor the 1.4 GW Sweetwater figure. Watch for the first permits or utility agreements at the unlocated announced sites, and for a GPU fleet disclosure that moves meaningfully beyond 1,896 units.

IREN's capacity story has grown in scale, while near-term deliverables have repeatedly landed later than first stated, usually without the company saying so. The November 2021 framing of 15.2 EH/s across all sites by September 2023 (14393) is the clearest case. Through 2023 the company's own updates sat at 5.5 EH/s (12689, 12712), and it then set a series of smaller interim targets: 9.1 (12465), 7.0 and 9.4 (12137, 12138), and 10 EH/s (6522, 6150). It reported 20 EH/s installed only in September 2024 (5689). The original target was eventually exceeded, but roughly a year late. The company re-anchored to new milestones rather than reconciling the miss. The ambition kept compounding throughout: 31 EH/s for Q4 2024 (5691), a Childress envelope that grew from 600 MW (12176) to a further 500 MW (6412), a 1,400 MW West Texas program at Sweetwater (6523), and a GPU fleet that went from 568 H100s (6262) to 1,896 units (5712).

The Childress abatement record is the strongest independent evidence, because those commitments are filed with the county rather than made in press releases. The March 2022 agreements required 300 MW by end-2024 (31399) and 200 MW by end-2025 (31409), with Phase 1 at a 100 MW minimum (31139). The October 2024 filings carry the same 300 MW and 200 MW tranches, but due a year later, at end-2025 (31375) and end-2026 (31332, 31334). The total megawatts held while the schedule moved right, in a public document. That counts as acknowledgement to regulators, though not as an explicit admission to investors. The $1.4B improvement estimate (31389) and the land addition for Reinvestment Zone #2 (31386, 31374) show real expansion of the site footprint. The record contains no interconnection-queue or on-the-ground evidence for Sweetwater's 1,400 MW, so that figure should be treated as announced capacity, not a scheduled delivery. The supplied record also breaks off mid-entry at claim 31364, so any later revisions have not been assessed here.

The next 90 days include the 31 December 2026 Childress Phase 3 commercial-operations deadline (31334). It is the first test of whether the restated schedule holds. Watch for a county compliance filing, or for another amendment that pushes the dates out again. Also watch for ERCOT energization or queue evidence that Sweetwater 1 can match its under-construction label. Finally, watch whether the Horizon buildout at Childress is reported in megawatts actually energized, rather than in contracted or potential capacity.

IREN's delivery record splits into two eras. Through 2024 the company set hashrate and buildout targets and largely hit them. The 15.2 EH/s ambition from November 2021 (claim 14393) is marked verified. So are the 80 MW Childress expansion promised for mid-2024 (claims 6495, 6410), the T21 shipping window (6408), the 568-GPU commissioning (6262), the 50 MW 2024 build (6153) and the 31 EH/s year-end target (5691). Interim figures moved steadily upward, from 2.3 to 3.7 to 5.5 EH/s across 2022 and 2023 (13824, 13604, 12827), then to 9.1, 10 and 20 EH/s (12465, 6150, 5689). The 20 EH/s 'pathway' (6342) was overtaken within a year. The one soft spot is that the June 2023 target of 9.1 EH/s was restated as 9.4 EH/s and then 10 EH/s (12138, 6522) without a firm date attached, which makes slippage in that window hard to measure rather than evidence of it.

Capacity framing has scaled far faster than delivery. Childress went from a 100 MW Phase 1 minimum in the 2022 abatement filing (31139) to 600 MW of potential (12176). The West Texas footprint jumped from 1,400 MW at Sweetwater (6523) to 2.75 GW of secured power (31046). These are power-envelope claims, not energized load, and the record contains no dated commitment for Sweetwater 1 or 2.

The strongest independent anchor is the Childress County reinvestment-zone paperwork. The 2024 agreement binds IREN to a Phase 3 minimum of 200 MW with commercial operations by 2026-12-31 (31332, 31334, 31345). That is a contractual, externally enforceable date, and it is now roughly ten weeks away. The supplied record contains no ERCOT interconnection-queue entries, building permits or on-site reporting for Sweetwater, so its pace rests on company statements alone. The claim list also cuts off mid-entry at 31050 (April 2025), so timelines for Horizon and the AI-cloud business after that point could not be assessed here.

Over the next 90 days, three things matter. First, whether IREN reports Childress Phase 3 commercial operations before the December 31 deadline or seeks an amendment from the county. Second, whether Sweetwater 1 energization gets a firm date that ERCOT filings corroborate. Third, whether the 2.75 GW figure starts being broken into dated, contracted tranches rather than restated as a single envelope.

IREN's record shows a company whose headline capacity has grown steadily, mostly in public, while the metric it reports has changed. Early disclosures track hashrate, from a 15.2 EH/s fleet target (14393) through operating levels of 2.3 (13824), 3.7 (13604), 5.5, 9.1 (12465) and 20 EH/s (5689) to 37 EH/s (31043). The 5.5 EH/s figure was restated across at least six releases in spring 2023 (12784, 12827, 12712, 12699, 12689), which inflates the apparent volume of progress. The unit then shifts to megawatts and sites. Childress grew from a 100 MW minimum commitment (31139) to 600 MW potential (12176) and 500 MW energized (31048). West Texas grew from a 1.4 GW Sweetwater plan (6523) to 2.75 GW of claimed secured power (31046). GPU capacity rose from 568 H100s (6261) to 1,896 units (5712), then stayed flat at that figure into April 2025 (31047). That plateau matters given the Horizon pivot toward AI workloads. On dates, near-term promises have generally held. The 80 MW Childress expansion (6495, 6410), the 2024 50 MW build (6153), the Q4 2024 31 EH/s target (5691), Phase 5 energization (31049) and Horizon 1 commissioning (31050) are all marked verified. Two items deserve scrutiny. The 2021 Bitmain 14.5 EH/s purchase commitment (14392) remains unverifiable even though the aggregate target beside it was verified. Sweetwater 1's April 2026 energization (31052) is marked verified, yet the project is still listed as under construction six months later. That suggests energization of substation capacity rather than delivered data hall capacity, a distinction the company's language does not always draw. Independent corroboration in this record is thin. The Childress County reinvestment-zone filings (31139 to 31145, 31153 to 31156) are the only public-record anchors, and they support the Childress land and Phase 1 commitments, not the larger figures. The record contains no ERCOT interconnection queue entries, permits or site reporting for Sweetwater 1, Sweetwater 2 or the 2.75 GW total, and no capacity figures for Kiowa, Oklahoma, Bundey or Redbank. The supplied record also ends mid-claim in April 2025 (30435), leaving roughly eighteen months unexamined. Over the next 90 days, watch for confirmation that Sweetwater 1 has moved from energized grid capacity to operating IT load. Also watch for ERCOT filings or a signed interconnection milestone for Sweetwater 2 against its 2027 date (31053), any GPU fleet disclosure above 1,896 units, and whether the four announced non-Texas sites receive megawatt figures backed by utility or permitting documents.

IREN's delivery record is best read as two stacked trajectories: a hashrate build that moved in steps, and a power-and-land position that grew much faster than the compute placed on it. Operating hashrate rose from about 2.3 EH/s (13824) to 3.7 EH/s (13604), then held at 5.5 EH/s across at least seven restatements through spring 2023 (12784, 12689). Targets moved from 9.1 (12465) to 9.4 (12138) to 10 EH/s (6522, 6150). Installed capacity then reached 20 (5689), the 31 EH/s Q4 2024 target (5691), and 37 EH/s (31043). Power ambitions scaled faster. Childress went from a 100 MW Phase 1 floor (31139) to 600 MW potential (12176) and 500 MW energized (31048). Sweetwater added 1.4 GW (6523, 31052) and a further 600 MW (31045), for 2.75 GW secured in West Texas (31046). The GPU pivot is the outlier: 568 H100s (6261) became 1,896 units (5712), and that count was unchanged by April 2025 (31047).

Dated promises have generally held. The 80 MW Childress expansion (6495, 6410), the 2024 50 MW build (6153), the 31 EH/s target, Childress Phase 5 (31049), Horizon 1 (31050) and the April 2026 Sweetwater 1 energization (31052) are all marked verified. One tension deserves re-checking. The 2021 claim of 15.2 EH/s by September 2023 (14393) is tagged verified, yet the record shows 5.5 EH/s operating through mid-2023 and sub-10 EH/s targets afterward. Meanwhile, the companion Bitmain contract (14392) remains unverifiable. That target looks like it was reframed rather than met on schedule, with no acknowledgement in the claims. Childress Phase 1's commercial-operations deadline (31141) was never dated.

Independent corroboration is thin in this file. The reinvestment-zone land purchase (31153, 31156) points to county records, and Sweetwater 1 has a street address. There are no permit, ERCOT interconnection-queue or site-visit entries, so the verified tags appear to rest largely on company reporting.

The record also ends mid-entry at claim 30435 in April 2025, roughly eighteen months before today. That gap is the first priority. Over the next 90 days, confirm whether Sweetwater 1 actually energized at the full 1.4 GW after its April 2026 milestone. Track Sweetwater 2's path toward its 2027 date (31053) and whether the liquid-cooled expansion beyond 50 MW (31051) has a stated size and date. Check whether the GPU fleet has moved past 1,896. Finally, see whether Kiowa, Oklahoma, Bundey and Redbank acquire capacity figures and queue positions or stay as placeholders.

IREN's trajectory is a pivot followed by a scale-up. Until 2024 the company framed capacity in hashrate, and those targets climbed steadily: 9.1 and 9.4 EH/s in 2023 (12465, 12138), 10 EH/s by May 2024 (6150), 31 EH/s for Q4 2024 (5691), and an installed 37 EH/s by April 2025 (31043). Over the same period the unit of ambition moved to megawatts. Childress went from a 100 MW Phase 1 minimum (31139) to 600 MW of potential (12176) to a 500 MW energized site (31048). The West Texas footprint grew to 2.75 GW of secured power (31046), anchored by the 1.4 GW Sweetwater 1 first floated in November 2023 (6523). The GPU fleet grew more modestly, from 568 units (6261) to 1,896 (5712, 31047). Most of the growth is new projects stacked on top of earlier ones, not existing figures being revised upward. There is no visible case of a headline number being quietly cut. One exception needs reconciling: the 2021 target of 15.2 EH/s by September 2023 (14393) is marked verified, yet IREN's own mid-2023 statements put operating capacity near 5.5 EH/s (12689, 12712). Either the verification or the date mapping is wrong.

Dated promises have generally held. The 80 MW Childress expansion due by mid-2024 (6495, 6410), the 2024 50 MW build (6153), Phase 5 by June 2025 (31049), Horizon 1 by year-end 2025 (31050), and Sweetwater 1 energization by April 2026 (31052) are all verified. That record is strong, but energization is not delivery. Sweetwater 1 is still listed as under construction, so the verified milestone means grid power reached the site, not that 1.4 GW of load is running. The 2022 Childress commercial-operations deadline (31141) never carried a date, so it cannot be scored.

The record contains no independent corroboration in the form of permits, ERCOT interconnection queue entries, or site reporting. The verification flags are the only outside check, and the sources behind them are not shown. The Sweetwater 2 grid agreement (31045) in particular should be matched against ERCOT queue data before its 2027 energization (31053) is treated as firm. The supplied record also ends at an April 2025 entry (30435), so any capacity claims made since then are not assessed here.

Over the next 90 days, watch for disclosed operating load or a named customer at Sweetwater 1, ERCOT milestones for Sweetwater 2, and Horizon phases beyond the first. Also watch whether Kiowa, the Oklahoma campus, Bundey and Redbank receive their first dated capacity claims. Those projects are listed with no figures at all, which makes them the least accountable part of the portfolio.

IREN's record reads as two distinct growth curves stitched together. Through 2023 the hashrate story stalled: operating capacity sat at 5.5 EH/s across at least six restatements from February to April 2023 (13337 through 12689). The original 15.2 EH/s target for September 2023 (14393) was effectively reset to 9.1 EH/s (12448) and then 9.4 EH/s (12138) without a clear public reconciliation. That claim's verified tag sits uneasily beside the company's own interim figures and deserves re-examination. From 2024 the cadence changed. The 20 EH/s installed base (5689) became 31 EH/s by year-end (5691), then 37 (31043), 40 (30434), and a 50 EH/s target for mid-2025 (30375), which is recorded as met. Power ambitions scaled faster still. Childress went from 600 MW potential (12176, 12186) to roughly 500 MW energized plus a 150 MW Phase 5 (31048, 31049). The West Texas pipeline grew from 1.4 GW at Sweetwater 1 (6523) to 2.75 GW secured (31046) and a 2 GW hub framing (30386). GPU capacity grew from 568 H100s (6261) to 1,896 units (5712) and then held flat into April 2025 (31047), even as the AI pivot moved to the center of the narrative.

On dates, the record since late 2023 is notably clean. Childress 80 MW (6495, 6410), the 2024 GPU deliveries (6262, 5712), Phase 5 (31049), Horizon 1 (31050) and Sweetwater 1 energization by April 2026 (31052) are all marked verified. Sweetwater 2's late-2027 date has been repeated consistently (31053, 30385).

The weak point is independent corroboration. Apart from a street address for Sweetwater 1, nothing in this record cites permits, ERCOT interconnection queue entries, or site reporting. The 600 MW Sweetwater 2 grid agreement (31045) rests on company disclosure alone. Kiowa, the Oklahoma campus, Bundey and Redbank carry no capacity or cost figures at all. The record supplied here also ends mid-entry at claim 30380, so later claims may already settle some of these questions.

Over the next 90 days, watch for the fiscal Q1 2027 results. The questions to ask are how much of Sweetwater 1's 1.4 GW is actually drawing load rather than merely energized, and whether Horizon phases beyond 50 MW are being commissioned. Also check whether the GPU fleet figure finally moves materially past 1,896. Then look for ERCOT or Oncor filings that corroborate Sweetwater 2's schedule, and for any first capacity numbers on the four announced but unquantified sites.

IREN's capacity narrative has grown in steps, and the record shows the early targets were reset rather than met. The November 2021 plan for 15.2 EH/s by September 2023 (14393) collides with the company's own April 2023 operating figure of 5.5 EH/s (12784, 12712, 12689). The 'verified' tag on 14393 therefore looks like a labeling problem worth auditing. After that, the goalposts moved upward roughly every two quarters: 9.1 EH/s tied to an 80MW build (12448), then 9.4 (12138), 10 (6341), a 20 EH/s pathway (6342), 20 installed by September 2024 (5689), 31 for Q4 2024 (5691), 37 and 40 in April 2025 (31043, 30434), and 50 by mid-2025 (30375). The later hashrate targets appear to have landed close to schedule. The earlier miss was replaced with new numbers and never acknowledged. Power ambitions scaled the same way, from 600MW of Childress potential (12176) to a 1,400MW Sweetwater 1 (6523) and 2.75GW of secured West Texas power (31046). The GPU story moved more slowly: 568 units (6262) became 1,896 (5712), and that figure was unchanged in April 2025 (31047).

On timelines, the 80MW Childress expansion due mid-2024 (6495, 6410) and the 2024 50MW build (6153) carry verified tags. Childress Phase 5 (31049, 30380) was dated to June 2025. The keystone date is Sweetwater 1 energization in April 2026 (31052, 30384). That deadline passed six months ago, yet the project is still listed as under construction, and the claim record ends in May 2025 with nothing confirming or revising the date. Horizon 1's end-2025 commissioning (31050) is in a similar position: the site is only partially operational, and there is no follow-up claim.

This record contains no independent corroboration of any kind: no ERCOT interconnection queue entries, transmission filings, county permits, or site reporting. The Sweetwater and Childress megawatt figures therefore rest on company statements alone. The verified tags cover delivery outcomes, not the underlying grid capacity.

Over the next 90 days, the priority is to establish whether Sweetwater 1 has energized, either fully or in part. The best sources are ERCOT generation and load interconnection status, Oncor or AEP transmission dockets, and IREN's next quarterly filing. A further check is whether Horizon's liquid-cooled capacity beyond 50MW (31051) has a dated commitment. Any restatement of Sweetwater 2's 2027 energization (31053) would be the first test of whether IREN acknowledges slippage openly or simply resets the target, as it did in 2023.

IREN's capacity narrative has scaled in two distinct phases, and the second has largely displaced the first. The early mining roadmap was the weaker of the two. The 15.2 EH/s target for September 2023 (claim 14393) is tagged verified, yet mid-2023 operating figures sat at 5.5 EH/s (12689), and the October 2023 target was a more modest 9.4 EH/s (12138). That looks like a quiet reset rather than delivery. The verification label on 14393 deserves scrutiny, because the record contains no explicit acknowledgement of the shortfall. From 2024 onward, the hashrate trajectory tightened considerably. It ran from roughly 20 EH/s installed (5689), to a 31 EH/s Q4 2024 target that was met (5691), to 40 EH/s by spring 2025 (30434, 29796), with 50 EH/s by mid-2025 marked verified (30375).

The power-capacity story has inflated more than it has slipped. Childress went from 600 MW potential (12176) to a 750 MW full-build framing tied to a second bulk transformer (29799). The West Texas site was announced at 1,400 MW (6523) and has since been reframed as a 2 GW Sweetwater hub (30386, 29805). That hub figure is simply Sweetwater 1 plus a 600 MW Sweetwater 2 slated for late 2027 (30385). The discrete Childress phases (6495, 6153, 30380) are all tagged verified, which suggests the company executes well on incremental builds at a site it already controls.

Horizon 1 is the one unresolved conflict in the record. The same Q4 2025 commissioning is tagged delayed in 30381 and verified in 29800, eight days apart. That inconsistency should be reconciled against primary sources before either label is cited.

Independent corroboration is thin. The record contains no permits, no ERCOT large-load interconnection entries, and no site reporting, so the verified tags rest on company disclosures. The claim history also stops in May 2025, which leaves a seventeen-month gap. Kiowa, the Oklahoma campus, Bundey and Redbank appear as announced projects with no capacity or timeline claims at all.

Over the next 90 days, the first priority is confirmation that Sweetwater 1 actually energized around April 2026 (30384), ideally from ERCOT or Oncor filings rather than company statements. After that, watch for a settled Horizon 1 commissioning date, evidence that the Childress second transformer arrived and lifted the site past 500 MW, and whether any of the four announced sites acquires a megawatt figure or an interconnection position. Any upward restatement of the Sweetwater total before Sweetwater 1 is visibly loaded should be read as narrative expansion, not delivery.

IREN's capacity story is one of steady upward restatement, though the early record overpromised. The November 2021 target of 15.2 EH/s by September 2023 (14393) is tagged verified, yet the company's own run of statements through spring 2023 repeated a 5.5 EH/s operating figure (12784, 12712, 12689). That tag deserves scrutiny, because the plateau suggests the original pace was not met on schedule. From mid-2023 onward the cadence becomes more credible: 9.1 EH/s tied to the 80 MW Childress build (12465, 12444), 10 EH/s by May 2024 (6150), 20 EH/s installed by September 2024 (5689), a Q4 2024 step to 31 EH/s (5691), 40 EH/s by April 2025 (30434), and a 50 EH/s mid-2025 target recorded as met (30375, 29814). Power ambitions have grown faster than hashrate. Childress moved from a 600 MW potential (12186, 12176) to 750 MW capex framing (29828), and Sweetwater grew from a 1.4 GW West Texas site (6523) into a 2 GW hub with a second 600 MW phase (30386, 30385).

Near-term dates have mostly held. The Childress 80 MW expansion (6495, 6410), the H100 commissioning (6262), the H200 purchase (5711) and Phase 5 construction (30380) all carry verified tags. The exceptions matter more than the successes because they sit on the AI-facing projects that drive the current narrative. Horizon 1's H2 2025 commissioning is marked delayed (30381). Sweetwater 1's April 2026 energization is flagged delayed in two filings (29839, 29804) but verified in a third (30384). The record does not resolve that inconsistency, and it is not clear the company acknowledged a slip.

Independent corroboration is thin in this record. No permits, ERCOT interconnection queue entries or on-the-ground reporting are attached, so most large-MW claims remain company-stated. The Sweetwater substation timeline in particular should be checked against ERCOT and transmission-provider filings rather than accepted from investor materials.

Over the next 90 days, three things matter most. The first is whether IREN states plainly that Sweetwater 1 is energized and at what load. The second is whether Horizon reaches full commissioning or receives a new date. The third is whether the late-2027 Sweetwater 2 target (29840) and the 750 MW Childress envelope are reaffirmed, trimmed or quietly dropped from guidance.

IREN's capacity story has expanded in two distinct phases, and the second is far larger than the first. Through 2023 the narrative was hashrate. The November 2021 target of 15.2 EH/s by September 2023 (claim 14393) gave way to a stepwise ladder of 5.5 EH/s (12784), 9.1 EH/s (12443), 9.4 EH/s (12138) and 10 EH/s (6522). That ladder later accelerated to 20, 31, 40 and 50 EH/s targets (5689, 5691, 30434, 30375). The power-envelope story then overtook it. Childress went from 600 MW potential (12176) to an implied 750 MW once a second bulk transformer arrives (29799). Sweetwater went from a 1.4 GW West Texas development (6523) to a 2 GW hub split across two sites (30386, 29805). Announced scale has only grown. No project on record has been resized downward.

Date discipline is mixed. The 2021 hashrate promise is tagged verified, but the company's own operating figures of roughly 5.5 EH/s in April 2023 (12712) sit well short of 15.2 EH/s ahead of the September 2023 deadline. That tag deserves scrutiny rather than acceptance. The Childress 80 MW and 50 MW phases (6410, 6153) and the 50 EH/s target (30375) appear to have landed on schedule. Horizon 1 is the clearest internal inconsistency: the same H2/Q4 2025 commissioning is recorded as delayed in one filing (30381) and verified in another (29800), and Horizon remains only partially operational. Sweetwater 1's April 2026 energization (30384) is tagged verified, yet the project is still listed as under construction six months past that date. This suggests energization of the substation rather than delivery of usable data hall capacity, a distinction the company has not drawn explicitly.

Independent corroboration is thin in this record. Nothing here cites ERCOT interconnection queue entries, county permits or on-the-ground reporting for Sweetwater, and there are no capacity claims at all after May 2025. That leaves roughly seventeen months of trajectory unsupported by the dataset. The announced Kiowa, Oklahoma, Bundey and Redbank sites carry no capacity, cost or date claims whatsoever.

Over the next 90 days, watch for three things. First, whether IREN reports energized megawatts at Sweetwater 1 in distinction from contracted or substation capacity. Second, whether the second Childress transformer is installed and the 750 MW figure appears in operating rather than potential terms. Third, whether Sweetwater 2's late-2027 target (30385) is restated or quietly shifted. ERCOT queue updates and Nolan County filings are the most direct external checks available.

IREN's record traces a company whose announced capacity has grown repeatedly and has not been cut back. The early hashrate story moved from a 15.2 EH/s target (14393) through stepwise operating milestones of 2.3, 3.7 and 5.5 EH/s (13824, 13604, 12784), then 10 EH/s (6150), 31 EH/s by late 2024 (5691) and a 50 EH/s mid-2025 target (30375). Power capacity grew the same way. Childress went from 600 MW of potential (12176) to 750 MW backed by a second bulk transformer (29799). The West Texas build started as 1,400 MW (6523) and is now framed as a 2 GW Sweetwater hub (30386, 29805), with Sweetwater 2 adding 600 MW for late 2027 (30385). None of these upward revisions came with a disclosed reason beyond demand, which matters given the later pivot toward AI hosting at Horizon.

On dates, the record is mostly favorable. The Childress 80 MW expansion (6495, 6410), the 2024 50 MW build (6153), GPU commissioning (6262) and the Sweetwater 1 April 2026 energization (30384) are all tagged verified. There are two exceptions. First, the 15.2 EH/s target for September 2023 (14393) carries a verified tag, yet IREN's own mid-2023 statements describe a 9.1 EH/s expansion path (12443). That tag deserves a second look rather than being taken at face value. Second, Horizon 1 is listed as delayed in one entry (30381) and as verified for Q4 2025 delivery in a near-identical entry eight days later (29800). Until that conflict is resolved, Horizon's timeline should be treated as unsettled.

Independent corroboration is the weakest part of the file. The claims cite BC Hydro connections (12198) and substation works at Sweetwater (29805), but the record contains no ERCOT interconnection queue entries, county permits or site reporting for the Texas expansions. There are also no capacity claims at all for Kiowa, the Oklahoma campus, Bundey or Redbank. The most recent claim dates to May 2025, so the file has gone roughly seventeen months without an update while the company's largest commitments were supposedly being delivered.

Over the next 90 days, the first thing to watch is what IREN has actually energized at Sweetwater 1 versus the full 1,400 MW, and whether that is described as power available or load drawn. Beyond that, look for evidence that Childress has reached 750 MW, a clear account of Horizon capacity beyond the first 50 MW, and any first site-level numbers, permits or queue filings for the four announced-only projects.

IREN's headline numbers have grown steadily, but the record shows targets being quietly reset rather than met on the original schedule. The November 2021 plan for 15.2 EH/s by September 2023 (claim 14393) carries a verified tag, yet the company's own filings through April 2023 describe roughly 5.5 EH/s operating (12784, 12689). The next targets were 9.1 EH/s (12443), then 9.4 EH/s (12138), then 10 EH/s (6522), which was reached only by May 2024 (6150). In effect the 2021 goal arrived about a year late, and no claim in the record says it was missed. After that point, delivery speeds up and becomes more believable. The company reported 20 EH/s (5689), a 31 EH/s Q4 2024 target (5691), and 40 EH/s by April 2025 (30434). Unlike the 2021 goal, these milestones line up with dated equipment purchases (6490, 6407) and dated Childress construction (6495, 6153), which makes the hashrate story more credible from 2024 onward. Power capacity has expanded in a similar way. Childress started at 600 MW (12176) and later became 750 MW once the second transformer was added (29799). Sweetwater started as 1.4 GW (6523) and is now a 2 GW hub (30386, 29805). The 80 MW Childress expansion did meet its June 2024 deadline (6410). The AI pivot is the weakest part of the record. Horizon 1 carries two conflicting statuses: claim 30381 marks commissioning as delayed, while 29800 marks the same Q4 2025 delivery as verified. That conflict needs to be resolved before Horizon is counted as delivered capacity. Several sites, including Kiowa, Oklahoma, Bundey and Redbank, sit on the books with no stated capacity, cost or date. The record has no independent corroboration of any kind: no ERCOT interconnection queue entries, no Oncor or county permits, and no site reporting. Every capacity figure here is the company's own statement. The record also ends in May 2025, so the past seventeen months are unobserved. For the next 90 days, the first priority is Sweetwater 1. Its promised April 2026 energization (30384) is now six months past, so check ERCOT and utility records for proof it actually happened. Second, look for a clear Horizon 1 commissioning statement with a named customer and the contracted megawatts. Third, look for substation progress at Sweetwater 2 that supports its late-2027 date (30385). Finally, watch whether mining hashrate is flat or falling while megawatts move to GPU hosting; that would show where the capacity is actually going.

IREN's announced footprint has expanded steadily, mostly by adding new targets rather than walking old ones back. Hashrate guidance climbed from 2.3 EH/s (13824) through a long plateau at 5.5 EH/s in early 2023 (12784, 12689), then to 10 EH/s (6150), 20 EH/s (5689), 31 EH/s (5691), 40 EH/s (30434) and a 50 EH/s mid-2025 target (30375, 29814). Power claims grew faster. Childress went from 600 MW of potential (12186) to a 750 MW capex frame (29828). The West Texas site went from 1.4 GW (6523) to a 2 GW Sweetwater hub once Sweetwater 2's 600 MW was added (30386, 30385). The GPU line grew from 568 H100s (6261) to 1,896 units (5712). These are additive expansions, not restatements.

Date performance divides by asset type. Near-term mining and Childress build-outs mostly landed on schedule: the 80 MW expansion (6495, 6410), the 50 MW 2024 build (6153), Phase 5 (30380) and the 31 EH/s and 50 EH/s targets are all marked verified. The exception is the earliest one. The 15.2 EH/s target for September 2023 (14393) is tagged verified even though the company was reporting about 5.5 to 8.3 EH/s around that date (12712, 6493). That tag deserves a re-check. The AI and grid-scale milestones are weaker. Horizon 1's H2 2025 commissioning is marked delayed (30381). Sweetwater 1's April 2026 energization is tagged verified in 30384 but delayed in 29839 and 29804, so the record contradicts itself on the company's single most important milestone. The supplied record also ends mid-2025, partway through claim 29841. Any later acknowledgement or revision of these dates is not visible here.

The record contains no independent corroboration: no ERCOT interconnection queue entries, Oncor or transmission filings, county permits, or site reporting for Sweetwater, Horizon or the newer announced sites (Kiowa, the Oklahoma campus, Bundey, Redbank). Every capacity figure here comes from the company, and the verified tags apply to dates, not to independent checks of megawatts actually delivered.

Over the next 90 days, the first priority is settling Sweetwater 1. April 2026 has passed, so either an energization notice exists or a revised date should be disclosed. ERCOT or utility records should show which. Next is whether Horizon's commissioning is complete and running AI workloads under contract, rather than simply being energized. Then come early-stage permitting or queue filings for Sweetwater 2 that are consistent with a late-2027 date (29840). Finally, watch whether any of the four capacity-less announced sites receives a megawatt figure or a date.

IREN's announced footprint has expanded in steps, and the targets have generally been revised upward rather than quietly reduced. Childress was first described as a 600MW site (12176, 12186). Later filings framed it as 500MW of additional capacity (6412), then 500MW fully operational across Phases 1-4 (30379), and most recently a 750MW ceiling contingent on a second bulk transformer (29799). The West Texas project shows a similar pattern. It was introduced at 1,400MW (6523) and later reframed as a 2GW Sweetwater hub (30386, 29805), with a 600MW second phase targeted for late 2027 (30385). Hashrate guidance followed the same arc, from 10 EH/s (6522, 6341) through 20 and 31 EH/s (5689, 5691) to a 50 EH/s mid-2025 target (30375).

Delivery has been more uneven than the verification tags suggest. The original 15.2 EH/s target for September 2023 (14393) carries a verified status. Yet the company's own disclosures put operating capacity at 5.5 EH/s through spring 2023 (12689) and 8.3 EH/s in November 2023 (6493). That gap deserves reconciliation rather than a clean pass. Later milestones held more consistently: the Childress 80MW expansion (6495, 6410), the 2024 50MW build (6153), and the Q4 2024 31 EH/s step (5691). Horizon 1 is the clearest unresolved item. The same H2 2025 commissioning promise is recorded as delayed in one claim (30381) and verified in another (29800), and the record does not explain the discrepancy. Sweetwater 1's April 2026 energization (30384) is marked verified. That milestone is grid energization, not live compute load.

The record contains no independent corroboration. It includes no ERCOT interconnection queue entries, county permits, or third-party site reporting, and the substation and fiber-loop statements for Sweetwater (29805, 29806) are company-sourced. The claim history also ends in May 2025. That leaves roughly seventeen months without logged guidance, an unusually long silence for a company that previously updated capacity monthly.

Over the next 90 days, four things matter. First, whether IREN discloses energized-versus-utilized megawatts at Sweetwater 1, and whether a named offtaker is attached. Second, a definitive statement on Horizon 1's commissioning date. Third, ERCOT queue or transmission filings that would support the late-2027 Sweetwater 2 timeline. Fourth, confirmation that the Childress transformer has arrived, since the 750MW figure depends on it.

IREN's record shows a company whose headline numbers have grown faster than its disclosure discipline. In bitcoin mining, the November 2021 target of 15.2 EH/s by September 2023 (14393) is marked verified. Yet the company's own operating figures through spring 2023 sat at 5.5 EH/s (12784, 12712, 12689), and 10 EH/s was not stated until May 2024 (6150). That verification deserves a second look. After that, the hashrate climb was steep and largely delivered: 20 EH/s installed by September 2024 (5689), the Q4 2024 step to 31 EH/s (5691) and the 50 EH/s mid-2025 target (29814) both verified. Power ambitions grew in parallel. Childress was described as a 600MW site in September 2023 (12186), carried an additional 500MW framing that December (6412), and moved to 650MW (29827, 29833, verified) and then a 750MW liquid-cooled design (29820, 29828). West Texas went from a 1.4GW development (6523) to a 2GW Sweetwater hub (29841), split into 1.4GW and 600MW phases (29822, 29823). Alongside this, the business pivoted toward AI compute. The GPU fleet went from 568 H100s (6261) to 1,896 units (5712), and the Horizon 1 50MW liquid-cooled build for Q4 2025 (29800) is verified.

Shorter-dated promises have generally held: the 80MW Childress expansion (6410), the 2024 GPU deliveries (6262), and the 50MW Childress build (6153). The clear exception is the project that matters most. Sweetwater 1's April 2026 energization is marked delayed across three separate claims (29804, 29822, 29839). The record contains no later company statement giving a revised date. As of mid-October 2026 that is a roughly six-month slip without a logged acknowledgement. The claim record also goes quiet after May 2025, which is itself notable for a company with this cadence of capacity announcements.

Independent corroboration is thin. The record contains no permit filings, ERCOT interconnection queue entries, or on-the-ground reporting for Sweetwater or the Childress expansions. The only external signals are status labels, with Sweetwater 1 listed as under construction at a specific street address. The announced pace therefore rests almost entirely on company statements and on IREN's earlier execution at smaller scale.

Over the next 90 days, the key question is whether IREN publicly restates a Sweetwater 1 energization date and attributes the slip to the substation, the utility, or equipment. Watch also for ERCOT or Oncor records confirming interconnection progress, for whether the late-2027 Sweetwater 2 date (29840) survives any Sweetwater 1 rescheduling, and for evidence that Horizon capacity beyond the first 50MW is actually contracted rather than designed.

IREN's announced footprint has expanded in nearly every dimension, and the denominator has shifted from hashrate to megawatts. Hashrate guidance climbed in steps: 9.1 EH/s (12448), 10 EH/s (6341) with a path to 20 (6342), 31 EH/s for Q4 2024 (5691), and 50 EH/s by mid-2025 (29814). The power story grew in parallel. Childress went from 600 MW of potential capacity (12176) to 650 MW (29827) and then to a 750 MW liquid-cooling design (29820, 29828). The West Texas site was announced at 1.4 GW (6523) and is now framed as a 2 GW Sweetwater hub (29841, 29840). Nothing has been formally shrunk, but the near-term liquid-cooled figures at Childress do not agree with each other: up to 50 MW of IT load in one claim (29837) and an initial 75 MW deployment in another (29838). That should be resolved before either number is treated as a commitment.

On dates, the record splits cleanly by horizon. Commitments within about twelve months have consistently been verified. These include the 80 MW Childress expansion (6495, 6410), the 2024 Childress build (6153), the GPU deliveries and fleet size (6262, 5711, 5712), the 31 EH/s and 50 EH/s targets, and Horizon 1's 50 MW by Q4 2025 (29800). The one large-scale, long-dated promise has not held. Sweetwater 1 energization for April 2026 was stated three times in the same disclosure (29804, 29822, 29839) and is now marked delayed. The claims on record contain no restated date or explanation, so as of mid-October 2026 the slip appears unacknowledged in this dataset. The record also ends in May 2025, so later disclosures may exist that have not been ingested.

Independent corroboration is thin. The verification flags come from the company's own subsequent reporting. The record contains no permit filings, ERCOT interconnection queue entries, or site reporting for Sweetwater or the Childress liquid-cooling conversion. The near-term delivery history is credible, but the gigawatt-scale pace is effectively unverified.

Over the next 90 days, the key question is whether IREN publishes a revised Sweetwater 1 energization date and says what caused the delay, whether at the substation, in the interconnection, or in equipment. The company's next quarterly update should also show whether Sweetwater 2's late-2027 date (29823) is being quietly re-baselined. A third thing to watch is evidence that Childress liquid-cooled capacity is moving beyond Horizon 1, particularly in ERCOT queue status and any contracted HPC tenant that would anchor the 750 MW design.

IREN's capacity narrative has scaled by orders of magnitude, and its unit of measure has changed along the way. Early claims were about hashrate. Operating capacity was 2.3 EH/s in August 2022 (13824) and plateaued at 5.5 EH/s through spring 2023 (12827, 12689). It then reached 10 EH/s in May 2024 (6150), 20 EH/s by September 2024 (5689), and 40 EH/s by May 2025 (29796), with a 50 EH/s mid-2025 target marked verified (29795). Since 2023 the headline unit has shifted to megawatts. Childress grew from 600 MW of potential (12176) to a 750 MW liquid-cooling redesign (29820, 29803), and the West Texas site grew from 1,400 MW (6523) to a 2 GW Sweetwater framing (29805). That growth in site scope is real on paper, but much of it is design and substation work rather than energized load.

Date discipline is mixed. Discrete, near-term Childress commitments held: the 80 MW expansion due mid-2024 (6495, 6410), the 2024 50 MW build (6153), the 31 EH/s Q4 2024 target (5691) and the H200 deployment (5711, 5712) are all marked verified. The earliest headline target is less clean. The 15.2 EH/s by September 2023 claim (14393) is marked verified, yet the company's own contemporaneous figures show 8.3 EH/s in late November 2023 (6493) and a 9.4 EH/s target (12138). That label deserves re-examination rather than acceptance. The most consequential date, Sweetwater 1 energization by April 2026 (29822), is now flagged delayed. The record contains no company claim acknowledging or re-dating that slip, which is a revision by silence. The claim record also stops at May 2025, roughly seventeen months ago.

The record contains no independent corroboration of the announced pace. There are no ERCOT large-load interconnection entries, no utility or substation filings, no county permits, and no on-the-ground reporting. The Sweetwater 1 street address and its under-construction status are the only physical anchors. Kiowa, the Oklahoma campus and Bundey carry no capacity or timeline claims at all.

Over the next 90 days, the key question is whether IREN publicly confirms a revised Sweetwater 1 energization date, and whether ERCOT or transmission-provider filings back it. Also watch for evidence that the initial 75 MW liquid-cooled deployment at Childress (29838) is drawing load, any refreshed hashrate disclosure that clarifies whether mining is being wound down in favor of AI capacity, and whether the Sweetwater 2 late-2027 date (29823, 29840) is restated or quietly extended.

IREN's record shows two expansions running in parallel. One is a bitcoin hashrate ramp that has largely delivered. The other is a power-and-AI build-out whose largest milestone has now slipped.

On capacity, the hashrate story is one of steady upward revision. Operating capacity sat at 5.5 EH/s through spring 2023 (12827, 12689). Targets then stepped through 9.1 (12448), 9.4 (12138), 10 (6341) and a 20 EH/s pathway (6342), and the company reported 40 EH/s installed with a 50 EH/s June 2025 target (29796, 29795). The power envelope grew faster still. Childress went from 600 MW (12176) to a 750 MW liquid-cooling redesign (29820, 29803). The 1,400 MW West Texas figure (6523) became a 2 GW Sweetwater complex split 1,400/600 MW (29805, 29822, 29823). Alongside this, the GPU narrative scaled from 568 H100s (6261) to 1,896 units (5712, 29807), plus initial liquid-cooled deployments of 50 to 75 MW at Childress (29837, 29838).

On dates, the dated near-term promises have mostly held. Childress 80 MW by mid-2024 (6495, 6410), the 2024 Childress build (6153), 31 EH/s in Q4 2024 (5691) and 50 EH/s by June 2025 (29795) are all marked verified. One caveat: claim 14393 carries a verified tag against a 15.2 EH/s September 2023 target, yet reported operating capacity was 8.3 EH/s in November 2023 (6493). That tag deserves re-examination. The significant miss is Sweetwater 1, whose April 2026 energization (29822) is now flagged delayed. The record contains no company statement acknowledging or re-dating it. Several items, including the 100 MW Childress early works (6496, 6413) and the Phase 5 150 MW block (29798), never carried firm dates.

On independent evidence, this record contains none: no permits, ERCOT or Oncor interconnection queue entries, or site reporting. Verified tags confirm reported outcomes, not physical pace. The claim stream also stops in May 2025, which leaves a roughly seventeen-month gap through the present.

Over the next 90 days, the key question is whether IREN formally re-dates Sweetwater 1 or confirms partial energization. Its fiscal first-quarter results, typically due in November, are the natural venue. Interconnection status in ERCOT filings would corroborate either answer. Further signals are the share of Childress actually converted to liquid cooling versus design-stage claims, and whether the late-2027 Sweetwater 2 date (29840) is quietly moved in step with Sweetwater 1.

IREN's record is one of steady upward revision, and the near-term promises have mostly been kept. The headline unit has also changed, from exahashes to megawatts and GPUs. The hashrate path runs from 5.5 EH/s of operating capacity in early 2023 (12827, 12712) through 10 EH/s in May 2024 (6150), 20 EH/s installed by September 2024 (5689), a verified 31 EH/s by year-end 2024 (5691), and a verified 50 EH/s by June 2025 (29795, 29814). Power ambitions grew in parallel. Childress went from 600 MW of potential (12176) to 650 MW (29827, verified) and then to a 750 MW liquid-cooling redesign (29803). The West Texas site grew from 1.4 GW (6523) into a 2 GW Sweetwater hub (29841). One oddity deserves a footnote: the 2021 target of 15.2 EH/s by September 2023 (14393) is marked verified, yet the company's own mid-2023 statements put capacity at 5.5 to 9.1 EH/s (12689, 12448). That verification looks generous. On dates, the Childress build-outs held. The 80 MW expansion by mid-2024 (6410, 6495), the 50 MW added in 2024 (6153), and Horizon 1's 50 MW liquid-cooled hall by Q4 2025 (29800) are all verified, as are the GPU deliveries (6262, 5712). The clear break is Sweetwater 1. The April 2026 substation energization (29804, 29839) is now flagged delayed, and the site still sits at under construction six months past that date. The supplied record ends in May 2025 and contains no later claim that resets the date, so the slip appears to have gone unacknowledged in the claims captured here. That gap matters because Sweetwater 1 is the single largest block of promised capacity. Independent corroboration is thin. The record carries no permits, no ERCOT interconnection queue entries, and no site reporting, so the pace at Sweetwater rests on company statements alone. Over the next 90 days, the first thing to watch is whether IREN publishes a revised Sweetwater 1 energization date and explains the delay, ideally alongside evidence of substation completion from ERCOT or Oncor filings. Next, check whether Sweetwater 2's late-2027 date (29840) is quietly moved to match. The Childress second bulk transformer (29799) is the gating item for the full 750 MW and should show up in delivery updates. The announced Kiowa, Oklahoma and Bundey projects also still lack any capacity figures. Concrete megawatt numbers or interconnection filings for those sites would be the first sign that they are more than options.

IREN's record shows two distinct growth phases, and the scale-up is real in both. In mining, stated operating capacity sat between 2.3 and 5.5 EH/s through mid-2023 (13824, 13604, 12689). It then stepped to 10 EH/s (6150), 20 EH/s installed (5689), the 31 EH/s Q4 2024 target (5691) and the 50 EH/s mid-2025 target (28860, 28347), with each later step marked verified. On power, Childress grew from 600 MW of potential (12176) to a 750 MW full-build framing (28866). The portfolio headline reached 2,910 MW grid-connected, with 660 MW operating in June 2025 (28868) and 810 MW a month later (28327). The GPU fleet went from 568 H100s (6261) to roughly 4.3k (28343), and Prince George is pitched as able to host more than 20k Blackwells (28344). One early entry deserves skepticism: the 2021 promise of 15.2 EH/s by September 2023 (14393) carries a verified tag, yet the company's own statements put operating capacity at 5.5 EH/s through that summer and still described 9.4 EH/s as a target in October 2023 (12138). The 14.5 EH/s Bitmain contract behind it (14392) remains unverifiable. That label should be rechecked rather than relied on.

On dates, the Texas buildout has largely held. The 80 MW Childress expansion (6495, 6410), the 2024 50 MW build (6153) and the 150 MW Phase 5 energization (28862) are all marked delivered on schedule. Sweetwater 1's April 2026 energization (28863) is also tagged verified. Horizon 1 has conflicting status, verified under 28349 but not yet verifiable under 28861, and the two need reconciling. Sweetwater 2's late-2027 date has been repeated without revision (28864, 28332). Kiowa, the Oklahoma campus and Bundey have no capacity or date claims on record.

The main weakness is the absence of independent evidence. The record contains no permits, no ERCOT or Oncor interconnection queue entries and no site reporting. The verified tags appear to rest on later company disclosures, not third-party confirmation. The claims also stop in July 2025, about fifteen months before this read.

Over the next 90 days, watch four things. First, whether Sweetwater 1 is carrying IT load or is only energized. Second, Horizon tenant and capacity disclosures that settle the 28349 and 28861 conflict. Third, ERCOT queue progress that supports or undercuts Sweetwater 2's late-2027 date. Fourth, whether any of the dormant announced sites acquire dated, sized commitments.

IREN's trajectory runs in two arcs. The bitcoin-mining arc was erratic early and disciplined later. The November 2021 plan for 15.2 EH/s by September 2023 (claim 14393) sits awkwardly beside the record that followed: operating capacity was restated at 5.5 EH/s at least six times between February and April 2023 (claims 13337, 12784, 12827, 12712, 12699, 12689). The goal was then reset to 9.1 EH/s (12448) and trimmed again in October 2023 to 7.0 EH/s, with 9.4 EH/s as a stretch target (12137, 12138). The verified tag on 14393 deserves re-examination, because the operating figures in the same record do not support 15.2 EH/s by that deadline, and the intervening resets were not framed as a retreat. From 2024 the pattern tightens. The 10 EH/s target (6341) was met by May 2024 (6150), 31 EH/s was guided for Q4 2024 (5691), and 50 EH/s was achieved on its June 2025 date (28860, 28347).

The power arc has expanded steadily rather than shrunk. Childress grew from a 600 MW potential site (12176) to 650 MW operating with a path to 750 MW (28348, 28866). Sweetwater was introduced at 1,400 MW in November 2023 (6523), and the portfolio now totals 2,910 MW of grid-connected power (28868). Alongside this, the company pivoted from 568 H100s (6261) to a fleet of roughly 4,300 GPUs (28343), with Prince George framed as able to host more than 20,000 Blackwells (28344). Near-term dates have mostly held: the Childress 80 MW and Phase 5 milestones landed as promised (6410, 28862). One inconsistency matters, though. Horizon 1's Q4 2025 delivery is graded not_yet_verifiable in 28861 but verified in 28349, and the project is still listed as only partially operational.

The supplied record contains no independent corroboration, such as ERCOT queue entries, county permits, or site reporting. It also ends in July 2025, partway through a claim. As a result, the April 2026 energization of Sweetwater 1 (28863) cannot be confirmed from this data, even though that date has now passed and the project remains under construction. Kiowa, Oklahoma and Bundey carry no stated capacity or dates.

Over the next 90 days, the key question is whether IREN's fiscal first-quarter disclosures explicitly confirm that Sweetwater 1's substation was energized and how much load it carries. Restated capacity without that confirmation would not settle it. Also worth watching are whether the late-2027 date for Sweetwater 2 (28864) holds, whether Horizon reaches full operation, and whether the announced-only sites gain interconnection filings to substantiate them.

IREN's record traces a company that has scaled its ambitions in two distinct registers and largely delivered against the dated ones. On the mining side, operating hashrate climbed from roughly 2.3 EH/s (13824) through the 5.5 EH/s plateau of spring 2023 (12784, 12712) to 20 EH/s (5689). The company then guided to 31 EH/s by end-2024 (5691) and 50 EH/s by mid-2025 (28860), and it reports reaching the latter (28347). Power ambitions grew faster still. Childress went from 600 MW of potential (12176) to a full 750 MW build-out backed by a second transformer (28866). The 1,400 MW West Texas site first flagged in late 2023 (6523) became Sweetwater 1, with a 600 MW Sweetwater 2 behind it (28864). Group-wide secured grid power reached 2,910 MW (28868), with operating capacity reported rising from 660 MW to 810 MW within a month (28868, 28327). Running alongside this is a GPU pivot, from 568 H100s (6261) to about 4.3k units (28343), with Prince George framed as able to host more than 20k Blackwell GPUs (28344). Nothing in the record shows announced capacity being quietly shrunk. Each restatement has been upward. Dated promises have mostly held. The Childress 80 MW expansion (6495, 6410), the 2024 50 MW tranche (6153), Phase 5 (28862) and Sweetwater 1's April 2026 energization (28863) are all marked verified. Two loose ends stand out. The original 14.5 EH/s Bitmain commitment (14392) remains unverifiable well past its 2023 deadline. Horizon 1's Q4 2025 delivery also carries inconsistent statuses across two near-identical claims (28861, 28349), which deserves reconciliation rather than assumption. The weak point is independent corroboration. Every capacity figure here is company-sourced. The record contains no ERCOT interconnection queue entries, Nolan or Childress County permits, or on-the-ground reporting. Energization at Sweetwater 1 also does not establish commissioned, revenue-generating load, and the project is still listed as under construction. Kiowa, the Oklahoma campus and Bundey carry no megawatt figures or dates at all. Over the next 90 days, watch the fiscal first-quarter update for actual load ramp and any tenant or GPU deployment at Sweetwater 1. Watch whether Horizon phases beyond the first are delivered or re-dated. ERCOT filings should show Sweetwater 2 tracking toward late-2027 energization (28332). Finally, watch whether the three announced-only sites gain capacity and schedule disclosures, or stay undefined.

IREN's record traces a company that has repeatedly enlarged its ambitions while largely keeping its dated commitments. Hashrate targets climbed from 9.1 EH/s (12448) to 9.4 (12138), then 10 (6522, 6341), 20 (6342), 31 (5691) and 50 EH/s (28860). The power ambition grew faster. Childress went from 600 MW potential (12176) to a full 750 MW once a second transformer arrived (28866), with 650 MW described as operating (28348). Sweetwater added 1.4 GW (6523, 28863) plus a 600 MW second phase (28864), bringing total grid-connected power to 2,910 MW (28868, 28345). The GPU line follows the same pattern, from 568 H100s (6261) to 1,896 units (5712), roughly 4,300 (28343), and stated room for more than 20,000 Blackwells at Prince George (28344). Nothing here has shrunk. The story has shifted from megawatts as hashrate to megawatts as AI capacity.

On timing, every claim that carries a promised date and has been scored is marked verified. That includes the 80 MW Childress expansion by mid-2024 (6495, 6410), 50 MW in 2024 (6153), Childress Phase 5 (28862), 31 EH/s in Q4 2024 (5691), 50 EH/s by June 2025 (28860) and Sweetwater 1 energization by April 2026 (28863). The undated 2023 hashrate targets are harder to judge. The 9.1 and 9.4 EH/s figures were replaced by a 10 EH/s goal that was reported reached only in May 2024 (6150), with no explicit acknowledgement of drift. Two inconsistencies stand out. The 15.2 EH/s target for September 2023 (14393) is marked verified even though stated operating capacity was 5.5 EH/s in April 2023 (12689). Horizon 1 also appears as both not yet verifiable (28861) and verified (28349). Both labels deserve a second look.

Independent corroboration is thin. The record holds no permit filings, ERCOT interconnection-queue entries or site reporting, so the verification labels are the only outside check. Sweetwater 2's late-2027 date (28864, 28332) rests entirely on company statements. Kiowa, the Oklahoma campus and Bundey are listed as announced with no capacity, cost or date at all.

Over the next 90 days, watch how much of Sweetwater 1's energized capacity is actually carrying load, since the site is still listed as under construction. Also watch whether liquid-cooled capacity at Childress grows beyond Horizon 1 (28867), and how many Blackwells Prince George has deployed against its stated ceiling. Any queue or permit filings for Sweetwater 2 would help, as would first sizing for the three unquantified sites. Note that the claim log reviewed here stops in early July 2025, so later statements could not be tested.

IREN's trajectory is one of consistent upward revision. Early targets were in bitcoin hashrate: 15.2 EH/s in late 2021 (claim 14393), then 9.1, 9.4 and 10 EH/s through 2023 (12465, 12138, 6522), then 20 EH/s in early 2024 (6342), 31 EH/s by year-end 2024 (5691), and 50 EH/s by mid-2025 (28860, with 28347 reporting it reached). Over the same period the company changed its unit of measure. Childress moved from an 80MW build (12461) to 600MW of potential (12186) and then 750MW of transformer-backed capacity (28866). The West Texas announcement added 1,400MW (6523). By mid-2025 the company claimed 2,910MW of grid-connected power (28868, 28345) and 810MW operating (28327). Capacity has not shrunk anywhere in the record. The scale-up is steep, and the narrative has shifted from mining to AI compute, from the WEKA and Poolside tie-ups (12005, 6285) to the Horizon 1 liquid-cooled hall (28861) and the Prince George Blackwell plan (28344, 28342).

On dates, the dated milestones mostly carry verified status. These include the Childress 80MW expansion by mid-2024 (6495, 6410), the GPU commissioning (6262), the 31 EH/s year-end target (5691) and Phase 5 energization (28862, with 28326 reporting completion). The original 2021 Bitmain contract claim (14392) remains not_yet_verifiable. Horizon 1 has inconsistent statuses for the same Q4 2025 deadline: 28349 is marked verified, while 28861 is still not_yet_verifiable. Sweetwater 1 is also unsettled. Its April 2026 energization (28863) is tagged verified, but the project record still lists it as under construction, so it is unclear whether energization actually occurred or the facility only received grid power without a load ramp. The record supplied here stops in July 2025, so it cannot show whether later statements revised any of these dates without saying so.

Independent corroboration is thin. The dataset contains no permit filings, ERCOT or AESO interconnection queue entries, or on-the-ground reporting. The 2,910MW grid-connected figure in particular rests only on company statements. Until that figure is matched to queue positions and signed interconnection agreements, the gap between operating and secured capacity should be treated as a company representation.

Over the next 90 days, through mid-January 2027, watch for evidence that Sweetwater 1 is carrying meaningful load rather than only being energized. Also watch whether Horizon 1's tenant-billed capacity matches the 50MW promised, and whether Horizon 2 (28330) gains a dated commitment. Finally, check whether Sweetwater 2's late-2027 target (28864) and the announced-only Oklahoma, Kiowa and Bundey sites gain permits or queue entries, or quietly slip.

IREN's ambitions have grown sharply, and the unit of measurement changed as the story moved from bitcoin mining to AI hosting. Hashrate claims run from 2.3 EH/s (13824) to 5.5 EH/s through spring 2023 (12784, 12712), then to targets of 9.1 and 10 EH/s (12465, 6522), 31 EH/s for Q4 2024 (5691) and 50 EH/s by mid-2025 (28860, 28320). The power figures grew even faster. The 80MW Childress build (12461) became a 600MW site potential (12186), then a 1.4GW West Texas development (6523), and by June 2025 a 2,910MW grid-connected portfolio with 660MW operating (28868). Since mid-2025 the company has framed growth in megawatts and GPU counts (28343, 28344) rather than exahash.

Most dated promises have held. The Childress 80MW expansion (6495, 6410), the 50MW 2024 build (6153), the Phase 5 150MW energization (28862) and the 50 EH/s target (28860) are all recorded as met. The Sweetwater 1 April 2026 energization (28863) is also marked verified, which matters most because it is the first gigawatt-scale delivery. The weak point is the earliest guidance. The 15.2 EH/s target for September 2023 (14393) is tagged verified, yet IREN's own 2023 disclosures show about 5.5 EH/s in April (12689) and a 9.4 EH/s target in October (12138). That looks like a slipped target that was quietly reset, and the verification label should be re-audited. The Bitmain contract claim (14392) also remains unresolved three years past its date. The Horizon 1 Q4 2025 delivery is tagged not_yet_verifiable in one entry (28861) and verified in two others (28349, 28329), so the status needs reconciling. A smaller arithmetic question also stands: 660MW operating fleet-wide (28868) sits uneasily next to 650MW operating at Childress alone weeks later (28348, 28326), given that the British Columbia sites are also running.

The record contains no permits, ERCOT interconnection-queue entries or site reporting. The pace is therefore supported by company statements and outcome tagging, not by third-party evidence. The claim list also stops at July 2025, so the last fifteen months of guidance are missing from this read.

Over the next 90 days, watch for three things. First, whether Sweetwater 1's energized capacity is matched by signed AI tenancy rather than just available power. Second, whether Horizon capacity beyond the first 50MW (28867) gets dated commitments. Third, whether Sweetwater 2's late-2027 energization (28864) shows up in ERCOT queue milestones. Progress on the Prince George Blackwell deployment (28342) will also show whether GPU installation is keeping up with power.

IREN's ambitions have grown steadily, and the unit it uses to describe them has changed along the way. Through 2023 the story was hashrate. Disclosures stepped from 2.3 EH/s (13824) to 3.7 (13604), then sat at 5.5 EH/s across a long run of near-identical restatements (12784, 12712, 12689). Targets then crept from 9.1 (12465) to 9.4 (12138) to 10 EH/s (6522), with no dates attached. That makes the drift hard to call a slip, but it was never framed as a revision either. Set against the 2021 anticipation of 15.2 EH/s by September 2023 (14393), those contemporaneous figures sit awkwardly beside that claim's verified tag and deserve a second look. The 14.5 EH/s Bitmain contract (14392) has simply gone quiet. From 2024 the hashrate line accelerated sharply, with 20 EH/s installed (5689), 31 promised for Q4 2024 (5691) and 50 reached by mid-2025 (28347, 28320). The headline unit then moved to megawatts and GPUs. Childress grew from 600 MW potential (12186) to 650 MW operating and a 750 MW build-out (28326, 28866). The West Texas 1,400 MW concept (6523) became Sweetwater 1, Sweetwater 2 at 600 MW was added (28864), and the portfolio reached 2,910 MW (28868). GPUs scaled from 568 H100s (6261) to roughly 4.3k (28343), with Prince George framed as able to host more than 20k Blackwells (28344). Dated promises have mostly held on the record supplied. The Childress 80 MW by mid-2024 (6410), the 2024 50 MW build (6153), Phase 5 (28862) and the 50 EH/s target (28860) all carry verified status. The weak point is evidentiary. The record contains no interconnection queue entries, no permits and no third-party site reporting, so verification appears to rest largely on IREN's own follow-on disclosures. The most consequential date, Sweetwater 1 energization by April 2026 (28863), is now about six months past, and nothing in the supplied claims confirms it. The record also ends abruptly at claim 28331. Over the next 90 days, watch for documented Sweetwater 1 energization with ERCOT corroboration rather than a company statement alone. Watch also for evidence that Horizon 1 (28329, 28349) delivered on its Q4 2025 date and has a named tenant, for the actual installed Blackwell count at Prince George against the 2,400-unit tranche (28342), and for any quiet resetting of the late-2027 Sweetwater 2 date (28864).

IREN's record describes a company whose headline unit of capacity has changed. Through 2023 the yardstick was exahash. A 15.2 EH/s target set in November 2021 (14393) sat alongside operating figures of 2.3 to 5.5 EH/s through April 2023 (13824, 13604, 12689). The target was then reset in steps: 9.1 EH/s (12465), 9.4 EH/s (12138), 10 EH/s (6341, 6150), 20 EH/s installed (5689), 31 EH/s for Q4 2024 (5691) and 50 EH/s reached by mid-2025 (28860, 28347). The 2021 target was met roughly two years late. The record marks 14393 verified without noting that it slipped, which readers should treat with caution. From 2024 the yardstick moves to megawatts and GPUs. Childress grew from 600 MW potential (12186) to a 750 MW full build (28866) with 650 MW operating (28348, 28326). Group grid-connected power reached 2,910 MW (28868, 28345), and the GPU fleet grew from 568 H100s (6261) to about 4,300 units (28343).

Near-dated promises at Childress have mostly held. The 80 MW expansion due by mid-2024 (6495, 6410), the 50 MW 2024 build (6153) and the 150 MW Phase 5 due June 2025 (28862, confirmed in 28326) all track. Labelling is inconsistent for Horizon 1: the same Q4 2025 commitment is marked not_yet_verifiable in 28861 but verified in 28349 and 28329. That should be reconciled before it is cited. The larger risk is Sweetwater. The 1.4 GW announced in November 2023 (6523) was later given an April 2026 energization date (28863), which has now passed. Sweetwater 1 is still listed as under construction, and nothing in the claims supplied here confirms energization or acknowledges a delay. The claims list supplied ends at July 2025, so later statements, including whatever IREN said about the April 2026 milestone, could not be assessed here.

The record contains no independent corroboration. There are no ERCOT large-load or interconnection filings, no Nolan County permits and no site reporting. The Sweetwater and Sweetwater 2 (28864) timelines therefore rest entirely on company statements. Kiowa, Oklahoma and Bundey carry no capacity or dates at all.

In the next 90 days, watch for three things. First, explicit confirmation of Sweetwater 1 energization, or a dated revision. Second, ERCOT filings that match the 1.4 GW and 600 MW blocks. Third, whether Horizon's liquid-cooled expansion beyond 50 MW (28867) gains a megawatt figure and a date. The 2,910 MW grid-connected figure should also be watched for any silent restatement. A further test is whether the Prince George Blackwell build-out (28342, 28344) is reported in installed units rather than capacity.

IREN's capacity narrative has grown in steps, each larger than the last, and each step has been reframed around a new unit of account. The 2021 ambition of 15.2 EH/s (14393) stalled around 5.5 EH/s through the first half of 2023 (12784, 12712, 12689). Expansion then resumed toward 9.1–10 EH/s (12465, 6341), reached 20 EH/s installed by September 2024 (5689), and hit 50 EH/s by mid-2025 (28860, 28347). Over the same period the story moved from hashrate to megawatts. Childress went from an 80MW build (12461) to 600MW potential (12186), then 650MW operating (28326), with a transformer sized for 750MW (28866). The portfolio headline became 2,910MW of grid-connected power (28868, 28345), anchored by Sweetwater 1's 1.4GW (6523). The GPU line has been rebased even more often, from 568 H100s (6261) to roughly 1,900 (5712), then about 4,300 (28343), and now a claimed footprint of more than 20,000 Blackwell GPUs at Prince George (28344). The direction has been consistently upward, but the metrics are not continuous, so direct comparison is hard.

On timing, IREN's record on dated promises is unusually clean. The Childress 80MW build by mid-2024 (6495, 6410), the 31 EH/s Q4 2024 target (5691), Phase 5 energization by June 2025 (28862), and Sweetwater 1 energization by April 2026 (28863) are all tagged verified. Smaller revisions went unremarked. Late-2023 targets moved among 7.0, 8.3 and 9.4 EH/s within weeks (12137, 6493, 12138). The Horizon 1 Q4 2025 delivery carries inconsistent verification status across restatements (28861 versus 28349 and 28329). The 2021 Bitmain 14.5 EH/s contract (14392) has never been closed out.

The independent evidence is thinner than the verification tags imply. The record provided contains no permit filings, ERCOT queue entries or site reporting, and it ends mid-2025. The verified flags therefore appear to rest mainly on IREN's own later disclosures. Sweetwater 1 is still listed as under construction even though its energization is marked verified, which suggests that energized grid capacity and usable IT load are not the same thing. Kiowa, the Oklahoma campus and Bundey have no capacity or dates attached.

Over the next 90 days, the first thing to watch is whether IREN reports contracted, revenue-generating AI load at Sweetwater 1 or only energized megawatts. The second is whether Sweetwater 2's late-2027 date (28864) gets interconnection documentation behind it. The third is whether self-mining hashrate is quietly reduced as Childress capacity moves to liquid-cooled AI halls (28867).

IREN's ambitions have grown sharply, and so has the unit of measure. In late 2021 the company was talking about 15.2 EH/s by September 2023 (14393). In 2023 and 2024 the language shifted to megawatts: Childress grew from a 600 MW potential site (12176, 12186) to a full 750 MW once a second transformer was on site (28866). By mid-2025 the company claimed 2,910 MW of grid-connected power across the portfolio (28868, 28345, 28328). The GPU story escalated in the same way, from 568 H100s (6261) to roughly 1,900 units (5712), then about 4,300 (28343), then a stated ability to host more than 20,000 Blackwell GPUs at Prince George (28344). Capacity has not shrunk anywhere in the record, but the headline metric keeps moving to whatever flatters the next step.

The early hashrate dates did not hold on the company's own numbers. Operating capacity sat at 5.5 EH/s through spring 2023 (12784, 12689), reached 8.3 EH/s by November (6493) and 10 EH/s only in May 2024 (6150). That is well short of the 15.2 EH/s promised for September 2023, so the verified tag on 14393 deserves a second look. The record since 2024 is better. The Childress 80 MW expansion (6495, 6410), Phase 5 at 150 MW (28862) and the 50 EH/s mid-2025 target (28860, 28347) are all marked delivered on time. Sweetwater 1's April 2026 energization (28863) is also marked verified. Horizon 1 is less clear: one Q4 2025 delivery claim is verified (28349) while a parallel one remains not yet verifiable (28861), and the project is still listed as only partially operational.

Independent corroboration is thin. The record has no permit filings, ERCOT interconnection-queue entries or site reporting, so the verification flags are the only outside check. Sweetwater 1 also carries a specific street address and an energized date but is still classed as under construction. That points to a gap between having power at the site and having revenue-generating compute running there. Note too that this claim log stops in July 2025, leaving more than a year unaccounted for.

Over the next 90 days, watch whether IREN puts a utilized-megawatt or contracted-tenant figure on Sweetwater 1, not just energized capacity. Watch whether Horizon moves to fully operational, and whether Sweetwater 2's late-2027 target (28864) gains visible interconnection milestones. Kiowa, the Oklahoma campus and Bundey are announced with no capacity or cost attached; any numbers there would be the next round of scope expansion.

IREN's trajectory is one of sustained upward revision rather than retreat, though the unit of ambition has changed. The earliest headline, a 15.2 EH/s fleet target by September 2023 (14393), was tagged verified, yet operating capacity sat at 2.3 to 3.7 EH/s through late 2022 (13824, 13604) and plateaued at 5.5 EH/s across a long run of near-identical spring 2023 statements (12784 through 12689). From there the hashrate story compounded: 9.1 EH/s in mid-2023 (12448), 10 EH/s by May 2024 (6150), 20 EH/s installed with 31 EH/s promised for Q4 2024 (5689, 5691), and 50 EH/s claimed by mid-2025 (28347). Power ambitions grew faster still. Childress went from 600 MW potential (12176) to 650 MW operating (28326) with another 450 MW flagged (20302), and the portfolio total reached 2,910 MW grid-connected (28345). Sweetwater 1's 1,400 MW figure has held unchanged since November 2023 (6523, 28331).

Dates have mostly held where they were specific. The Childress 80 MW expansion (6410), the 568-GPU commissioning (6262), the H200 fleet build (5712) and Horizon 1's Q4 2025 delivery (28329) all carry verified tags. The Sweetwater 1 energization target of April 2026 (28331) is also marked verified, but energization is not IT load, and the record does not show how much of that 1,400 MW is actually serving customers six months on. The 2023 Bitmain 14.5 EH/s contract (14392) and the Dell supply agreement due by March 2026 (4161) remain unverifiable, the latter now overdue for confirmation.

The weakest part of the record is independent corroboration. Verification tags aside, nothing here cites ERCOT large-load interconnection filings, Nolan County permits, or site reporting for Sweetwater, nor BC Hydro service confirmation beyond IREN's own description (12198). The pivot from mining to AI cloud, anchored by the $9.7 billion Microsoft contract (8112), raises the stakes on that gap, since GPU-ready capacity is far harder to stand up than miner racks. The supplied record also ends mid-entry at 20309, so late-2025 and 2026 claims may be missing.

Over the next 90 days, watch for disclosed IT load energized at Sweetwater 1 versus grid capacity, any confirmation that Dell hardware landed against the March 2026 date, Horizon build-out pacing tied to Microsoft delivery milestones, ERCOT queue or permit activity for Sweetwater 2's late-2027 target (28332), and whether the 50 EH/s mining base is held or quietly reallocated to AI load.

IREN's facility record splits into two eras, and they need different readings. The bitcoin-mining era opens with an anticipated 15.2 EH/s by September 2023 (claim 14393). The company's own later figures don't support that date. Operating capacity sat at 5.5 EH/s through April 2023 (12689, 12712), and the June 2023 target was only 9.1 EH/s (12443). Claim 14393 carries a verified tag, but on this record the date was effectively reset, never acknowledged as a miss. After that reset, the mining ramp became credible: 10 EH/s by May 2024 (6150), 20 EH/s installed by September 2024 (5689), and 50 EH/s reported in mid-2025 (28347, 28320). The 2024 Childress milestones (6410, 6153) and the Q4 2024 31 EH/s target (5691) are tagged verified, so the company largely delivered once targets were reset to nearer horizons.

The AI era scales the numbers by an order of magnitude. The GPU fleet goes from 568 H100s (6261) to roughly 1,900 (5712) to about 4,300 (28343), with Prince George framed for more than 20,000 Blackwells (28344). Grid-connected power is put at 2,910 MW (28345, 28328). Commercial backing now rests on a $9.7 billion Microsoft contract (8112) and a $5.8 billion Dell supply agreement due by March 2026 (4161), which is still not verifiable. The Childress numbers also need care. A 650 MW operating figure (28326) sits beside a 200 MW IT-load figure (20310). That is gross power versus deliverable IT capacity, not a contradiction, but headline megawatts should not be read as AI-ready capacity.

The central timing question is Sweetwater 1. Claim 28331 promised energization of 1,400 MW by April 2026. Six months past that date, the project is still listed as under construction, and no claim in this record confirms energization or acknowledges a slip. The record contains no permits, ERCOT interconnection-queue filings or on-the-ground reporting for Sweetwater, Kiowa, Oklahoma or Bundey. Independent support for the 2.9 GW narrative therefore rests on the company's own statements and the verified tags on earlier, smaller milestones. The supplied record also ends mid-entry at claim 20309, so later disclosures may be missing.

Over the next 90 days, watch for four things: an explicit statement on whether Sweetwater 1 has energized, and at what megawatt level; evidence that Dell hardware has been delivered and is producing revenue under the Microsoft contract; a dated plan for Horizon 2 (28330); and any interconnection or permitting filings that would move Sweetwater 2's late-2027 date (28332) and the earlier-stage sites beyond press-release status.

IREN's record describes a company that has kept expanding its stated ambitions, and whose dated promises mostly check out. The exception is the original 2021 hardware plan, which was never cleanly closed out. On capacity, the arc is steep but uneven. The November 2021 target of 15.2 EH/s by September 2023 (14393) and the matching 14.5 EH/s Bitmain purchase contract (14392) gave way to operating figures of 2.3 to 3.7 EH/s in 2022 (13824, 13604). Capacity then sat at 5.5 EH/s through at least five restatements in spring 2023 (12784, 12712, 12689). Only after that did it climb through 9.1 to 10 EH/s (12465, 6150), 31 EH/s by Q4 2024 (5691) and 50 EH/s by mid-2025 (28347). The power story compounded faster. Childress went from 600 MW of potential (12176) to 650 MW operating (28326), and the portfolio grew to 2,910 MW grid-connected (28345) and then 3 GW secured (20311). GPU counts rose from 568 units (6261) to a stated 100,000-unit deployment scale (20296), anchored by the $9.7 billion Microsoft contract (8112). That marks a strategic pivot from self-mining to AI infrastructure, not just more of the same. On dates, the verified record is strong. The 80 MW Childress expansion by mid-2024 (6495, 6410), the 2024 50 MW build (6153), the 31 EH/s Q4 2024 target (5691) and Horizon 1 by Q4 2025 (28329) all held. Sweetwater 1 energization by April 2026 (28331) is also marked verified, although the project itself still shows as under construction. The weak point is 2021. Claim 14392 remains unverifiable years past its deadline, and the 2023 plateau at 5.5 EH/s makes the verified label on 14393 worth re-examining. The 15.2 EH/s target appears to have been re-baselined rather than met on schedule, and the record holds no explicit acknowledgement of that. Independent corroboration is thin in this dataset. The BC Hydro connection (12198) and the Sweetwater street address are the only external anchors. No ERCOT queue entries, permits or site reporting are on record for Sweetwater 2, Kiowa, the Oklahoma campus or Bundey, and none of those projects carries a stated capacity. The record also cuts off mid-claim at 20304, so the latest British Columbia and Childress figures may be incomplete. Over the next 90 days, the key test is whether the 200 MW of Childress construction phases due by end-2026 (20295) is delivered and reported as energized, IT-loaded capacity rather than shell. The other markers are Microsoft-linked GPU commissioning milestones, any tenant or utilization disclosure for the energized Sweetwater 1, and whether Sweetwater 2's late-2027 date (28332) gains interconnection or permitting evidence.

IREN's capacity narrative has scaled up steadily and changed shape along the way. The company has moved from a hashrate story to a megawatt and GPU story. Bitcoin targets rose in stages, from 5.5 EH/s operating (12827) to 9.1 and 9.4 EH/s targets (12465, 12138), 10 EH/s (6150), 31 EH/s by Q4 2024 (5691), and 50 EH/s (28347). The power ambitions grew faster. Childress went from 600 MW potential (12186) to 650 MW operating (28326). The West Texas figure of 1,400 MW (6523) became a 2,000 MW Sweetwater Hub (20304), which looks like Sweetwater 1 plus the 600 MW Sweetwater 2 (28332) rather than new capacity. The GPU fleet went from 568 units (6261) to 1,896 (5712), roughly 4,300 (28343), and a stated 100,000-unit scale (20296) tied to the $9.7 billion Microsoft contract (8112). That last step is more than a twentyfold jump with no intermediate disclosures on the record.

Dated promises have largely held where they can be checked. The Childress 80 MW expansion (6495, 6410), the H100 commissioning (6262), the 31 EH/s target (5691), Horizon 1 (28329), and Sweetwater 1 energization in April 2026 (28331) are all marked verified. One tension stands out. The 2021 promise of 15.2 EH/s by September 2023 (14393) is tagged verified, yet the company's own 2023 disclosures put operating capacity at 5.5 EH/s and targets below 10 EH/s. That tag deserves re-audit. The related Bitmain 14.5 EH/s contract (14392) remains unverifiable, and the company quietly re-based its targets rather than addressing the shortfall.

The record contains no independent corroboration of the announced pace: no ERCOT interconnection queue entries, Nolan County or Childress County permits, or BC Hydro filings. The verification tags appear to rest on company follow-up statements. The grid-connected 2,910 MW figure (28345, 28328) in particular needs an outside check against ERCOT and BC Hydro records. The record also stops in November 2025, leaving an eleven-month gap through the Sweetwater 1 energization window.

Over the next 90 days, the key date is the Childress GPU deployment deadline of December 31, 2026 (20309). It is the first hard test of whether the Microsoft-linked GPU build matches the contract's revenue ramp. Watch for confirmation that Sweetwater 1 was actually energized at the promised scale rather than in a partial first tranche. Also watch for interconnection or construction evidence that Sweetwater 2 is on track for late 2027, and for any site assignment of the 110 MW British Columbia expansion (20303).

IREN's record follows two arcs, and they should be read separately. The first is the bitcoin buildout. A 15.2 EH/s target set in late 2021 (14393) is marked verified. Its companion hardware commitment (14392) is still not verifiable. Between them, the record shows a long plateau: from February through April 2023, at least six separate statements (13337, 12784, 12827, 12712, 12699, 12689) repeat the same 5.5 EH/s figure. Repeated restatements of one installed number like this usually mean deployment stalled and messaging filled the gap. Growth then resumed quickly: 9.1 to 9.4 EH/s (12443, 12138), 10 EH/s (6150), 31 EH/s by Q4 2024 (5691, verified), and 50 EH/s by mid-2025 (28347).

The second arc is the pivot to AI. Here the units change, and with them how comparable the claims are. GPU counts move from 568 (6261) to 1,896 (5712) to roughly 4,300 (28343), then to a stated deployment scale of 100,000 (20296). That last figure is roughly twenty times the installed base and has no date attached. Power claims grow in the same way. Childress goes from 600 MW of potential (12176) to 650 MW operating (28326). Sweetwater's 1,400 MW (6523) feeds into 2,910 MW grid-connected (28345) and then about 3 GW secured (20311). The jump from "potential" to "operating" to "secured" is never reconciled in the record. That is a framing revision rather than a disclosed one.

On timing, dated commitments have mostly held. The 80 MW Childress expansion (6495, 6410), the H100 commissioning (6262), 50 MW at Childress in 2024 (6153), Horizon 1 in Q4 2025 (28329) and Sweetwater 1 energization by April 2026 (28331) are all marked verified. The weaker spots are the undated ones. The 100 MW early works (6496, 6413) and Horizon 2 planning (28330) were never given dates, so they can neither slip nor be confirmed.

Independent corroboration is thin. The record contains no ERCOT queue entries, permits or site reporting for Sweetwater, Kiowa, Oklahoma or Bundey. The four sites other than Sweetwater remain announced with no capacity attached. The verified flags therefore rest largely on IREN's own later disclosures. The supplied record also ends mid-entry at 20304.

Over the next 90 days, the test is whether the 200 MW of Childress phases promised by year-end 2026 (20295) is energized and leased. Delivery against the Microsoft contract (8112) should become visible in revenue. The 110 MW British Columbia expansion (20303) needs dates. Sweetwater 2's late-2027 energization (28332) should show up in public interconnection filings.

IREN's capacity story has moved in two phases: a bitcoin-mining ramp measured in exahash, then an AI-infrastructure pivot measured in megawatts and GPUs. The first phase began with a 15.2 EH/s target for September 2023 (14393). Operating capacity then stalled near 5.5 EH/s across at least six restatements through April 2023 (12827, 12712, 12689), reached 8.3 EH/s by late 2023 (6493), and hit 10 EH/s only in May 2024 (6150). Along the way the goalposts were quietly reset to 9.1 (12465), 9.4 (12138) and 10 EH/s (6522), and the original target was never formally retired. The verified tag on 14393 deserves scrutiny, since the record implies a slip of roughly a year. After that, execution accelerated. The company reported 20 EH/s installed (5689), a 31 EH/s Q4 2024 target that held (5691), and 50 EH/s by mid-2025 (28347).

Power claims have scaled more consistently. Childress went from 600 MW potential (12176) to 650 MW operating (28326), within 810 MW total operating (28327) and roughly 2.9 GW grid-connected (28345). The Sweetwater figures reconcile cleanly: 1,400 MW for Sweetwater 1 (6523) plus 600 MW for Sweetwater 2 (28332) matches the 2,000 MW hub (20304). Dated infrastructure commitments since late 2023 have largely held. These include the Childress 80 MW expansion (6410), the 50 MW 2024 build (6153), Horizon 1 in Q4 2025 (28329) and Sweetwater 1 energization by April 2026 (28331). The GPU trajectory is steeper and less tested. The fleet grew from 568 H100s (6261) to about 4,300 units (28343), and the company now claims a 100,000-unit deployment ambition (20296) anchored by the $9.7 billion Microsoft contract (8112).

Independent corroboration is thin. Every verification in this record appears to rest on company disclosures. No ERCOT queue position, building permit or site reporting is attached for Sweetwater (883 FM 611) or Horizon, and Kiowa, Oklahoma and Bundey have no capacity claims at all. The Dell partnership entry (20308) is also incomplete in the record.

The key date in the next 90 days is the December 2026 Childress GPU deployment deadline (20309), which is the first hard test of the Microsoft-linked build. Watch whether Sweetwater 1 progresses from energization to live IT load, whether Horizon 2 (28330) acquires a date, and whether reported hashrate starts falling as power is reallocated to AI. Any silent change to the late-2027 Sweetwater 2 target (28332) would echo the 2023 pattern.

IREN's record traces a company that grew its headline capacity aggressively, then changed what it was measuring. The original 2021 target of 15.2 EH/s by September 2023 (14393) did not survive contact with the build. Operating hashrate sat at 5.5 EH/s through spring 2023 (12827, 12689), and the target was then restated as a ladder of smaller near-term goals: 9.1 EH/s (12448), 9.4 EH/s (12138), 10 EH/s (6522, 6341). Fifteen EH/s was effectively reached only by the 20 EH/s disclosure of September 2024 (5689), roughly a year late, and the record shows no acknowledgement of that slip. After the reset, the mining build accelerated sharply: 31 EH/s was promised for Q4 2024 (5691) and 50 EH/s was reported by mid-2025 (28347, 28320). Power figures have mostly grown rather than been revised down. Childress went from a 600 MW potential (12176) to 650 MW operating (28326), with another 450 MW now flagged (20302). Grid-connected power rose from 2,910 MW (28345) to a round 3 GW (20311). The bigger story is the shift from hashrate to GPUs: 568 H100s (6261) became 1,896 units (5712), then about 4.3k (28343), and finally the $9.7 billion Microsoft contract (8112), backed by a parent guarantee on Dell purchase obligations (4164).

The company's near-term dated promises have generally held. The 80 MW Childress expansion (6495, 6410), the 2024 GPU commissioning (6262) and the Q4 2024 hashrate target (5691) are all tagged verified. The larger, longer-dated items are a different matter. Horizon 1's Q4 2025 delivery (28349, 28329) is still unverifiable, although the project is listed as partially operational. Sweetwater 1's April 2026 energization (28331) is now more than five months overdue. The project is still marked under construction, and the November 2025 disclosures neither restate nor revise that date.

Nothing in this record independently confirms the AI-era pace. There are no ERCOT interconnection or large-load entries, no permits, and no on-the-ground reporting for Sweetwater, Horizon, or the announced Kiowa, Oklahoma and Bundey sites, none of which carries a capacity figure.

Over the next 90 days, the most important signal is whether IREN confirms Sweetwater 1 energization, or restates its date, in its next quarterly update. ERCOT queue and energization records should be checked against claim 6523's 1,400 MW. Watch also for evidence that Horizon capacity is actually delivering to Microsoft under 8112, for any scheduling detail on the additional 450 MW at Childress, and for whether Sweetwater 2's late-2027 date (28332) moves alongside its sister site.

IREN's record shows two separate expansion stories. The first is a bitcoin-mining hashrate story that consistently ran behind its own promises. The second is an AI-compute story that has grown much faster than any delivery evidence in the file. The November 2021 target of 15.2 EH/s by September 2023 (claim 14393) was not met. Operating capacity was still being reported at 5.5 EH/s through April 2023 (12784, 12689) and 9.1 EH/s in June 2023 (12465), and the 10 EH/s level was only described as achieved in May 2024 (6150). The company never reconciled the original target in later disclosures. It simply reset the goal to 10, then 20, then 31 EH/s (6341, 6342, 5691). The 20 EH/s installed figure in September 2024 (5689) suggests the pace did pick up once Childress came online. However, the 31 EH/s Q4 2024 target, the 80MW Childress build due mid-2024 (6495, 6410), and the 50MW 2024 build (6153) all remain unresolved in this record even though their dates have passed.

The power figures have grown steadily and without stated reasons. Childress went from 600MW (12176) to 750MW (20301). The 1,400MW West Texas figure (6523) became 2,000MW at Sweetwater (20304), within a 3,000MW secured portfolio (20311). GPU counts went from 568 (6261) to 1,896 (5712) to a stated 100,000-unit scale (20296), anchored by the $9.7 billion Microsoft contract (8112) and the $5.8 billion Dell purchase (20308, 4161). The Dell agreement carried a March 2026 milestone that has passed with no recorded outcome.

The record contains no independent corroboration of any kind: no permits, no ERCOT large-load queue entries, no BC Hydro filings, and no site reporting. The Sweetwater 1 entry has a street address and an under-construction status, but nothing third-party backs up the energization timing. Every capacity figure here is self-reported. The claim list is also cut off after 20251, so later disclosures may change this picture.

The next 90 days are decisive because three commitments come due on 31 December 2026. These are the 200MW Horizon/Childress build-out for Microsoft (20294, 20295, 4158), the GPU deployment schedule at Childress (20309), and the British Columbia GPU transition (20249). Watch whether the next quarterly update reports energized megawatts and revenue-generating GPU clusters, rather than secured power. Also watch for whether Microsoft acceptance milestones are disclosed, whether Dell shipment volumes are confirmed, and whether ERCOT or county records show Sweetwater energization. Given the history, an unacknowledged slip into 2027 would fit the pattern; an on-time, verifiable handover would break it.

IREN's trajectory splits into two eras, and the headline numbers have consistently ratcheted upward rather than shrunk. In the mining era, the November 2021 target of 15.2 EH/s by September 2023 (14393, backed by the 14.5 EH/s Bitmain contracts in 14392) was not met. Stated capacity sat at 3.7 EH/s in September 2022 (13604) and 5.5 EH/s through spring 2023 (12784, 12689). The goalpost was then reset to 9.1 EH/s (12465), 9.4 EH/s (12138) and 10 EH/s (6522), which was reported as operating in May 2024 (6150). The record never reconciles that against the original deadline, so the slip of roughly a year looks unacknowledged rather than revised. Capacity then accelerated to 20 EH/s by September 2024 (5689), with 31 EH/s promised for Q4 2024 (5691). Site power followed the same inflationary arc: Childress went from 600 MW (12176) to 750 MW (20301), and West Texas went from 1,400 MW (6523) to a 2,000 MW Sweetwater figure (20304) inside a 3,000 MW secured portfolio (20305). Intermediate Childress build promises were also redefined. The 80 MW expansion was due by mid-2024 (6495, 6410), and a separate 50 MW tranche was dated to end-2024 (6153). No delivery confirmation for either appears in the record, and several past-due claims remain tagged not_yet_verifiable, which is itself a gap. The AI pivot replaces hashrate with contracted load: a $9.7 billion Microsoft agreement (8112) tied to 200 MW at Childress/Horizon by end-2026 (20294, 20295, 4158), funded by a $5.8 billion Dell purchase (20308, 4161) and framed against a 100,000-GPU ambition (20296). The 200 MW figure appears under both Childress (20310) and Horizon (4159), so readers should treat these as one commitment, not two. No independent corroboration was supplied: no ERCOT or BC Hydro queue entries, county permits, or site reporting. The announced pace therefore rests on company statements alone. The supplied record also truncates mid-entry at claim 20251. Over the next 90 days, the end-2026 Microsoft delivery date (20294) and the British Columbia GPU transition (20249) both come due. Watch three things. First, whether quarterly disclosures show energized megawatts at Horizon rather than construction phases. Second, whether the Dell deliveries promised by March 2026 (4161) are confirmed as received and installed. Third, whether any Sweetwater interconnection or energization milestone surfaces in third-party filings to support the 2,000 MW claim.

IREN's capacity story has two distinct phases, and the record treats them unevenly. In mining, the 15.2 EH/s target due by September 2023 (14393) was missed badly. Operating capacity sat near 2.3 to 3.7 EH/s through late 2022 (13824, 13604) and plateaued at 5.5 EH/s through spring 2023 (12827, 12698). The goal was then reset in steps: 9.1 EH/s (12443), 9.4 EH/s (12138), 10 EH/s (6522, 6341), and a 20 EH/s pathway (6342). The resets did eventually land. 10 EH/s was reported in May 2024 (6150) and 20 EH/s installed by September 2024 (5689). The 31 EH/s Q4 2024 target (5691), however, is flagged delayed, and the record contains no later hashrate figure confirming it. The original miss was never explicitly reconciled; it was overtaken by newer, smaller targets.

Power figures have only grown. Childress went from 600 MW (12197) to an additional 500 MW (6412) and is now framed as 750 MW total with 450 MW of additional capacity (20301, 20302). The West Texas site grew from 1,400 MW (6523) to 2,000 MW at Sweetwater (20304), within a 3 GW secured portfolio (20305, 20311). Delivery has lagged these numbers. The 80 MW Childress build due mid-2024 (6495, 6410), the 50 MW 2024 build (6153), the 568-GPU commissioning (6262), and the 1,896-unit H200 fleet (5712) are all marked delayed.

The November 2025 pivot to the $9.7 billion Microsoft contract (8112) and the $5.8 billion Dell GPU order (20308) raises the stakes on Childress. That site's 200 MW IT load (20310) is promised by year-end 2026 (20309).

No independent evidence appears in this record: no ERCOT interconnection queue entries, no Nolan or Childress County permits, no BC Hydro filings, and no site reporting. Every project row still reads announced with blank capacity, even sites the company says have operated since 2019 to 2023 (12193, 12194). The supplied claim list also ends mid-entry after 20303, so later claims are not reflected here.

Over the next 90 days, the key test is whether Childress shows energized GPU halls before the December 31 deadline in 20309. Watch for Microsoft revenue recognition, ERCOT energization milestones for Sweetwater, and any quiet restatement of the 31 EH/s or 750 MW figures.

IREN's facility record has two phases, and the second has largely replaced the first. From 2021 to 2024 the headline was hashrate. A 15.2 EH/s target for September 2023 (14393), backed by Bitmain contracts (14392), was missed. Reported capacity then rose in steps: 2.3 EH/s (13824), 3.7 EH/s (13604), a long plateau at 5.5 EH/s through spring 2023 (12712, 12689), and 10 EH/s by May 2024 (6150). Targets moved along the way. A 9.1 EH/s goal in June 2023 (12448) was restated in October as 7.0 EH/s, with a path to 9.4 (12144, 12138). The 20 EH/s installed by September 2024 (5689) was real growth, but the 31 EH/s promised for Q4 2024 (5691) is tracked as delayed. From late 2025 the story is AI compute: a $9.7 billion Microsoft contract (8112), a 100,000-GPU scale claim (20296), a $5.8 billion Dell purchase (20308), and 200 MW of Childress IT load due by end-2026 (20294, 20310). The earlier GPU steps, 568 H100s by mid-2024 (6262) and a 1,896-unit fleet by end-2024 (5712), are both flagged delayed. That makes the jump to six-figure GPU counts a change in kind, not an extension of a proven ramp.

Childress dates have been revised more than they have held. An 80 MW expansion was due by June 2024 (6495, 6410). It sat alongside a 100 MW early-works claim (6496, 6413) and was later restated as 50 MW by end-2024 (6153). All are marked delayed, and none of the restatements explains the change from the prior figure. Site power potential has moved from 600 MW (12176) to an added 500 MW (6412) to 750 MW total (20301), again without reconciliation. The record also has no claims at all between September 2024 and November 2025, so the move from mining to AI hosting is visible only at its two ends.

The file contains no independent corroboration: no permits, no ERCOT or BC Hydro queue entries, no site reporting. None of the eleven project entries lists a capacity or cost. The 1,400 MW West Texas figure (6523) and the 110 MW British Columbia expansion (20303) are therefore unverified company statements.

Over the next 90 days, the key test is whether Horizon/Childress shows energized, commissioned capacity against the end-2026 Microsoft schedule (4158, 20309). Specific signs to look for are Dell delivery tranches, any amendment to Microsoft milestones, and the first revenue recognition from the contract. The other item is whether an ERCOT interconnection filing appears for Sweetwater. The final claim in this record (4164) is truncated and should be re-pulled before publication.

IREN's record describes two companies. The first is a bitcoin miner whose hashrate targets ran well ahead of delivery. The second is an AI infrastructure developer whose power claims have grown faster than any dated commitment attached to them. The original plan of 15.2 EH/s by September 2023 (14393, backed by Bitmain contracts in 14392) was delayed. The company was installing 5.5 EH/s in April 2023 (12784, 12712) and reported 10 EH/s only in May 2024 (6150). Hashrate did eventually compound, reaching 20 EH/s installed by September 2024 (5689). But the Q4 2024 step to 31 EH/s (5691) is again flagged delayed. The pattern is consistent: capacity arrives, roughly two to four quarters late.

On power, the scaling is upward and largely undated. Childress went from 600 MW potential (12176, 12186) to 750 MW total (20301), with another 450 MW described as additional (20302). The West Texas site grew from 1,400 MW (6523) to 2,000 MW at Sweetwater (20304), feeding a 3,000 MW secured portfolio (20305, 20311). Dated construction commitments have not held. The 80 MW Childress expansion due mid-2024 (6495, 6410) slipped, the 50 MW build due by end-2024 (6153) slipped, and both GPU tranches slipped: 568 H100s by mid-2024 (6262) and the H200 fleet expansion to 1,896 units (5711, 5712). The record holds no restated dates for these items. The 2025 disclosures replace them with larger, mostly undated figures, such as 100,000 GPUs (20296) and the 110 MW British Columbia expansion (20303). That is a revision without a reconciliation.

The record contains no independent corroboration: no ERCOT large-load interconnection filings, no Nolan or Childress County permits, no BC Hydro tariff documents, and no site reporting. The secured-power figures are therefore company-stated only. That matters most for Sweetwater, where 2,000 MW is an order of magnitude beyond anything IREN has energized.

The next 90 days are decisive because the Microsoft-linked 200 MW at Childress carries a 2026-12-31 date (20294, 20310). The record is truncated before that claim's status, so it should be confirmed before publication. It anchors the $9.7 billion contract (8112) and the $5.8 billion Dell GPU purchase (20308). Watch for Dell delivery confirmations, energization of Childress IT load, the fiscal Q1 results commentary on Microsoft acceptance milestones, and any ERCOT queue or county filings that would turn the Sweetwater figure from claim into evidence. On the record so far, a slip past year-end would fit the pattern more than break it.

IREN's capacity narrative has grown by roughly an order of magnitude, but the numbers mostly describe options rather than deliveries. In bitcoin mining, the original 15.2 EH/s target for September 2023 (claim 14393) was missed by about a year. Reported operating capacity sat at 5.5 EH/s through spring 2023 (12784, 12698), moved through a series of intermediate targets of 9.1, 9.4, and 10 EH/s (12465, 12138, 6522), and reached 20 EH/s only in September 2024 (5689). The follow-on 31 EH/s goal for Q4 2024 (5691) is itself flagged as delayed. Power-side figures have grown faster than anything visibly built. Childress went from 600 MW (12176) to 750 MW (20301), Sweetwater was introduced at 1,400 MW (6523), and the North American portfolio is now framed as 3,000 MW secured (20311).

Date discipline is weak and is rarely acknowledged. The 80 MW Childress expansion due mid-2024 (6495, 6410), the 50 MW Childress build due end-2024 (6153), the 568-GPU H100 commissioning (6262), and the H200 purchase and resulting fleet size (5711, 5712) all carry delayed status. None of them has a later claim that resets the date. Instead, the record goes silent from September 2024 to November 2025 and resumes with a new framing: the $9.7 billion Microsoft contract (8112), the $5.8 billion Dell GPU agreement (20308), and a 200 MW Horizon IT load (4159) promised by end-2026 (4158, 20309). That pattern looks more like replacing old commitments than closing them out. The record also cuts off mid-entry at 20295, so the Childress phasing detail is incomplete.

The independent evidence here is thin. The record has no permits, ERCOT interconnection queue positions, or site-level reporting for Childress, Horizon, or Sweetwater. Every project is still listed as announced, with capacity and cost blank, including Childress, which the company describes as operating since April 2023 (12196). Until queue data or construction filings are attached, the 750 MW and 3,000 MW figures should be read as company assertions.

The next 90 days run directly into the 2026-12-31 Horizon and Microsoft deadline. The key signals are whether IREN reports energization and customer acceptance of discrete Horizon phases, whether Dell shipments are tied to recognized Microsoft revenue, and whether Sweetwater gains a verifiable ERCOT interconnection milestone. Also worth watching is whether the end-2026 date is restated explicitly or quietly reworded, as earlier targets were.

IREN's record shows ambitions that kept growing while delivery kept slipping. The early bitcoin-mining targets set the pattern. A 15.2 EH/s fleet promised for September 2023 (14393), backed by 14.5 EH/s of Bitmain contracts (14392), is marked delayed. Operating capacity rose from 2.3 EH/s (13824) to 3.7 EH/s (13604), then held at 5.5 EH/s through a run of near-identical April 2023 statements (12784, 12712, 12689). After that, the next target moved instead of being met: 9.1 EH/s (12448), then 7.0 and 9.4 EH/s (12137, 12138), then 10 EH/s (6522, 6341). The company did report 20 EH/s installed by September 2024 (5689), but the 31 EH/s promised for Q4 2024 (5691) is again flagged delayed. The power figures grew the same way. Childress went from 600 MW (12176) to 750 MW (20301), and the West Texas site went from 1,400 MW (6523) to a 2,000 MW Sweetwater Hub (20304). None of these figures comes with a stated energization date.

Most of the dated promises have slipped, and the record rarely acknowledges it. The Childress 80 MW build described as underway in June 2023 (12444) was later given a June 2024 delivery date (6495, 6410) and missed it. The 50 MW Childress block for 2024 (6153), the T21 shipments (6408), the 568-GPU commissioning (6262) and the H200 fleet (5711, 5712) all carry delayed flags. We see no restated dates for any of them, so the misses appear to have been absorbed rather than explained.

The AI pivot raises the stakes considerably. The $9.7 billion Microsoft contract (8112) rests on Horizon's 200 MW (4159) being delivered by end-2026 (4158, 20309, 20295). The $5.8 billion Dell supply agreement that feeds it (4161) already shows a March 2026 milestone as delayed.

The record contains no independent corroboration. There are no ERCOT interconnection-queue entries, Texas permits, BC Hydro filings or site reporting, so every capacity figure here is self-reported. The claim list also appears truncated, which means later revisions may be missing.

Over the next 90 days, the key question is whether IREN shows evidence that Horizon is being commissioned in phases before the December 31, 2026 deadline. Beyond that, watch for any disclosure of Dell delivery volumes, any ERCOT or utility filing that supports Sweetwater's 2,000 MW, and whether Microsoft revenue recognition begins on schedule or the deployment window is quietly extended.

IREN's record traces two expansions, and the second is far larger than the first. The bitcoin-mining phase climbed in steady increments. It went from 2.3 EH/s (13824) to 3.7 EH/s (13604), then to a 5.5 EH/s figure that was restated almost weekly through spring 2023 (13348, 12726, 12689). After that came targets of 9.1 EH/s (12443), 9.4 EH/s (12138) and 10 EH/s (6522, 6341), then 20 EH/s installed (5689) and a 31 EH/s Q4 2024 goal (5691). The power claims grew faster than the hashrate. Childress went from 600 MW (12176, 12186) to 750 MW (20301). The West Texas site went from 1,400 MW (6523) to a 2,000 MW Sweetwater hub (20304), and the total grid-connected portfolio reached 3,000 MW (20305). The 2025 pivot to AI compute ties that land bank to a $9.7 billion Microsoft contract (8112) and a 200 MW Horizon deployment (4159, 20294).

The company has a consistent record of missing its dated commitments. The founding 15.2 EH/s target for September 2023 (14393) and its Bitmain supply (14392) slipped. The 80 MW Childress expansion due by mid-2024 (6495, 6410) slipped, as did the 50 MW 2024 build (6153), the first H100 commissioning (6262), the 1,896-GPU fleet (5712) and the 31 EH/s target (5691). The interim hashrate goals were not formally withdrawn. They were overtaken by newer, larger numbers, so the targets were reset without the company acknowledging the misses. The Dell supply milestone of March 2026 (4161) is already flagged as delayed, and that is the first warning on the Microsoft build path.

The file contains no independent corroboration. There are no ERCOT interconnection queue records, no Nolan or Childress County permits, no BC Hydro filings and no site reporting. The power totals and operating-since dates (12196, 12195, 12193) therefore rest only on company statements. The claims listing also ends at 20301, so any later updates are not assessed here.

The next 90 days fall exactly on the deadline window. Horizon's GPU deployment and the Childress construction phases are both promised by December 31, 2026 (4158, 20295, 20309). The key test is whether the fiscal Q1 update reports energized megawatts and accepted GPU capacity at Horizon, or just shipments. Other things to watch are whether the delayed Dell deliveries close, whether Microsoft revenue recognition begins, and whether ERCOT queue status and county filings for Sweetwater match the 2,000 MW claim. A quiet shift of the 200 MW date into 2027 would fit the company's past pattern of resetting targets.

IREN's record is really two stories: a bitcoin-mining buildout that reached its targets late, and an AI-infrastructure pivot whose biggest commitments now come due. The mining trajectory was consistently optimistic on timing. The November 2021 target of 15.2 EH/s by September 2023 (14393, 14392) is marked delayed. Operating capacity instead grew from 2.3 EH/s (13824) to 3.7 EH/s (13604), then stalled at 5.5 EH/s through spring 2023, with that figure repeated across more than a dozen near-identical claims (12784 through 12698). The goal was then reset to 9.1 EH/s (12465), 9.4 EH/s (12138) and 10 EH/s (6522, 6341). IREN reached 10 EH/s by May 2024 (6150) and reported 20 EH/s installed that September (5689). Capacity did scale substantially, but roughly a year behind the original schedule, and the 31 EH/s Q4 2024 target (5691) is also marked delayed. Facility-level dates show the same pattern: the 80 MW Childress expansion due mid-2024 (6495, 6410) and the 50 MW 2024 build (6153) both slipped. The first GPU deliveries, 568 H100s (6262) and the H200 fleet expansion to 1,896 units (5711, 5712), also missed their stated dates. None of these slips carries an explicit revision in the record; later claims simply restate new figures. The announced power envelope has only expanded. Childress went from 600 MW (12176, 12197) to 750 MW (20301), Sweetwater was introduced at 1,400 MW (6523), and total secured grid-connected power is now 3,000 MW (20305). There are no claims between September 2024 and November 2025, so the record does not show whether the 31 EH/s target was ever met. This dataset contains no independent corroboration: no permits, ERCOT or BC Hydro queue entries, or site reporting beyond the company's own BC Hydro partnership statements (12182, 12198). Every project also lacks populated capacity and cost fields, so the pace cannot be verified from outside evidence here. The claim list also cuts off mid-entry at 20302. In the next 90 days, the key test is the Microsoft contract (8112): 200 MW at Horizon with GPU deployment by December 31, 2026 (20294, 4158, 20309). That deadline falls just after this window closes. Watch for energization and commissioning disclosures at Horizon, progress on the roughly 100,000-GPU deployment (20296), and Dell delivery milestones under the guaranteed purchase agreement (4164). Given IREN's history of unacknowledged slips, watch as well for quiet changes to the Microsoft schedule.

IREN's record shows a mining company whose ambitions consistently ran ahead of its delivery, and which has since restated itself as an AI infrastructure developer at a much larger scale. The first hashrate target of 15.2 EH/s by September 2023 (claim 14393), backed by Bitmain contracts (14392), is flagged as delayed. Actual operating capacity moved from 2.3 EH/s (13824) to 3.7 EH/s (13604), then held at 5.5 EH/s through at least eight near-identical statements in spring 2023 (12827 through 12689). The June 2023 target of 9.1 EH/s (12465) was reset in October to a nearer 7.0 EH/s alongside a 9.4 EH/s goal (12144, 12138). The 10 EH/s milestone was reached by May 2024 (6150), and the 31 EH/s Q4 2024 target (5691) is marked delayed. Most of these targets arrived, but late, and the company rarely said so directly.

Childress follows the same pattern and adds a quiet shrinkage. It was described as a 600 MW site in September 2023 (12176). An 80 MW expansion was promised for mid-2024 (6495, 6410) and slipped. By May 2024 the 2024 build had become 50 MW (6153), which also slipped, and the record never explains the change. The early GPU commitments, 568 H100s (6262) and a 1,896-unit fleet (5712), are also delayed.

After a year with no claims on record, the November 2025 disclosures are a step change. They include a $9.7 billion Microsoft contract (8112), 200 MW of Childress IT load (20310), an additional 450 MW at Childress (20302), 2,000 MW at the Sweetwater Hub, up from 1,400 MW for West Texas in 2023 (20304 against 6523), and 3,000 MW of total secured power (20305). The $5.8 billion Dell supply commitment due March 2026 (4161) is already flagged delayed. The claim describing the Childress construction phases (20295) is cut off in this record, so its timeline cannot be assessed here.

Nothing in the record independently corroborates any of this. There are no permits, no ERCOT large-load queue entries, and no site reporting. Every project, including Childress, which claim 12196 says has operated since April 2023, is still listed as announced with no capacity recorded. Kiowa, the Oklahoma campus and Bundey have no claims at all.

Over the next 90 days, the key test is whether the Horizon deployment for Microsoft stays on its December 2026 date (4158, 20294). The other things to watch are whether the delayed Dell hardware has actually shipped, whether the next quarterly report restates Childress and Sweetwater phasing, and whether Sweetwater energization can be confirmed in ERCOT or county filings.

IREN's record shows capacity ambitions that have repeatedly outrun delivery, although the company has eventually delivered on its bitcoin-mining headlines. The November 2021 target of 15.2 EH/s by September 2023 (claim 14393), along with the matching Bitmain contract (14392), is marked delayed. Operating capacity rose from 2.3 EH/s (13824) to 3.7 EH/s (13604), then held at 5.5 EH/s through at least eight near-identical restatements between February and April 2023 (13337 through 12689). The June 2023 figure of 9.1 EH/s (12465) was reframed by October as 7.0 EH/s of self-mining with a 9.4 EH/s target (12144, 12138). The 10 EH/s mark was reached in May 2024 (6150), and 20 EH/s was reported in September 2024 (5689), roughly a year after the original 15.2 EH/s deadline. The same pattern holds at smaller scale. The 80 MW Childress expansion promised by mid-2024 (6495, 6410), the 50 MW Childress build by year-end 2024 (6153), the 568-GPU H100 commissioning (6262), the 1,896-unit GPU fleet (5712) and the 31 EH/s Q4 2024 target (5691) are all marked delayed, and none was followed by a restated date in the record. Power claims have grown much faster than delivered load. Childress went from 600 MW (12197) to an additional 500 MW (6412) and later an additional 450 MW (20302). The West Texas site grew from 1,400 MW (6523) to 2,000 MW at Sweetwater (20304), rolling up to 3 GW of secured grid-connected power (20305). The record has a fourteen-month gap between September 2024 and November 2025, and then the pivot arrives all at once: the $9.7 billion Microsoft contract (8112), 200 MW for Microsoft by end-2026 (20294, 4158), and a $5.8 billion Dell supply agreement (4161) whose March 2026 date is already marked delayed. The record contains no independent corroboration of any kind, such as permits, ERCOT or BC Hydro interconnection queue entries, or site reporting. The 3 GW figure and the Sweetwater and Childress expansions therefore rest entirely on company statements. The Kiowa, Oklahoma, Bundey and Sweetwater 2 projects carry no dated claims at all, and the supplied record is truncated after claim 20295. Over the next 90 days, the key question is whether the 200 MW Microsoft deployment at Horizon/Childress is energized and accepted by the 2026-12-31 date. Also watch for disclosure on how much of the Dell hardware has actually shipped, and for ERCOT large-load filings or energization notices at Sweetwater 1 that would move the 2 GW figure from assertion toward evidence.

IREN's announced footprint has grown by an order of magnitude, but the growth is mostly a change of unit and ambition rather than a record of delivery against dated promises. The mining-era arc runs from 2.3 EH/s operating (13824) through 5.5 EH/s (12827) and 10 EH/s (6150) to 20 EH/s installed by September 2024 (5689). That real progress still missed the original 15.2 EH/s target for September 2023 (14393), and the 31 EH/s promised for Q4 2024 (5691) is also marked delayed. On the power side, Childress went from 600 MW potential (12176) to 750 MW total (20301), Sweetwater from 1,400 MW (6523) to 2,000 MW (20304), and the secured North American portfolio now stands at 3,000 MW (20311). GPU ambitions went from 568 H100s (6261) to a 1,896-unit fleet (5712) to a stated 100,000-unit deployment scale (20296). The headline numbers have only ever moved up.

Dated facility promises have held poorly. The 80 MW Childress expansion was described as under construction in June 2023 (12449) and then promised for mid-2024 (6495, 6410). Both of those claims are now delayed. By May 2024 the Childress build for that year was being described as 50 MW by year-end (6153), which is also delayed. That looks like a quiet downsizing of the near-term tranche that was never reconciled with the earlier 80 MW figure. The first GPU commissioning target (6262) and the H200 fleet expansion (5712) slipped as well. The record also has a gap from September 2024 to November 2025, so it is unclear how the pivot from mining to AI hosting changed which Childress capacity was prioritized.

Every item in this record is company-sourced. It contains no permits, ERCOT or BC Hydro interconnection-queue entries, or site reporting that would independently confirm the 750 MW Childress or 2,000 MW Sweetwater figures. Energization of large-load capacity in Texas is visible in ERCOT filings, so the absence of that evidence here is a gap to close, not proof of anything. The supplied record also ends mid-entry at claim 20305.

Over the next 90 days, the decisive test is whether the 200 MW of Childress IT load tied to the $9.7 billion Microsoft contract (8112, 20294, 20295) is energized and handed over by the December 31, 2026 target. Watch for a phased partial delivery framed as on schedule, any change to the Microsoft revenue-start language, and whether Sweetwater 1 shows interconnection milestones that match its 2,000 MW headline.

IREN's record shows a company whose headline capacity has grown with each re-framing, while dated delivery promises have often lagged. The original 2021 plan targeted 15.2 EH/s by September 2023 (claim 14393, backed by the Bitmain contracts in 14392). Both are now marked delayed. By April 2023 the company was reporting 5.5 EH/s, repeated almost verbatim across a dozen filings (12784 through 12698). The goal was then reset rather than reconciled: 9.1 EH/s in June 2023 (12448), then 7.0 and 9.4 EH/s on the same day in October 2023 (12137, 12138), then 10 EH/s (6522). That figure was reached by May 2024 (6150), and 20 EH/s was installed by September 2024 (5689). This is real growth, but it arrived against a moving target rather than the one first set. The 31 EH/s promised for Q4 2024 (5691) is marked delayed, and the record has no later statement confirming it.

Power-side figures follow the same pattern. Childress went from 600 MW of potential (12176) to 750 MW (20301). Meanwhile, the dated 80 MW expansion due by mid-2024 (6495, 6410) and the 50 MW build due by end-2024 (6153) are both marked delayed. The 1,400 MW West Texas figure (6523) has appeared without an interim milestone. On the AI side, the first GPU tranches slipped: the 568 H100s were due mid-2024 (6262), and the H200 purchase and 1,896-unit fleet were due by end-2024 (5711, 5712). The contracted position then jumped sharply to the $9.7 billion Microsoft agreement (8112) and a 100,000-GPU deployment scale (20296). None of the delayed dates has a matching claim in which the company acknowledges or reschedules them. The record also has a gap between September 2024 and November 2025, and it is truncated after claim 20304.

Every claim in this file is company-stated. The record contains no permits, ERCOT large-load interconnection filings, BC Hydro service agreements, or on-the-ground reporting to confirm the pace at Childress, Sweetwater, or the British Columbia expansion (20303). That independent evidence is what separates announced megawatts from energized ones, and it is missing here.

The next 90 days bring the most consequential deadline in the file. Horizon's 200 MW for the Microsoft deployment is promised by 31 December 2026 (4158, 4159, 20294, 20295). Watch for three things: energization and commissioning disclosures at Horizon, any change to Microsoft revenue-recognition timing, and whether the company reports Childress construction phases in delivered megawatts rather than potential megawatts. Given the history, a quiet restatement is more likely than an announced slip, so any wording change in quarterly filings deserves close reading.

IREN's capacity story has grown in two distinct stages, and the second is much larger than the first. The bitcoin-mining stage started at a 15.2 EH/s target (14393). That target was then presented as a series of incremental milestones: 2.3 (13824), 3.7 (13604), 5.5 (12712), 9.1 (12448), 10 (6341), and a 20 EH/s pathway (6342), before an installed 20 EH/s was stated in September 2024 (5689) alongside a 31 EH/s Q4 goal (5691). The site-level power figures grew faster than the operating ones. Childress went from 600 MW (12176) to 750 MW (20301). A separate 1,400 MW West Texas figure tied to Sweetwater (6523) appeared with no interim steps. The second stage, AI compute, moves the story to a different scale: a 200 MW Horizon IT load (4159, 20294), a $9.7 billion Microsoft contract (8112), and a 100,000-GPU deployment figure (20296). Many claims repeat one disclosure across several releases, so the record looks denser than the underlying commitments are.

Dated promises have not held well. The original 15.2 EH/s target for September 2023 (14393) was reached roughly a year late. The 80 MW Childress expansion due by mid-2024 (6495, 6410), the 50 MW Childress build for end-2024 (6153), the 568-GPU H100 commissioning (6262), and the 1,896-unit GPU fleet (5712) are all marked delayed. None of these slips shows up in the record as a revised date. Later disclosures instead move on to larger numbers. The record also has no entries between September 2024 and November 2025, so it cannot show whether the 31 EH/s target (5691) was met.

The supplied record contains no independent corroboration: no ERCOT or BC Hydro interconnection queue positions, no county or state permits, and no on-the-ground construction reporting for Horizon, Sweetwater, or the Oklahoma and Kiowa entries. Every project on file is still listed as announced, with no capacity or cost filled in. On this record alone, the pace is asserted, not evidenced. The input also ends partway through the November 2025 disclosures (after 20304), so later claims may exist that this read does not cover.

The next 90 days end at the 2026-12-31 deadline for the Microsoft deployment and the 200 MW Childress construction phases (4158, 20295). Watch whether IREN's upcoming quarterly update reports energized Horizon capacity in megawatts and GPUs accepted by the customer, not just GPUs delivered. Also watch whether the Dell-guaranteed hardware (4164) is shown as installed, and whether any Sweetwater interconnection milestone surfaces in ERCOT filings. A quiet shift of the Microsoft schedule into 2027, like the unacknowledged slips in 2024, would be the main signal to flag.

IREN's record reads as a company whose headline numbers have grown faster than its delivery against dated promises. On hashrate, the 15.2 EH/s target set in late 2021 for September 2023 (14393) was flagged as delayed, and the company spent most of 2023 restating a 5.5 EH/s installed base (12827, 12712, 12689) while moving its next target from 9.1 (12448) to 9.4 (12138) to 10 EH/s (6522). It did reach 10 EH/s by May 2024 (6150) and reported 20 EH/s by September 2024 (5689), but the 31 EH/s Q4 2024 goal (5691) is marked delayed. A later 50 EH figure (4034) arrives without any reconciliation of that miss. Power ambitions scaled more sharply still: Childress at 600 MW (12176), Sweetwater 1 at 1,400 MW (6523), Sweetwater 2 at 600 MW (20252). None of the eleven projects on record carries a stated capacity or cost at the project level, so these site figures cannot yet be tied to specific builds.

Dated facility commitments have slipped consistently. The 80 MW Childress expansion due mid-2024 (6495, 6410), the 50 MW Childress build due end-2024 (6153), the H100 commissioning (6262), the H200 fleet expansion (5712, 5711), and the Dell GPU supply due March 2026 (4161) are all recorded as delayed. The pattern is quiet revision rather than public reset: later disclosures pivot to larger AI-cloud framing, including the $9.7 billion Microsoft contract (8112) and 200 MW Horizon (4159, 20250), without restating the missed milestones. Some dollar fields, such as 12208 and 12136, appear to carry unit errors and should not be read literally.

The record contains no independent corroboration of any kind. There are no permits, no ERCOT or BC Hydro interconnection-queue entries, and no on-the-ground reporting. The announced pace is therefore supported only by company statements. The most pressing gap is Sweetwater 1, whose 1,400 MW substation energization was promised by April 30, 2026 (20251). Five months later, it remains unconfirmed.

Over the next 90 days, watch four things. First, whether IREN confirms Sweetwater 1 energization or revises its date. Second, whether Dell deliveries and Horizon deployments track the Microsoft schedule due by end-2026 (4158). Third, progress on the British Columbia GPU transition, which is also due by end-2026 (20249). Fourth, whether ERCOT filings or local reporting surface that can independently anchor the Sweetwater and Childress megawatt figures.

IREN's record shows a company whose unit of ambition has changed more than once. Early claims were measured in exahashes. The November 2021 target of 15.2 EH/s by September 2023 (14393) is marked delayed, and the April 2023 installed figure sat at 5.5 EH/s (12784, 12712). The ladder then climbed through 9.4 (12138) and 10 EH/s (6150) to 20 EH/s installed in September 2024 (5689), and 50 EH/s by late 2025 (4034, 4021). The Q4 2024 target of 31 EH/s (5691) is flagged delayed, but the later 50 EH/s claim implies the shortfall was eventually overtaken rather than acknowledged. The more consequential shift is to power and GPUs. Site ambitions grew from 600 MW at Childress (12186) to 1.4 GW in West Texas (6523) and 1.6 GW in Oklahoma (16667). The GPU count moved from 568 units (6261) to roughly 2,100 in November 2025 (4033), then to 99,900 installed or on order by March 2026 (3422). That last figure blends delivered and contracted hardware, which makes it a weak measure of physical progress.

Dated promises have fared poorly. The 80 MW Childress expansion due mid-2024 (6495, 6410), the 50 MW Childress build for 2024 (6153), the early GPU commissioning window (6262), and the H200 purchase (5711, 5712) are all recorded as delayed. None of these has a corresponding restatement in the record. The Dell supply agreement tied to March 2026 (4161) is also marked delayed. That matters because it feeds the Microsoft contract (8112), whose GPU services deployment at Horizon is scheduled through year-end 2026 (4158) on a 200 MW IT load (4159).

No independent corroboration appears in this record. There are no ERCOT or SPP interconnection entries, county permits, or site reporting to test the Sweetwater and Oklahoma figures, so those numbers should be treated as company-sourced only. The Oklahoma ramp runs to 2028 (16668), which leaves ample room for silent revision.

Over the next 90 days, the key test is whether Horizon's Microsoft capacity is energized and billing by the 2026-12-31 date in claim 4158, and whether IREN confirms that the Dell hardware under 4161 has actually shipped. Watch whether the 99,900-unit figure is split into installed and ordered counts. For Sweetwater and Oklahoma, watch for interconnection or permitting filings; their absence would suggest the gigawatt totals are pipeline rather than schedule.

IREN's record shows headline figures that have grown faster than its delivery dates have held. The earliest forward target, 15.2 EH/s by September 2023 (claim 14393), is marked delayed. Operating hashrate instead climbed in steps: above 2.3 EH/s (13824), then 3.7 EH/s (13604), then a 5.5 EH/s figure restated across roughly ten filings from February to April 2023 (13348 through 12689), then 9.1 EH/s tied to an 80 MW build (12465). Later marks of 10 EH/s (6150), 20 EH/s (5689) and 50 EH/s (4034) show real growth. However, the Q4 2024 target of 31 EH/s (5691) is also delayed, and the record contains no claim between September 2024 and November 2025 that explains the bridge.

Power ambition has scaled more sharply than execution. Childress went from an 80 MW expansion due mid-2024 (6495, 6410, both delayed) and a 50 MW build due by end-2024 (6153, delayed) to 600 MW of site potential (12186). Separate announcements added 1,400 MW in West Texas (6523) and 1,600 MW in Oklahoma (16667). GPU counts moved from 568 (6261) to 1,896 (5712) to 2,100 (4033), then to 99,900 units installed or on order (3422) about three months later. That last figure blends delivered and contracted hardware and should not be read as operating capacity. Several promised dates passed without a revised date appearing anywhere in the record, and that silent slippage is the pattern that deserves the most scrutiny.

Independent corroboration is absent from this dataset. None of the eleven listed projects carries a stated capacity or cost. There are no permit filings, ERCOT or other interconnection-queue entries, or on-the-ground reports to test the announced pace against. Until such evidence appears, the Sweetwater, Oklahoma and Horizon figures remain company-sourced.

Over the next 90 days, the central test is Horizon. The Microsoft GPU deployment schedule runs to year-end 2026 (4158), against a 200 MW IT load (4159) that underpins the $9.7 billion contract (8112). The Dell supply agreement, already past its March 2026 date (4161), is the gating dependency. Watch for three things: disclosure of how many of the 99,900 GPUs are actually energized, any restated Horizon milestones, and interconnection or permitting filings for Oklahoma ahead of its 2028 ramp (16668).

IREN's facility record traces two distinct growth curves, and the second is far steeper than anything the first ever delivered. On the bitcoin side, the original 15.2 EH/s target for September 2023 (claim 14393) missed its date, yet operating capacity did scale steadily afterward: 5.5 EH/s through spring 2023 (12784, repeated across more than a dozen near-identical filings), 10 EH/s by May 2024 (6150), 20 EH/s by September 2024 (5689), and 50 EH/s by late 2025 (4034, 4021). The trajectory is real, but the dates are consistently optimistic; the 31 EH/s Q4 2024 target (5691) is itself marked delayed. The AI pivot moves in a different order of magnitude. GPU counts rose from 568 H100s (6261) to 1,896 (5712) to 2,100 (4033), then jumped to 99,900 units "installed or on order" by March 2026 (3422). That phrasing merges delivered hardware with purchase commitments, and the gap between the two is the central question for this company. Announced power follows the same escalation: Childress at 600 MW (12186) plus 500 MW more (6412), Sweetwater at 1,400 MW (6523), and an Oklahoma campus at 1,600 MW with a ramp running to 2028 (16667, 16668).

Timelines have slipped repeatedly and without restatement in the record. The 80 MW Childress expansion due June 2024 (6495, 6410), the 50 MW Childress build for end-2024 (6153), the H100 commissioning (6262), and the $5.8 billion Dell GPU agreement due March 2026 (4161) are all flagged delayed, and none carries a revised date. Several named projects, including Kiowa, Bundey, Sweetwater 2 and the Sweetwater Hub, have no capacity or timeline claims attached at all.

This record contains no independent corroboration. There are no permits, no ERCOT or SPP interconnection queue entries, and no site reporting, so every megawatt figure rests on company statements. That matters most for Sweetwater and Oklahoma, where the announced scale runs well ahead of demonstrated delivery.

Over the next 90 days, the key test is the Horizon deployment for Microsoft, scheduled through December 2026 (4158) and backed by the $9.7 billion contract (8112). Watch for evidence that Dell hardware has actually shipped, for any breakout of installed versus ordered GPUs, for energization at Sweetwater, and for interconnection filings that would anchor the Oklahoma ramp in something beyond press releases.

IREN's record traces a shift from bitcoin self-mining to AI compute, with capacity figures that have grown far faster than the company's record of meeting its own dates. Hashrate claims rise from just over 2.3 EH/s (13824) to 5.5 EH/s (12784), 10 EH/s (6150), 20 EH/s (5689) and 50 EH/s (4034). Site-power claims grow even more steeply: Childress at 600 MW (12186), then 500 MW of additional capacity there (6412), a 1,400 MW West Texas development tied to Sweetwater (6523), and a 1,600 MW Oklahoma campus (16667). GPU claims move from 568 H100s (6261) to 1,896 units (5712) and 2,100 (4033), then to 99,900 'installed or on order' (3422). That last metric combines deployed and contracted hardware, so it measures commitments rather than delivered capacity.

Delivery dates have repeatedly slipped. The 15.2 EH/s target for September 2023 (14393) was missed, and the fleet stood near 5.5 EH/s that spring (12712). The 10 EH/s mark was not claimed until May 2024 (6150). The 80 MW Childress expansion promised for mid-2024 (6495, 6410), the 50 MW Childress build for end-2024 (6153), the 31 EH/s Q4 2024 target (5691) and the H200 purchase (5711) are all flagged delayed. None of these misses was followed by a claim acknowledging a revised date. Instead, each new announcement set a larger headline number, most visibly the 50 EH/s figure (4034), which appeared without any reconciliation to the missed 31 EH/s milestone. The Dell supply agreement, due by March 2026 (4161), is also delayed. The record contains no claims between September 2024 and November 2025, so the company's performance over that period cannot be assessed from this dataset. Many entries are also near-duplicates (12784, 12827, 12712, 12726), which makes the record look more active than the underlying progress.

This record contains no independent evidence: no permits, no ERCOT or SPP interconnection queue entries, and no on-the-ground reporting for Sweetwater, Horizon or the Oklahoma campus. The multi-gigawatt figures therefore rest entirely on company statements. The operational histories for Mackenzie, Prince George and Childress (12194, 12195, 12196) are the most verifiable claims and are also the smallest.

The key test in the next 90 days is the Microsoft GPU Services deployment at Horizon, scheduled for end-2026 (4158), since the $9.7 billion contract (8112) depends on it. Watch whether the delayed Dell hardware is disclosed as received, whether the 99,900 figure is broken out into installed and on-order units, and whether filings show interconnection or permitting progress for Sweetwater and Oklahoma ahead of the 2028 ramp (16668).

IREN's stated ambitions have grown by an order of magnitude, but most of that growth sits in forward-looking figures rather than delivered capacity. The mining story is easiest to follow. The November 2021 target of 15.2 EH/s by September 2023 (14393), backed by the Bitmain contracts in 14392, was missed; operating capacity sat at 5.5 EH/s through spring 2023 (12784, 12698). Guidance was then reset in smaller steps: 9.1 (12448), 9.4 (12138), 10 (6522). The 10 EH/s figure was reported as reached in May 2024 (6150), 20 EH/s by September 2024 (5689), and 50 EH/s by late 2025 (4034, 4021). The original target was roughly a year late, and after that the company largely met its incremental targets. The exception is the 31 EH/s year-end 2024 goal (5691), which is flagged as delayed.

The timelines for power and buildings have been less reliable. The 80 MW Childress expansion due by mid-2024 (6495, 6410) and the 50 MW Childress build for 2024 (6153) are both flagged as delayed. No later claim in the record acknowledges either slip or gives a new date. Meanwhile, the stated site envelopes kept growing: 600 MW at Childress (12186), 1.4 GW in West Texas (6523), and now 1.6 GW in Oklahoma (16667), with a ramp that runs to 2028 (16668). The AI pivot follows the same shape. The GPU fleet went from 568 H100s (6261, delivery flagged delayed in 6262) to 1,896 (5712, also delayed), then 2,100 (4033), then 99,900 "installed or on order" (3422). That last figure blends delivered and contracted units, so it should not be read as operating capacity.

The record contains no independent corroboration: no ERCOT or Oklahoma interconnection queue entries, no building permits, and no site reporting. Sweetwater 2, Kiowa and Bundey have no capacity or timeline claims at all. All of the pace assessment above therefore rests on company statements, and the gigawatt-scale figures should be treated as site potential, not committed load.

The next 90 days bring a real test. The Microsoft deployment at Horizon (4158) is due by December 31, 2026. It is tied to the 200 MW Horizon IT load (4159) and to the Dell supply agreement (4161), whose March 2026 milestone is already flagged as delayed. Three things are worth watching: whether IREN reports energized megawatts at Horizon rather than GPUs on order, whether revenue recognition under the $9.7 billion contract (8112) begins on schedule, and whether any interconnection or permitting filings for Sweetwater or Oklahoma surface to support the 2028 ramp.

IREN's headline capacity has grown by an order of magnitude, but the path has been uneven and the unit of ambition has changed. Early mining targets ran well ahead of delivery: the 15.2 EH/s goal set for September 2023 (14393) came due while operating capacity sat near 5.5 EH/s (12827), and it was reached only in 2024. After that the curve steepened. Capacity went from 10 EH/s in May 2024 (6150) to 20 EH/s by September (5689), then to 50 EH/s by late 2025 (4034, 4021). The latest figure is 46 EH/s installed (3423). That is the first reported contraction, and the record does not explain it, though it fits a reallocation of power toward GPU workloads. Power ambitions have scaled faster still. Childress went from 600 MW (12176) to Sweetwater at 1,400 MW (6523) and now an Oklahoma campus at 1,600 MW (16667). On the GPU side, Horizon's 200 MW IT load (4159) is anchored by a $9.7 billion Microsoft contract (8112).

Near-term dates have slipped consistently. The 80 MW Childress expansion due mid-2024 (6495, 6410), the 50 MW Childress build due end-2024 (6153), H100 commissioning (6262), the H200 purchase and the 1,896-GPU fleet target (5711, 5712), and the 31 EH/s Q4 2024 target (5691) are all flagged delayed. Some targets were also revised without acknowledgement. The October 2023 releases gave both 7.0 and 9.4 EH/s (12137, 12138) against the earlier 9.1 EH/s plan (12465), with no reconciliation. The Dell supply agreement tied to March 2026 (4161) is also marked delayed.

This record contains no permits, interconnection-queue filings, or site reporting for any project. Most project entries lack capacity or cost fields entirely, so neither the Sweetwater and Oklahoma megawatt figures nor the 2028 Oklahoma ramp (16668) can be checked independently. Operational claims for the British Columbia sites (12194, 12195) and Childress (12196) are self-reported.

Over the next 90 days, the test is Horizon: whether GPU services for Microsoft deploy by the end-2026 date (4158), which depends on the delayed Dell hardware flow. Watch for an explanation of the drop from 50 to 46 EH/s, and for ERCOT or Oklahoma utility filings that would give the Sweetwater and Oklahoma megawatt claims some independent footing.

IREN's record reads as two distinct delivery stories stitched together, and the gap between them is widening. The bitcoin mining arc is the better-evidenced one. The 15.2 EH/s target set in late 2021 for September 2023 (claim 14393) was missed by a wide margin: stated operating capacity sat at 5.5 EH/s through spring 2023 (12827, 12698). The company then reset the goal in steps, first to 9.1 EH/s (12465), then 9.4 (12138), then 10 EH/s (6522). The 10 EH/s level was eventually stated in May 2024 (6150), and 20 EH/s by September 2024 (5689). None of these resets explicitly acknowledges the original 2023 date. The Q4 2024 target of 31 EH/s (5691) is flagged delayed, yet the fleet later reached 50 EH/s (4034, 4021). Mining capacity did scale, but on a schedule set by later announcements rather than the first ones.

The facility layer shows the same pattern. The Childress 80 MW expansion due by mid-2024 (6495, 6410) and the 50 MW build due by end-2024 (6153) both slipped. Meanwhile the headline power figures kept rising: 600 MW at Childress (12186), a further 500 MW (6412), 1,400 MW at Sweetwater (6523), and now a 1,600 MW Oklahoma campus with a ramp running to 2028 (16667, 16668).

The AI pivot carries the largest gap between ambition and demonstrated delivery. The stated GPU fleet grew only from a targeted 1,896 units (5712, itself delayed) to 2,100 (4033). Against that base, the company has announced a $9.7 billion Microsoft contract (8112), a $5.8 billion Dell supply agreement that has already missed its March 2026 date (4161), and a 50,000-unit B300 purchase (11693). That is roughly a twenty-fold scale-up with no interim proof point on record.

The record contains no independent corroboration. There are no permits, no ERCOT or SPP interconnection queue entries, and no on-the-ground construction reporting. Every capacity figure here is self-reported and should be weighted accordingly.

Over the next 90 days, the decisive test is the Horizon deployment for Microsoft, promised by 31 December 2026 (4158, 4159). Watch for evidence of energized IT load at Childress, confirmation of Dell hardware delivery and acceptance, financing disclosures tied to the B300 order, and any Oklahoma or Sweetwater filings in the relevant interconnection queues.

IREN's announced scale has grown by orders of magnitude, but the unit of ambition has shifted. The 2021 plan of 15.2 EH/s by September 2023 (14393), backed by 14.5 EH/s of Bitmain contracts (14392), was not met on schedule. Reported capacity climbed from 2.3 EH/s (13824) to 5.5 EH/s (12827) and 10 EH/s by May 2024 (6150), then accelerated to 20 EH/s (5689) and 50 EH/s by late 2025 (4034, 4021). The 3423 filing reports 46 EH/s installed, the first reported decline. It coincides with the pivot to GPU cloud, so it reads as a reallocation of power rather than an operational setback, but the company has not framed it that way. Power claims have followed the same arc: Childress grew from an 80 MW build (12449) to 600 MW of potential (12197), Sweetwater was introduced at 1,400 MW (6523), and Oklahoma at 1,600 MW (16667).

Dated promises have held poorly. The 80 MW Childress expansion due mid-2024 (6495, 6410), the 50 MW Childress build due end-2024 (6153), the H100 commissioning (6262), the H200 purchase (5711), the 31 EH/s Q4 2024 target (5691) and the Dell GPU supply milestone of March 2026 (4161) are all flagged delayed. Later disclosures move on to new totals rather than reconciling these misses. Delivery has tended to arrive, but later than stated, and the company has not acknowledged revisions.

This record contains no independent corroboration: no ERCOT or SPP interconnection queue positions, no county or state permits, and no on-the-ground reporting for Sweetwater, Horizon or Oklahoma. The megawatt figures are therefore company assertions only. Oklahoma's 2028 power ramp (16668) especially should be treated as aspirational until a utility agreement or queue filing appears.

For the next 90 days, the key test is whether Horizon's 200 MW IT load (4159) is on track to support the Microsoft deployment due by end-2026 (4158), which underpins the $9.7 billion contract (8112). Watch for confirmation that the delayed Dell hardware has been delivered, any Sweetwater 1 energization date, Oklahoma interconnection or permitting filings, and whether hashrate falls further as mining sites are converted.

The scaling story divides cleanly into two regimes. Through 2021-2024 the unit of ambition was hashrate, and the first big number did not survive contact with delivery: the 15.2 EH/s anticipated for September 2023 (claim 14393), underwritten by binding Bitmain contracts (claim 14392), met an actual base of 2.3 EH/s in August 2022 (claim 13824) and roughly 9.1 EH/s by mid-2023 (claims 12443, 12465). Both 2021 claims carry a delayed flag, but the record shows no moment where the shortfall was named; instead the target was re-based forward — 9.4, then 10, then a pathway to 20 (claims 12138, 6341, 6342), then 31 by Q4 2024 (claim 5691, delayed), arriving at 50 EH/s only in late 2025 (claims 4034, 4021). Roughly four years late against the original curve, reached by moving the goalposts rather than acknowledging them.

From late 2025 the denominator shifts to megawatts and contracted revenue, and the numbers get much larger: Childress at 600 MW (claims 12176, 12186), Sweetwater at 1400 MW (claim 6523), Horizon at 200 MW of combined IT load (claim 4159), and Oklahoma at 1600 MW with a power ramp stretching to 2028 (claims 16667, 16668). The Microsoft contract (claim 8112) and the Dell supply agreement (claim 4161) give this phase counterparty validation the mining era lacked — but claim 4161 is already flagged against its March 2026 date.

On independent corroboration, the honest answer is that this record does not contain any. There are no permit filings, interconnection queue positions, or site-visit reports here; the only external anchors are commercial counterparties and the BC Hydro connection (claims 12182, 12198), all of which are announcement-side rather than verification-side. Note also the structural tell: the overwhelming majority of capacity claims carry no promised-by date at all, while nearly every claim that does carry one is flagged delayed. Undated capacity is unfalsifiable capacity.

For the next 90 days, the live tests are the Microsoft GPU Services deployment schedule due 31 December 2026 (claim 4158) and the H2 2026 GPU deployment phases (claim 11694). Watch whether the year-end deadline is confirmed, quietly restated as a 2027 ramp, or left undated — the last of those would be the most informative outcome.

IREN's record divides cleanly into two eras, and the delivery discipline differs sharply between them. The 2021–2024 bitcoin phase opened with a forward hashrate promise that arrived roughly two years late in substance (14392, 14393), and the interval between promise and delivery was filled with restatement rather than progress: the same 5.5 EH/s figure was issued across at least eight separate communications from February through April 2023 (13337, 13348, 12827, 12784, 12726, 12712, 12699, 12698, 12689) before any new number appeared. That pattern — re-announcing a static figure as though it were news — is the single most reliable tell in this dataset, and it recurred around the Childress 80MW block, which was announced, re-announced, and then carried a mid-2024 date that slipped (6495, 6410). Capacity itself did scale, and substantially, from low single-digit EH/s to the twenties and then toward fifty (5689, 4034, 4021). But the most recent installed figure sits below the headline capacity figure that preceded it (3423 versus 4021), and no acknowledgement of that step-down appears in the record. It may be a definitional shift between nameplate and installed; it should be treated as unexplained until the company explains it.

The post-2025 AI phase has larger numbers and thinner verification. The Microsoft agreement (8112) and the Dell supply commitment (4161) are counterparty-anchored, which is meaningful, but the Dell obligation already carries a slipped date, and the demonstrated GPU fleet (4033) remains small relative to the dollar figures attached to it. Power announcements have outrun everything else: Childress at 600MW (12176), West Texas at 1,400MW (6523), Oklahoma at 1,600MW with a ramp reaching 2028 (16667, 16668). The record contains no permit filings, interconnection queue positions, or independent site reporting — only the BC Hydro connection (12198) predates the Texas and Oklahoma scale-up. Absent those, announced pace is unverified rather than contradicted.

Over the next ninety days, the Horizon deployment date (4158) is the binding test, since it falls inside the window and is tied to a named customer. Watch also whether installed hashrate returns to the fifty figure, whether Dell deliveries are re-dated, and whether Oklahoma produces any utility or county filing to substantiate a 2028 ramp.

The shape here is a company whose ambition has compounded far faster than its delivery cadence, but whose disclosure has been structured in a way that makes that gap hard to measure. The earliest dated commitment on record (14393, with its hardware underpinning in 14392) set a hashrate target for late 2023 that the operating figures of that period (12827, 12726, 12698) show was not met on time; it was effectively reached a year or more late, and then vaulted past. Scaling since has been genuine in direction — the power narrative moves from a single 600 MW Texas site (12176, 12197) to a 1,400 MW West Texas ambition (6523) and a 1,600 MW Oklahoma campus (16667) — but the announced footprint has grown roughly an order of magnitude while the project register itself carries no populated capacity or cost against any of the eleven named sites, which is a disclosure gap worth naming.

On dates, the pattern is consistent and narrow: when IREN attaches a deadline, it tends to slip. Childress tranches (6495, 6410, 6153), miner shipment (6408), GPU commissioning (6262), the H200 purchase and resulting fleet count (5711, 5712), and the Q4 2024 capacity step (5691) are all flagged late. Most other claims carry no promise date at all, which renders them unfalsifiable rather than met. Mid-2023 targets were also restated at shifting values across adjacent disclosures (12465, 12138, 6493) without visible reconciliation.

Nothing in this record is independent. There are no permits, interconnection queue entries, or third-party site observations attached to any claim, so the pace rests entirely on company statements; ERCOT queue positions for the Sweetwater and Childress assets and Oklahoma's ramp filings against 16668 are the obvious external checks.

Over the next ninety days, watch whether the Dell supply agreement (4161), already past its stated date, is confirmed closed; whether the Microsoft deployment schedule (4158) — the first commitment with a counterparty able to enforce it — shows interim milestones against the 200 MW Horizon load (4159); and whether the 99,900 GPU figure (3422) is ever split into installed versus merely ordered. The record has also been silent since early March.

The arc here is one of consistent directional honesty paired with consistently optimistic timing. IREN's earliest capacity commitment (14392, 14393) targeted 15.2 EH/s by September 2023; the company was in fact operating around 5.5 EH/s through the spring of that year (12827, 12726), a shortfall of roughly two years rather than two quarters. What follows is not retrenchment but serial upward revision — 9.1, then 9.4, then 10, then 20 and 31, and by late 2025 a stated 50 EH/s (4034, 4021). Power ambition scaled faster still, from 80 and 100 MW Childress tranches (12449, 6496, 6413) against a 600 MW site envelope (12176), to a 1,400 MW West Texas footprint (6523), to a 1,600 MW Oklahoma campus (16667). The announced portfolio has grown by roughly two orders of magnitude while the actual delivery unit has never exceeded the 50 to 100 MW tranche (6153, 6495).

Dates are where the record is weakest. The 80 MW Childress expansion (6495, 6410), the T21 shipments (6408), the 568-GPU commissioning (6262), the 50 MW 2024 build (6153), the H200 purchase and 1,896-unit fleet (5711, 5712), the 31 EH/s Q4 2024 step (5691), and the Dell supply agreement (4161) all carry missed dates. The pattern is not denial so much as displacement: a subsequent release announces a larger number, and the prior deadline is never reconciled against it. Readers tracking the headline figure would not notice; readers tracking the calendar would.

Independent corroboration is largely absent from this record, which is itself the finding. There are no permits, no interconnection queue entries, and no third-party site reporting in the file. What substitutes is counterparty exposure — the BC Hydro connection (12198), the PodTech acquisition (12181, 12192), the Microsoft contract (8112) and the Dell parent guarantee (4164) — commitments that carry external signatures and are harder to quietly restate.

Over the next ninety days, the load-bearing item is the Microsoft GPU deployment schedule at Horizon, due December 2026 (4158, 4159). Watch whether the 99,900 GPUs on order (3422) begin converting to installed units, whether the Dell slippage past March 2026 is ever addressed directly, and whether Oklahoma shows early works consistent with a 2028 ramp (16668). Note also that this record terminates mid-entry at 11694, leaving the most recent quarter unverified.

The shape of IREN's record is a company that has repeatedly missed the near term while being roughly right about the far term, and has rarely marked the difference. The founding promise of 15.2 EH/s by September 2023 (14393), underwritten by what was described as binding Bitmain tonnage (14392), arrived late and light; the company was still restating 5.5 EH/s as current operating capacity across at least nine separate disclosures between February and April 2023 (13348, 13337, 12827, 12784, 12726, 12712, 12699, 12689, 12698), a plateau presented each time as movement. Ten EH/s, the 2024 target (6341), landed only in May 2024 (6150) — eight months past the date attached to a figure fifty percent larger. What followed, though, was genuine: 20 EH/s by September 2024 (5689) and 50 EH/s by late 2025 (4034, 4021). The trajectory bent upward; the dated promises along the way did not hold.

The quiet revisions sit at the construction layer. Childress 80MW was due first half of 2024 (6495, 6410), and by May 2024 the full-year Childress target had been reframed as 50MW (6153) without reconciling the earlier 80MW and 100MW tranches (6496, 6413). Meanwhile the land-and-power number inflates faster than anything energized — 600MW at Childress (12176, 12186), 1,400MW at Sweetwater (6523), 1,600MW in Oklahoma (16667) with no intermediate milestone before a 2028 ramp (16668).

Independent corroboration is uneven by site. The retroactively stated energization dates for Canal Flats, Mackenzie, Prince George and Childress (12193, 12194, 12195, 12196) are checkable against BC Hydro interconnection and ERCOT records and have not been contradicted. Sweetwater 1 carries a specific street address, which implies filed permits; Kiowa and the Oklahoma campus carry no address at all, and should be read as queue positions rather than projects until one appears.

Through mid-December, three things resolve. The Microsoft deployment schedule expires 31 December 2026 (4158) against Horizon's 200MW IT load (4159); slippage there is contractual, not presentational. The Dell agreement passed its March 2026 mark (4161), so reconcile delivered GPUs against the 2,100 figure (4033). And Mackenzie's H2 2026 GPU phases (11694) fall inside the window, offering the cleanest near-term test of whether execution now matches announcement.

IREN's disclosure history shows a company that consistently arrives at its announced numbers, but later than the calendar it first published, and with the unit of measure quietly changing underneath. The earliest commitment, 15.2 EH/s underwritten by binding Bitmain contracts by September 2023 (claims 14393, 14392), was missed by a wide margin; the fleet was still at 5.5 EH/s through most of spring 2023, a figure restated almost verbatim across at least six releases in April alone (12784, 12712, 12699, 12689). That repetition is itself a signal: capacity announcements were recycled as news during a period when the physical build was flat. The number the company promised for 2023 was not reached until roughly 2024 (6150), and then overshot decisively — 20 to 31 EH/s in late 2024 (5689, 5691) and 50 EH/s by late 2025 (4034, 4021). Direction of travel is up and the scaling is real; the dates are the weak part.

The pattern with dates is slippage of one to two quarters, absorbed silently at the next quarterly update rather than acknowledged. Childress 80 MW promised for June 2024 (6495, 6410), the 50 MW 2024 tranche (6153), the 31 EH/s Q4 target (5691), the H200 purchase and 1,896-GPU fleet (5711, 5712), and the initial 568-GPU commissioning (6262) all carry that signature. The same mechanism is now live on a far larger balance: the Dell agreement (4161) is already behind its March 2026 date, and it is the physical input to both the Horizon 200 MW IT load (4159) and the Microsoft deployment schedule running to December 2026 (4158).

The record supplied here is entirely company-sourced. Nothing in it is corroborated by permits, ERCOT or BC Hydro interconnection filings, or site reporting, and the BC assets (12192, 12181) were acquired with capacity already standing rather than built. Over the next ninety days, watch whether Horizon energizes on the 2026 schedule despite the Dell slip, whether the 99,900-GPU figure (3422) is disaggregated into installed versus ordered, whether Sweetwater 2 and Sweetwater Hub ever receive named capacity of their own, and whether Oklahoma's 1.6 GW (16667) is matched to a queue position consistent with a 2028 ramp (16668).

The shape of this record is a company whose announced denominator has grown far faster than its delivered increments. The 2021 commitment was 15.2 EH/s of hashrate underwritten by binding Bitmain contracts by September 2023 (14393, 14392); the actual path ran 2.3 (13824), 3.7 (13604), then stalled at 5.5 EH/s through at least seven separate restatements between February and April 2023 (13337, 13348, 12827, 12784, 12712, 12726, 12699, 12698, 12689). That cluster is worth naming plainly: the same number re-announced repeatedly is not progress, and the original target was missed by roughly a year before being superseded rather than reconciled. Power framing escalated similarly, from an 80 MW build (12461, 12444) to 600 MW at Childress (12176, 12186), 1,400 MW at Sweetwater (6523), and 1,600 MW in Oklahoma (16667) — a twentyfold expansion of the headline while construction still lands in 50 to 100 MW tranches (6496, 6413, 6153).

The dating pattern is the clearest tell. Nearly every claim carrying an explicit deadline has slipped: the 80 MW Childress expansion promised for June 2024 (6495, 6410), T21 shipments (6408), GPU commissioning (6262), the 50 MW 2024 build (6153), the H200 purchase and resulting 1,896-unit fleet (5711, 5712), and the 31 EH/s Q4 2024 step (5691). Conversely, the largest figures — 50 EH/s (4034, 4021), 200 MW at Horizon (4159), 1,600 MW in Oklahoma (16667) — carry no promised-by date at all. Risk has migrated into undated claims.

Independent corroboration is thin in this record. Nothing here attaches a permit, an ERCOT or BC Hydro interconnection position, or site reporting to any milestone; the BC Hydro relationship appears only as company characterization (12182, 12198). The 99,900 GPU installed-or-on-order figure (3422) is a two-order-of-magnitude jump from 2,100 units ten months earlier (4033) and is the single least externally anchored step on record.

Over the next ninety days, watch whether Horizon energizes against the end-2026 Microsoft schedule (4158, 8112), whether the already-delayed Dell agreement (4161) is reconciled with the 99,900-unit claim, whether Oklahoma produces queue or permit filings consistent with a 2028 ramp (16668), and whether 50 EH/s is simply restated flat, as 5.5 once was.

IREN's capacity story scales up impressively in aggregate while the underlying delivery record is more uneven than the headline numbers suggest. The founding promise — 15.2 EH/s across all sites by September 2023, underwritten by binding Bitmain contracts (claims 14392, 14393) — was missed by a wide margin; the company was guiding to 7.0 and then 9.4 EH/s in October 2023 (12137, 12138), roughly a two-year slip that was never framed as one. The intervening period shows a different kind of noise: the 5.5 EH/s figure was restated across at least six separate releases between February and April 2023 (13337, 13348, 12827, 12784, 12726, 12712, 12699, 12689), producing announcement volume without capacity movement. From there the ramp is real — 9.1 EH/s (12448), 10 (6341, 6150), 20 to 21 (5689, 5690), 50 by late 2025 (4034, 4021) — but note the quiet step down to 46 EH/s installed in March 2026 (3423). That is a four-exahash gap between stated capacity and installed capacity five months apart, reconcilable through definitional drift rather than disclosed shortfall, and it deserves a direct question.

The date discipline is the clearest signal. Nearly every claim carrying a hard deadline is flagged delayed: the 80MW Childress expansion due mid-2024 (6495, 6410), T21 shipments (6408), the 568-GPU commissioning window (6262), the 50MW 2024 build (6153), the H200 purchase and 1,896-unit fleet target (5711, 5712), and the 31 EH/s Q4 2024 step (5691). Undated claims are simply stated. The pattern is a company that hits directional targets late and avoids committing to dates where it can.

Nothing in this record is independently corroborated. The Childress 600MW and Sweetwater 1,400MW figures (12176, 12186, 6523) are potential capacity, and Kiowa, Oklahoma, Bundey and Sweetwater 2 carry no capacity or cost at all. ERCOT large-load interconnection entries and Childress County permitting are the only things that would confirm pace.

Over the next 90 days, watch the Microsoft GPU deployment schedule against its end-2026 marker (4158, 8112), the already-slipped Dell agreement (4161), B300 delivery evidence (11693), and whether installed hashrate recovers toward 50 EH/s or 46 becomes the new baseline.

The defining feature of IREN's disclosure history is not that capacity targets were missed but that the unit of account kept changing before misses could be adjudicated. The 2021 listing-era commitment of 15.2 EH/s by September 2023 (14393, backed by binding Bitmain contracts in 14392) resolved into roughly 5.5 EH/s (12784, 12712, 12699), a shortfall of about two-thirds that was never characterized as such. Instead the 5.5 figure was restated in at least six separate releases across April 2023 (12827, 12726, 12690, 12698, 12689), an announcement cadence that outpaced actual progression. From mid-2023 the trajectory genuinely steepened: 9.1 EH/s (12448), 10 EH/s (6341), a stated pathway to 20 (6342), 31 EH/s targeted for Q4 2024 (5691), and 50 EH/s by late 2025 (4034, 4021). That is real order-of-magnitude growth, but it arrives alongside a migration of the headline metric from hashrate to megawatts (600 MW at Childress in 12176 and 12186, 1,400 MW in West Texas in 6523) and then to GPU counts and contracted dollars (568 H100s in 6261, 1,896 units in 5712, 2,100 in 4033, 50,000 B300s in 11693, $9.7B with Microsoft in 8112, $5.8B with Dell in 4161). Each pivot resets the scoreboard.

The slippage pattern is consistent and narrow: equipment and commissioning dates move, site energization tends to arrive. Childress 80 MW slipped from the June 2024 anchor twice (6495, 6410), T21 shipments slipped (6408), GPU commissioning slipped (6262), the 50 MW 2024 build slipped (6153), and the H200 purchase and resulting fleet size slipped past December 2024 (5711, 5712). None of these were retracted; they were superseded.

What this record does not contain is external corroboration. The operational-since assertions (12193, 12194, 12195, 12196) are retroactive and self-reported. The 600 MW and 1,400 MW Texas figures should be legible in ERCOT large-load interconnection filings, and the BC Hydro arrangements (12182, 12198) sit against a province with a documented history of pausing crypto load connections.

Over the next ninety days, the binding test is the Microsoft deployment schedule due 31 December 2026 (4158) against Horizon's 200 MW IT load (4159), with Dell already flagged late (4161). Watch whether the B300 phasing (11694) is restated, and whether 50 EH/s holds flat as power is redirected.

IREN's announced capacity has scaled by roughly an order of magnitude, but the path shows a recurring pattern of targets being overtaken rather than met on time. The November 2021 plan for 15.2 EH/s by September 2023 (14393), backed by Bitmain contracts (14392), is marked delayed. Operating capacity instead stepped from 2.3 EH/s (13824) to 3.7 (13604), then held at 5.5 EH/s through a long run of near-identical restatements in early 2023 (12784 through 12689). The targets kept moving: 9.1 EH/s (12448), then 9.4 (12138), then 10 EH/s (6522), which was reported as reached in May 2024 (6150). The company claimed 20 EH/s installed by September 2024 (5689) and 50 EH/s by late 2025 (4034, 4021). The original 15.2 figure was surpassed, but about a year behind schedule.

The dated facility milestones are weaker. The Childress 80 MW expansion due by mid-2024 (6495, 6410), the 50 MW Childress build for 2024 (6153), the 31 EH/s Q4 2024 target (5691), and the H100 and H200 GPU deliveries (6262, 5711, 5712) are all flagged delayed. None of these has a later claim in the record that acknowledges or re-dates it. Each was superseded by a larger headline instead. The same pattern shows in site power: Childress was presented at 600 MW (12176, 12186), then with an additional 500 MW (6412), and West Texas at 1,400 MW (6523). None of the project records carries populated capacity or cost fields, so these megawatt figures cannot be reconciled against a baseline.

This record contains no independent corroboration. There are no ERCOT interconnection-queue entries, no Nolan or Childress County permits, and no on-the-ground reporting. Some unit tags also look unreliable, such as 12208 recording GPU capacity in USD. The pace of the AI pivot therefore rests on company statements alone. That pivot includes the $9.7 billion Microsoft contract (8112), the $5.8 billion Dell supply agreement (4161), the 200 MW Horizon IT load (4159), and 50,000 B300 GPUs (11693).

Three things matter most over the next 90 days. First, whether Horizon's Microsoft deployment stays on track for its 2026-12-31 date (4158). Second, whether IREN explains the status of the Dell deliveries that were due by March 2026 and are now delayed (4161). Third, whether the Mackenzie B300 phases slated for H2 2026 (11694) show up as energized, revenue-generating capacity in the next quarterly report. Grid-queue and permit filings for Sweetwater would be the first external test of the 1,400 MW narrative.

IREN's record describes a company whose headline ambitions have repeatedly outrun delivery, then caught up later. The earliest hashrate target, 15.2 EH/s by September 2023 (claim 14393), backed by a 14.5 EH/s Bitmain contract (claim 14392), was missed by a wide margin. Operating capacity sat at 5.5 EH/s through spring 2023 (claims 12712, 12689). The company then reset to a sequence of smaller interim targets: 9.1 EH/s (claim 12465), 9.4 EH/s (claim 12138), and 10 EH/s (claims 6341, 6150). The 20 EH/s pathway flagged in January 2024 (claim 6342) was reached by September 2024 (claim 5689). The 31 EH/s Q4 2024 target (claim 5691) slipped, and 50 EH/s appears only by late 2025 (claims 4034, 4021). The trajectory is real growth, but the stated dates mostly marked the start of a slide rather than a finish line.

The physical build shows the same pattern. The Childress 80 MW expansion promised by mid-2024 (claims 6495, 6410) and the 50 MW tranche promised by end-2024 (claim 6153) are both marked delayed. Their follow-on disclosures do not acknowledge the slip. Site-level power framing has also shifted. Childress was described as 600 MW total potential (claims 12186, 12176), then as having an additional 500 MW (claim 6412). Sweetwater 1 carries a 1,400 MW figure (claim 6523) that has no dated delivery milestone on record.

The AI pivot follows the same rhythm at larger scale. Early GPU commitments of 568 H100s (claim 6262) and a 1,896-unit fleet (claim 5712) both slipped. The fleet was only about 2,100 units by November 2025 (claim 4033). By March 2026 the company claims 99,900 GPUs installed or on order (claim 3422), without separating the two. The Dell supply agreement tied to the Microsoft contract (claims 4161, 8112) already missed its March 2026 date.

This record contains no independent corroboration: no ERCOT large-load or interconnection filings, no county permits, no site reporting. Announced pace at Childress, Horizon and Sweetwater should be treated as company-asserted. Oklahoma, Kiowa and Bundey carry no claims at all.

Over the next 90 days, watch four things. First, whether Horizon's 200 MW IT load (claim 4159) is energized against the end-2026 Microsoft deployment date (claim 4158). Second, whether IREN discloses the installed-versus-ordered split behind claim 3422. Third, whether Dell deliveries have caught up. Fourth, whether any interconnection or energization evidence surfaces for Sweetwater 1.

IREN's announced scale has grown by more than an order of magnitude, and the unit of ambition has changed. Through 2023 and 2024 the headline figure was bitcoin hashrate. The 15.2 EH/s target for September 2023 (claim 14393) was followed by operating figures stuck at 5.5 EH/s through April 2023 (claims 12784, 12689). Interim targets of 9.1, 9.4 and 10 EH/s followed (claims 12465, 12138, 6341), with 10 EH/s confirmed in May 2024 (claim 6150) and 20 EH/s installed by September 2024 (claim 5689). The company eventually far exceeded the original goal, reporting 50 EH/s by late 2025 (claim 4034), but hit it roughly a year late. The later story is GPUs: 568 H100s (claim 6261), 1,896 units (claim 5712), 2,100 units in November 2025 (claim 4033), then a $9.7 billion Microsoft contract (claim 8112), 50,000 B300s (claim 11693) and 99,900 GPUs "installed or on order" by March 2026 (claim 3422). That last figure is about 48 times the count reported three months earlier, and the blended wording hides how much of it exists physically.

Dated promises have mostly slipped. The 80MW Childress expansion due by mid-2024 (claims 6495, 6410), the 50MW 2024 Childress build (claim 6153), the H100 commissioning (claim 6262), the 31 EH/s Q4 2024 target (claim 5691) and the Dell delivery due March 2026 (claim 4161) all carry delayed flags. None has a follow-up claim acknowledging a revised date. Instead, subsequent disclosures move to larger numbers, and the original target drops out of the narrative. Site-level power figures show the same inflation. Childress went from 600MW potential (claim 12186) to Sweetwater's 1,400MW (claim 6523), and the project register carries no capacity or cost for any of its eleven sites.

The record contains no independent corroboration: no permits, no ERCOT or BC Hydro interconnection queue entries, and no on-the-ground reporting. The BC Hydro relationship (claims 12198, 12182) is self-described. The announced pace therefore rests entirely on company statements. That matters most at Sweetwater and at the unsized Oklahoma, Kiowa and Bundey sites.

Over the next 90 days, the test is Horizon. The Microsoft deployment is due by end-2026 (claim 4158) against a 200MW IT load (claim 4159). Watch for a disclosed split of installed versus ordered GPUs behind claim 3422, confirmation that the delayed Dell hardware has landed, and any ERCOT queue or energization evidence for Sweetwater before capacity figures there are treated as more than intent.

IREN's capacity story splits into two eras, and the numbers only make sense when read that way. The mining era began with a 15.2 EH/s target for September 2023 (14393), backed by 14.5 EH/s of Bitmain contracts (14392). Both are flagged delayed. Through April 2023 the company repeatedly reported 5.5 EH/s (12712, 12689). Later milestones were stated as fresh targets: 9.1 EH/s (12465), then 7.0 and 9.4 EH/s (12144, 12138), then 10 EH/s (6341). None of these was framed as a reset of the original goal. The fleet did grow, reaching 10 EH/s by May 2024 (6150), 20 EH/s by September 2024 (5689) and 50 EH/s by late 2025 (4034, 4021). It then fell to 46 EH/s installed in March 2026 (3423), the first contraction on record. That is consistent with power being reallocated to GPUs.

The GPU side scaled far more abruptly. The count went from 568 H100s (6261) to about 1,896 after the H200 purchase (5712), to 2,100 (4033), then to 99,900 'installed or on order' (3422). That last figure mixes deployed and contracted hardware, so it cannot be compared cleanly with the earlier ones. Power announcements show the same escalation: 600 MW at Childress (12176), 1,400 MW at Sweetwater (6523) and 200 MW of IT load at Horizon (4159). These are anchored by the $9.7 billion Microsoft contract (8112) and the $5.8 billion Dell agreement (4161).

Date discipline has been weak. The 80 MW Childress expansion due mid-2024 (6495, 6410), the 50 MW 2024 build (6153), the H100 commissioning (6262), the Q4 2024 31 EH/s target (5691) and the Dell milestone of March 2026 (4161) are all flagged delayed. We found no explicit acknowledgement of any of these slips in the company's own claims.

The record contains no independent corroboration. It has no ERCOT interconnection-queue entries for Sweetwater or Childress, no building or air permits, and nothing from BC Hydro beyond IREN's own partnership statements (12198, 12182). The announced pace therefore rests entirely on company disclosure. Note also that the claim list provided ends mid-entry at 11698, so the latest disclosures are not assessed here.

Over the next 90 days, check whether Horizon phases are energized on the schedule attached to the Microsoft deployment (4158). Look for Dell delivery confirmation against the delayed March milestone, and for the 99,900-unit figure to be split into installed and ordered. Watch for Sweetwater energization evidence in ERCOT filings, and for whether hashrate keeps falling as mining sites are converted.

IREN's record reads as two companies sharing one set of sites. The Bitcoin-mining line scaled in steps that repeatedly trailed the headline. The 15.2 EH/s target for September 2023 (14393), backed by the Bitmain contracts in 14392, is marked delayed. Operating capacity sat at 5.5 EH/s through spring 2023 (12712, 12689) and was restated toward 9.1 and then 9.4 EH/s (12465, 12138). The 10 EH/s milestone arrived around May 2024 (6150), 20 EH/s by September 2024 (5689), and the Q4 2024 target of 31 EH/s slipped (5691). By late 2025 IREN reported 50 EH/s (4034, 4021), but the March 2026 figure of 46 EH/s installed (3423) is the first contraction in the series. The filings do not explain it, but it coincides with power being redirected to GPUs.

The GPU line is the steeper curve. It rose from 568 H100s (6261) to 1,896 units after the H200 purchase (5712), then to roughly 2,100 (4033), and then to 99,900 installed or on order (3422). That last figure mixes delivered and contracted hardware, so it measures commitment rather than delivery. The Microsoft contract (8112) and the Dell supply agreement (4161) sit behind it, and the Dell milestone for March 2026 is already flagged delayed.

On timelines, the pattern is consistent. Near-dated physical deliverables at Childress have slipped: the 80 MW expansion due mid-2024 (6495, 6410), the 50 MW build for 2024 (6153), and the GPU commissioning dates (6262, 5711). The large envelopes have stayed intact and been enlarged: 600 MW potential at Childress (12176), 1.4 GW at Sweetwater (6523), and 200 MW of Horizon IT load (4159). Several listed projects, including Kiowa, the Oklahoma campus and Bundey, carry no capacity or timing claims at all. The record provided also ends partway through the 2026 entries.

This claim set contains no independent corroboration. It has no ERCOT or BC Hydro interconnection-queue entries, no building permits and no site reporting. The pace therefore rests entirely on company statements, and those statements have a documented slip rate on near-term dates.

Over the next 90 days, the key test is claim 4158, which schedules the Microsoft deployment at Horizon for completion by end-2026. The November quarterly update should say how many Horizon phases are energized, whether the Dell hardware has shipped in full, and whether the Mackenzie GPU phases (11694) are on track. Energization evidence for Sweetwater should also appear by then. Watch too whether mining hashrate falls further below 46 EH/s without comment.

IREN's record shows two businesses on different arcs. The bitcoin mining side shows long-run scale-up with repeated near-term slippage. The GPU and AI side began with small purchases and has recently jumped to contract values that dwarf everything before it.

On mining, the original 2021 target of 15.2 EH/s by September 2023 (14393, backed by the Bitmain contracts in 14392) is marked delayed. Operating capacity was 2.3 to 3.7 EH/s through late 2022 (13824, 13604). It then sat at 5.5 EH/s across roughly ten near-identical statements from February to April 2023 (13337 through 12698), and that repetition overstates activity. Interim targets kept being reset: 9.1 EH/s (12465), 9.4 (12138), 10 (6522). The company reported 10 EH/s in May 2024 (6150), still short of the 2021 figure eight months after its deadline. The 31 EH/s Q4 2024 goal (5691) is also marked delayed. By late 2025 IREN reported 50 EH/s (4034, 4021), well past the original ambition, but every dated milestone along the way missed. There is no explicit acknowledgement of those misses; new targets simply replaced old ones.

At Childress, stated site power went from 600 MW (12186, 12197) to 500 MW of additional capacity (6412), and it is unclear whether these figures overlap. The 80 MW expansion promised for mid-2024 (6495, 6410) and the 50 MW 2024 build (6153) are both delayed. The GPU track follows the same pattern. The 568 H100s due mid-2024 (6262) and the 1,896-unit fleet due end-2024 (5712) both slipped, and the stated count was only 2,100 units in November 2025 (4033). Weeks earlier the company announced the $9.7 billion Microsoft contract (8112), 200 MW at Horizon (4159), and a $5.8 billion Dell agreement (4161), whose March 2026 date is already marked delayed. The claimed 50,000-GPU purchase (11693) implies more than a twentyfold jump in fleet size.

The record contains no independent evidence: no permits, no ERCOT large-load or interconnection queue entries, no site reporting. Every project, including sites claimed operational since 2019 to 2023 (12193, 12194, 12196), still carries the status "announced" with no capacity or cost fields. The 1,400 MW Sweetwater figure (6523) is the least corroborated claim on file. The claim list also appears truncated at 11694, so later revisions may be missing.

Over the next 90 days, watch for delivered Dell hardware, since 4161 is past due. Watch for evidence that Horizon is on track for the December 31, 2026 Microsoft schedule (4158), and for confirmation of the H2 2026 GPU phases (11694). Also watch for any ERCOT or permit filings that would substantiate Sweetwater's energization timeline.

IREN's capacity story has grown by orders of magnitude, but the dates attached to it have rarely held. The 2021 target of 15.2 EH/s by September 2023 (14393), backed by Bitmain contracts (14392), was missed. Operating capacity sat at 5.5 EH/s through spring 2023, restated in roughly a dozen near-identical disclosures (12784, 12712, 12689), and reached 10 EH/s only in May 2024 (6150). The target was eventually exceeded, with 20 EH/s in September 2024 (5689) and 50 EH/s by late 2025 (4034, 4021), but about a year late. The Q4 2024 step to 31 EH/s (5691) is itself flagged delayed.

The Childress build shows quieter revision. The 80 MW expansion promised for mid-2024 was reaffirmed in December 2023 with the same date (6495, 6410) and marked delayed. A 100 MW early-works framing (6496, 6413) then gave way to a 50 MW construction goal for 2024 (6153), which also slipped. None of this was reconciled on the record. Meanwhile, headline site potential kept growing: 600 MW at Childress (12186), 500 MW more (6412), and 1,400 MW in West Texas (6523). The record contains no Sweetwater timeline to test that figure against. Several listed projects, including Kiowa, the Oklahoma campus, Bundey, Sweetwater 2 and the Sweetwater Hub, have no capacity or date claims at all.

GPU commitments follow the same pattern at a larger scale. The 568 H100 delivery for mid-2024 (6262) and the 1,896-unit fleet for end-2024 (5712) both slipped, and only 2,100 units were reported by November 2025 (4033). The company then signed a $9.7 billion Microsoft contract (8112), a $5.8 billion Dell supply agreement already past its March 2026 date (4161), and a 50,000-unit GPU purchase phased through H2 2026 (11693, 11694). That is more than a twentyfold jump from a base with a consistent slippage record.

The record includes no independent corroboration, such as ERCOT large-load interconnection filings, county permits or site reporting. Every capacity figure here is self-reported, and the Sweetwater megawatts in particular should be checked against public queue data. The claim list also appears truncated after 11694.

For the next 90 days, three items matter. The first is whether Horizon's 200 MW of IT load (4159) is visibly energizing against the December 2026 Microsoft deployment schedule (4158). The second is whether IREN confirms Dell deliveries or quietly moves the date. The third is whether the next quarterly update restates the H2 2026 GPU phasing or revises it without comment.

IREN's capacity story splits into two eras, and the numbers move differently in each. The bitcoin-mining era opened with an ambitious target of 15.2 EH/s by September 2023 (14393), backed by Bitmain contracts for 14.5 EH/s (14392). By April 2023 the company was reporting 5.5 EH/s installed (12689, 12698). The goal was then reset in steps: 9.1 EH/s tied to an 80 MW build (12465), 9.4 EH/s (12138), and 10 EH/s (6341), which it reached in May 2024 (6150). The original September 2023 deadline was not met, and nothing on record acknowledges the miss. Later growth was real and fast: 20 EH/s installed by September 2024 (5689) and 50 EH/s by late 2025 (4034, 4021). The March 2026 figure of 46 EH/s installed (3423) is the first decline on record. That fits capital and power moving toward GPUs.

The AI era scales much faster. The GPU plan grew from 568 H100s (6261) to 1,896 units (5712), then 2,100 (4033), and then a 50,000-unit B300 agreement (11693). On power, Childress moved from 600 MW potential (12176) to 750 MW combined with the 500 MW addition (6412), and Sweetwater 1 was framed at 1,400 MW (6523). Horizon's 200 MW IT load (4159) is now tied to a $9.7 billion Microsoft contract (8112).

The record on dates is weak. The 80 MW Childress expansion (6495, 6410), the H100 commissioning (6262), the 50 MW 2024 build (6153), the H200 fleet (5712), the 31 EH/s Q4 2024 target (5691), and the Dell supply milestone (4161) are all marked delayed. One inconsistency also stands out: Childress is described as operational since April 2023 (12196) while construction was still underway in June 2023 (12444). This is probably partial energization, but it is not stated.

The record contains no independent evidence. There are no ERCOT large-load interconnection filings, no county or state permits, no BC Hydro service confirmations beyond the company's own partnership claims (12198), and no on-the-ground reporting. The announced pace therefore rests on IREN's word. Given the delay rate, that is a thin basis. The supplied record also appears to cut off after claim 3423, so any later disclosures are not assessed here.

In the next 90 days, watch whether the Horizon deployment stays on track for the December 2026 Microsoft schedule (4158). Watch whether the Dell delivery slippage (4161) is resolved or rolls into 2027, and whether B300 phases at Mackenzie through H2 2026 (11694) are confirmed as delivered or quietly re-phased. Two further signals would matter: a further drop in hashrate below 46 EH/s, and ERCOT queue entries for Sweetwater, which would give the first external check on the 1.4 GW claim.

IREN's record shows a company whose unit of ambition has changed more than once, and this makes the trajectory hard to score on a single axis. The earliest target, 15.2 EH/s across all sites by September 2023 (14393), was not met. Stated operating capacity was 5.5 EH/s in April 2023 (12689). Targets were then restated lower and closer in, at 9.1 EH/s (12443) and 9.4 EH/s (12138). The 10 EH/s mark was not claimed until May 2024 (6150). None of these restatements refers back to the original commitment. Hashrate later climbed to 20 EH/s (5689), then to 50 EH/s by late 2025 (4034, 4021). It fell to 46 EH/s installed in March 2026 (3423). That decline is most plausibly read as power being reallocated to GPUs, but the record does not say so explicitly.

The GPU line scales far more steeply. It went from 568 H100s (6261) to 1,896 units after the H200 purchase (5712), then 2,100 in November 2025 (4033). Four months later the company claimed 99,900 units 'installed or on order' (3422). That figure blends deployed and contracted hardware, so it should not be read as operating capacity.

Dated facility promises have fared poorly. The 80 MW Childress expansion due mid-2024 (6495, 6410) is flagged delayed. So are the 50 MW of 2024 Childress construction (6153), the H100 commissioning (6262), the H200 delivery (5711), the Q4 2024 31 EH/s target (5691), and the Dell supply agreement milestone of March 2026 (4161). Power-envelope figures such as 600 MW at Childress (12176) and 1,400 MW in West Texas (6523) are site potential, not energized load. No project record carries populated capacity or cost fields, and Kiowa has no location.

The file contains no independent corroboration. There are no permits, no ERCOT large-load interconnection entries, no BC Hydro filings, and no site reporting. The announced pace therefore rests entirely on company statements. The claim list is also truncated after 11698, so later 2026 claims were not assessed.

Over the next 90 days, the central test is the Horizon deployment for Microsoft, scheduled by year-end 2026 (4158) against a 200 MW IT load (4159). That deployment underpins the $9.7 billion contract (8112). Also watch whether the Mackenzie H2 2026 GPU phases (11694) are reported as energized, and whether the 99,900-unit figure is split into installed and pending units. The next quarterly update should show whether Dell deliveries have caught up, and ERCOT queue filings for Sweetwater would be the first external check on the Texas buildout.

IREN's headline numbers have grown by more than an order of magnitude, but the metric being headlined has changed along the way. Through 2024 the story was hashrate, climbing from 2.3 EH/s (13824) past 10 EH/s (6150) to 20 EH/s installed (5689), with a 31 EH/s target (5691) and later a 50 EH/s capacity figure (4034, 4021). The March 2026 disclosure gives 46 EH/s installed (3423). That is either a real step down or a quiet change from capacity to installed, and the record does not say which. GPUs have since replaced hashrate as the lead number: 568 H100s (6261), then 1,896 units (5712), 2,100 (4033), then 99,900 installed or on order (3422), anchored by a 50,000-unit B300 agreement (11693). Merging installed and ordered units into one figure makes the jump hard to judge. Power claims have scaled the same way, from an 80 MW build (12449) to 600 MW at Childress (12186), a further 500 MW there (6412), and 1,400 MW in West Texas (6523).

Dated promises have not held well. The 80 MW Childress expansion due by mid-2024 is flagged delayed (6495, 6410), as are the T21 shipments (6408), H100 commissioning (6262), the 50 MW Childress build for 2024 (6153), the H200 purchase (5711) and the 31 EH/s Q4 2024 target (5691). The 50 MW figure appeared after the 80 MW commitment without an explanation of how the two relate, which reads as a re-scope rather than an acknowledged slip. The 100 MW Childress early works (6496, 6413) have never carried a date. More recently, the Dell supply agreement tied to March 2026 (4161) is also marked delayed, and it underpins the Microsoft contract (8112).

The record contains no independent corroboration: no permits, no ERCOT or BC Hydro interconnection queue entries, and no site reporting. The BC Hydro relationship (12182, 12198) is company-stated. Sweetwater, Kiowa, the Oklahoma campus and Sweetwater Hub are listed as projects with no capacity, cost or dated claims attached, so the pace there cannot be tested at all.

Over the next 90 days, the key test is whether Horizon's 200 MW (4159) is delivering Microsoft GPU services by the end-2026 deadline (4158), since that deadline falls inside the window. Also watch whether Dell hardware delivery is confirmed, whether the Mackenzie H2 2026 phases (11694) are reported as energized, how installed GPUs are split from ordered ones, and whether the next hashrate figure continues below 50 EH/s. For Sweetwater, look for ERCOT queue entries or county filings that would support the 1,400 MW claim.

IREN's record describes a company that has changed what it measures as much as how much it has grown. Bitcoin hashrate rose from 3.7 EH/s (13604) to a 5.5 EH/s milestone repeated across roughly a dozen near-identical filings (12827, 12712, 12698). It then climbed to 10 EH/s achieved (6150), 20 EH/s installed (5689) and 50 EH/s of stated capacity (4034, 4021). The March 2026 figure of 46 EH/s installed (3423) sits below that 50 EH/s claim with no reconciliation on record, which reads as either a definitional shift between capacity and installed or a quiet contraction as power is redirected to GPUs. The GPU line has scaled far more steeply: 568 H100s (6261), a 1,896-unit fleet (5712), 2,100 units (4033), then 99,900 installed or on order (3422), anchored by a 50,000-unit B300 order (11693). Power ambitions moved the same way, from a 600 MW Childress site (12176) and 1,400 MW in West Texas (6523) to a 5 GW NVIDIA partnership (11077). None of these headline numbers has been carried into the project records, all of which still show blank capacity and cost.

Dated promises have a weak record. The 80 MW Childress expansion due mid-2024 (6495, 6410), the 50 MW Childress build for 2024 (6153), T21 shipments (6408), the H100 commissioning (6262) and the H200 purchase (5711) are all tagged delayed. The later 31 EH/s Q4 2024 target (5691) and the Dell supply agreement due March 2026 (4161) are also marked delayed. Nothing in the record shows IREN acknowledging these slips. Each new disclosure simply resets to a larger number. The record also has no claims between September 2024 and November 2025, exactly the window in which the AI pivot took shape.

No independent corroboration is on file. Every claim is company-stated, and there are no permits, ERCOT interconnection queue entries, BC Hydro filings or site reporting for Sweetwater, Horizon or Childress. The pace cannot be verified from this dataset. The input was also truncated after claim 11077, so later claims were not reviewed.

Over the next 90 days, the key test is Horizon's 200 MW (4159) against the Microsoft deployment deadline of 2026-12-31 (4158), which underpins the Microsoft contract (8112). The other things to watch are whether Dell hardware delivery (4161) closes, and whether the B300 phases slated for H2 2026 (11694) show up as energized load. Beyond those, look for any ERCOT or municipal filing that ties the Sweetwater megawatts to a real interconnection date, and for an explanation of the gap between 50 and 46 EH/s.

IREN's record describes a company that has changed what it is building, and the capacity figures track that change more clearly than they track delivery. Early 2023 claims repeat a single 5.5 EH/s milestone (12784, 12712, 12689). Targets then climb in steady steps: 9.1 EH/s tied to an 80 MW Childress build (12448, 12444), 10 EH/s (6522, 6341), a pathway to 20 EH/s (6342), 20 EH/s installed and 31 EH/s targeted for Q4 2024 (5689, 5691), and 50 EH/s by late 2025 (4034, 4021). The March 2026 figure of 46 EH/s installed (3423) is the first decline, and nothing on record explains it. A likely reading is that power is being reallocated to GPUs, but the company has not said so. GPU claims show the real inflection. They move from 568 H100s (6261) to 1,896 units (5712) to 2,100 (4033), then to 99,900 installed or on order (3422), anchored by a 50,000-unit purchase (11693) and the Microsoft and NVIDIA contracts (8112, 11076).

Dated promises have held poorly. The 80 MW Childress expansion due mid-2024 (6495, 6410), the T21 shipments (6408), the first GPU commissioning (6262), 50 MW at Childress in 2024 (6153), the H200 fleet (5711, 5712), 31 EH/s by year-end 2024 (5691), and the Dell supply due March 2026 (4161) are all flagged as delayed. Several large commitments were never dated at all and never revisited. These include the 100 MW Childress early works (6496, 6413), the 500 MW of additional Childress power (6412), and the 1,400 MW West Texas development tied to Sweetwater 1 (6523). The record also goes quiet between September 2024 and November 2025, which hides how the slipped 2024 targets were actually resolved.

The record contains no independent evidence for any of this. There are no permits, no ERCOT or BC Hydro interconnection entries beyond the stated partnership (12198), and no site reporting. All eleven projects carry blank capacity and cost fields, so the announced pace rests entirely on company statements. Sweetwater, the largest claimed power position, has no timeline claim of any kind.

Over the next 90 days, the key test is Horizon. Its 200 MW IT load (4159) must support Microsoft deployments promised by end-2026 (4158), and the delayed Dell hardware feeds that schedule directly. Watch the next operating update for three things: an explanation of the 50 to 46 EH/s drop, any energization or queue evidence for Sweetwater, and whether the phased GPU deployment through H2 2026 (11694) names specific sites and megawatts rather than unit counts.

IREN's facility story has scaled by more than an order of magnitude, and the unit of measure has changed along the way. In mid-2023 the anchor was an 80 MW build at Childress (12444, 12449), framed in hashrate terms of 5.5 to 9.1 EH/s (12827, 12443). By September 2023 Childress was described as a 600 MW site (12176), and two months later a 1,400 MW West Texas development tied to Sweetwater 1 appeared (6523). By 2026 the framing had moved to GPUs and megawatts of AI capacity: 99,900 GPUs installed or on order (3422), a 5 GW NVIDIA partnership (11077), and stepped targets of 480 MW by end-2026 (9290) and 1,210 MW by end-2027 (9293). Hashrate has quietly contracted. A 50 EH/s capacity figure (4034, 4021) became 46 EH/s installed by March 2026 (3423). That is plausibly a deliberate reallocation toward GPUs, but the record contains no claim explaining it.

The date record is weak. Nearly every claim that carried a specific deadline is marked delayed. That includes the Childress 80 MW expansion due mid-2024 (6495, 6410), the 50 MW 2024 build (6153), T21 miner shipments (6408), the first H100 commissioning (6262), the H200 purchase and resulting fleet size (5711, 5712), the 31 EH/s Q4 2024 target (5691), and the Dell GPU supply milestone for March 2026 (4161). Later disclosures generally set new, larger targets rather than reconciling the missed ones. The 31 EH/s miss, for example, is never addressed before 50 EH/s is cited. Most capacity claims also carry no delivery date at all, which limits how far they can be held to account.

No independent corroboration appears in the record. There are no permits, no ERCOT large-load interconnection filings, no BC Hydro service agreements beyond company-stated partnerships (12182, 12198), and no site reporting. Every project entry still lacks capacity and cost fields. The announced pace therefore rests entirely on company statements, and a delivery history this delay-prone argues for pulling ERCOT queue and Nolan County permit records for Sweetwater before accepting the 2027 figure.

The next 90 days bring the first real test of the AI-era targets. Watch the fiscal first-quarter update for progress against the Horizon 200 MW build (4159) and the Microsoft deployment schedule due by end-2026 (4158). Watch for whether the delayed Dell supply (4161) has closed. Check whether the 480 MW 2026 figure (9290) is restated, trimmed, or quietly rolled into 2027. Look for any first disclosure of energized load at Sweetwater.

IREN's disclosed ambition has grown by more than an order of magnitude in three years, while its record on dated near-term deliverables is consistently weak. In April 2023 the company described 5.5 EH/s of installed mining capacity (12712, 12689). It then cycled through 9.1, 9.4 and 10 EH/s targets (12448, 12138, 6522) before reporting 10 EH/s operating in May 2024 (6150) and 20 EH/s installed that September (5689). It cited 50 EH/s by late 2025 (4034, 4021). The March 2026 figure of 46 EH/s (3423) is the first reported contraction, and it coincides with the pivot to AI compute. The GPU line tells the more dramatic story. It began with 568 H100s (6261), then a planned fleet of 1,896 (5712) and about 2,100 units in late 2025 (4033). By March 2026 the company cited 99,900 installed or on order (3422), including a 50,000-unit B300 agreement (11693). Power framing followed the same curve: Childress at 600 MW (12186), 1,400 MW for West Texas (6523), a 5 GW NVIDIA partnership (11077, 9289), and 490 MW in Spain via Nostrum (9294).

Dated promises have not held well. The 80 MW Childress expansion due mid-2024 (6495, 6410), the 50 MW 2024 build (6153), the H100 commissioning (6262), the H200 purchase and fleet target (5711, 5712), the Q4 2024 31 EH/s goal (5691) and the Dell supply schedule (4161) are all flagged as delayed. None of these has a visible restatement in the record. Later disclosures instead moved to new, larger milestones, 480 MW by end-2026 (9290) and 1,210 MW by end-2027 (9293), with no reconciliation against the missed ones. Nothing indicates the misses were acknowledged, although the record here is truncated.

The record contains no independent corroboration of the announced pace. There are no permits, no ERCOT large-load interconnection entries, and no on-the-ground reporting. Every capacity figure is company-sourced. Kiowa has no stated location (9296), and the multi-gigawatt partnership claims come without site-level megawatt allocation.

Over the next 90 days, which run into mid-December, the key test is whether Horizon's 200 MW of IT load (4159) begins delivering Microsoft GPU services on the end-2026 schedule (4158). Also watch whether the H2 2026 GPU phases (11694) show up as revenue-generating installed units rather than orders. Dell deliveries, already past their date (4161), are the gating input for both. Further signals would be ERCOT or county filings for Sweetwater, any clarification of Kiowa's site, and closing terms for Nostrum. Finally, watch whether the 480 MW year-end figure is reaffirmed, quietly rebased, or absent from the next quarterly update.

IREN's disclosed ambitions have grown by roughly two orders of magnitude, but the growth has come mostly as new headline figures rather than delivered ones. In 2023 the story was an 80MW Childress build and a 600MW site ceiling (12449, 12197). By November 2023 it had added a 1,400MW West Texas project (6523). By May 2026 it claimed a 5GW secured power portfolio and a 5GW NVIDIA partnership (9182, 11077), plus 490MW via the Nostrum acquisition in Spain (9294). The stepped roadmap of 480MW in 2026 and 1,210MW in 2027 (9290, 9293) is the first framing specific enough to test. Mining capacity went the other way. Targets climbed from 9.1 to 50 EH/s (12465, 4034), then installed hashrate fell to 46 EH/s by March 2026 (3423). That drop is consistent with power being moved to GPUs, but the record contains no claim that says so. GPU counts moved from 2,100 units (4033) to 99,900 installed or on order (3422) in about three months. Because that figure mixes installed and ordered units, it does not measure deployment.

Dated promises have held poorly. The 80MW Childress expansion due by mid-2024 (6495, 6410), the 50MW 2024 build (6153), the H100 commissioning (6262), the H200 fleet target (5711, 5712), 31 EH/s by Q4 2024 (5691) and the Dell supply milestone of March 2026 (4161) are all tagged delayed. None of them has a later claim giving a revised date. Those misses were overtaken by larger announcements rather than addressed, which is the pattern to discount against. The record also has a gap between September 2024 and November 2025.

The record contains no independent corroboration: no ERCOT interconnection queue entries for Sweetwater, no BC Hydro service agreements beyond the partnership references (12198, 12182), no county permits or abatement filings, and no site reporting. The 5GW portfolio and the 2027 and 2028 steps (9296) therefore rest on company statements alone. The Sweetwater 1 timeline claim (9190) is truncated in this record and should be re-pulled before any conclusion about that site.

Over the next 90 days, the test is whether Horizons 2 through 4 are energized and serving Microsoft on the 2026 year-end schedule (9187, 4158). That is the first hard check on the 480MW target. Also watch whether the fiscal Q1 update separates installed GPUs from ordered ones, whether the delayed Dell deliveries get a stated new date, and whether regulatory filings show Sweetwater 1 energization and the grid status of the Nostrum capacity.

IREN's stated ambition has grown by roughly an order of magnitude, while the record of dated deliverables is thinner and more mixed than the headline figures suggest. In 2023 the story was a 600 MW Childress site (12176, 12197, 12186) and a 1,400 MW West Texas development later tied to Sweetwater 1 (6523). By 2026 it had become a 480 MW target for 2026 (9290), 1,210 MW for 2027 (9293), and a 5 GW secured power portfolio and NVIDIA partnership (9182, 9289, 11077). The GPU count shows the same pattern, rising from 568 H100s (6261) to 1,896 (5712), then 2,100 (4033), then 99,900 installed or on order (3422). Bitcoin mining capacity peaked at 50 EH/s (4034, 4021) and was later reported at 46 EH/s installed (3423). That decline is consistent with sites such as Canal Flats and Prince George being repurposed for AI workloads (9194, 9191), though the company has not presented it as a trade-off. One figure has not been reconciled: Horizon was described as 200 MW of combined IT load (4159), and six months later Horizon 1 alone was tied to a 300 MW Microsoft handoff (9186). The gap may reflect gross versus IT load, but no explanation is on record.

Dated promises have held poorly. The 80 MW Childress expansion due mid-2024 (6495, 6410), the 2024 50 MW build (6153), the first GPU commissioning (6262), the H200 fleet target (5712), and the 31 EH/s Q4 2024 goal (5691) are all marked delayed. None of these slips appears in a later claim with a revised date. The pattern has continued into 2026. The Dell supply milestone (4161), the Childress retrofit GPU delivery (9193), and the Mackenzie installation (9192) are all past their dates and flagged delayed, and later statements do not acknowledge the slips.

The record contains no independent corroboration. There are no permits, no ERCOT or BC Hydro interconnection queue entries, and no site reporting to test the claimed Sweetwater substation energization (9190), which carries no date, or the Horizon build pace. Until that evidence is gathered, the multi-gigawatt figures should be read as company assertions.

The next 90 days offer a clear test. Horizon 1's 300 MW Microsoft handoff is due by September 30 (9186), and Horizons 2 through 4 are due by year-end (9187). Watch for confirmation that the Microsoft contract (8112) has begun revenue recognition, for revised dates on the Childress and Mackenzie GPU deployments, and for ERCOT filings or county records showing Sweetwater 1 energization.

IREN's record describes a company that has changed what it is building more than how much. The 2023 filings framed West Texas as a 1,400 MW development (claim 6523) serving a bitcoin fleet heading to 10 and then 20 EH/s (6341, 6342). By 2026 the same land is described as AI cloud capacity: 480 MW in 2026 (9290), 1,210 MW or roughly 1.2 GW in 2027 (9293, 8102, 365, 8592), and a 750 MW eventual ceiling at Childress alone (8580). The headline megawatts have narrowed slightly, but the value per megawatt has been repriced around Microsoft (8112, 8563) and NVIDIA (11103, 9289). Mining has already peaked. Stated hashrate capacity reached 50 EH/s (4034, 4021), and the next disclosure reports 46 EH/s installed (3423). That is consistent with sites being reallocated to GPUs, though the company has not framed it as a reduction.

Date discipline is weak on near-term infrastructure. The 80 MW Childress expansion due mid-2024 (6495, 6410), the 50 MW 2024 build (6153), the 31 EH/s year-end target (5691), and the H100 and H200 commissioning windows (6262, 5712) are all marked delayed. The pattern continues into 2026: the Childress retrofit and Mackenzie GPU installs due by July (9193, 9192) and the Dell supply schedule (4161) have also slipped. None of these slips is followed by an explicit revised date in the record. The Horizon figures are internally inconsistent. Claim 9186 cites a 300 MW Horizon 1 handoff by September 30, while 685 describes Horizon 1 as 50 MW delivered, and 4159 and 8578 put all four Horizons at 200 MW combined. The Microsoft contract value also appears in mismatched units (8112 versus 686). Both issues should be reconciled against source documents before anyone relies on them.

The record contains no independent corroboration: no permits, no ERCOT or BC Hydro interconnection entries, and no site reporting. Sweetwater 1 energization (9190) carries no date, and Kiowa has no location. The 5 GW secured-power claim (9182) and the 2028 Sweetwater and Kiowa phase (9296) therefore rest on company assertion alone.

Over the next 90 days, the test is whether Horizons 2 through 4 reach Microsoft by year-end (9187, 8676, 8564). The fiscal first-quarter update should show whether that milestone and the 480 MW 2026 figure (9290) hold, or whether they quietly move into 2027. Watch also for a firm Sweetwater 1 energization date, revised delivery dates for Mackenzie and the Childress retrofit, and any ERCOT queue or county filing that would ground the 2027 gigawatt claim.

IREN's infrastructure narrative traces a pattern of expanding ambition paired with consistent delivery shortfalls that are rarely acknowledged in subsequent filings. In November 2023, the company announced 1400 MW of West Texas capacity and a 10 exahash-per-second bitcoin mining target (6523, 6522). Within weeks, it promised an 80 MW Childress expansion by June 2024 (6495, 6410)—marked delayed—and subsequent GPU procurement commitments for mid-2024 delivery (6261, 6262, also marked delayed). By September 2024, IREN claimed 20 EH/s installed and a 31 EH/s target for Q4 2024 (5689, 5691), with 1896 GPUs promised by December (5712). Today, six months past that Q4 deadline, the company states 46 EH/s installed capacity (3423)—silent on the 31 EH/s miss—and claims 99,900 GPU units on order (3422), a figure so large it signals a wholesale recalibration rather than incremental growth.

The GPU narrative shifted dramatically in November 2025 with a $9.7 billion Microsoft contract (8112), followed by announcements of 50,000 NVIDIA B300 purchase agreements and a $3.4 billion NVIDIA commitment (11693, 11103) in spring 2026. These claims reframed IREN from bitcoin-first to AI-infrastructure operator. However, foundational GPU infrastructure from 2024 remains undelivered. A Dell GPU agreement dated March 2026 remains marked delayed (4161). Horizon deployments—four 50 MW modules promised by December 2026—were committed in August (8564, 8589), giving only 16 weeks to deliver 200 MW of complex infrastructure. This aligns with IREN's prior pattern: construction timelines slip without public revision, while capacity targets reset annually at higher levels.

Independent validation is absent from the record. No permits, grid interconnection queue entries, or third-party site reporting anchor these claims. IREN's track record of delayed timelines across bitcoin mining, GPU procurement, and Childress expansion—none explicitly revised, all subsequently discussed as background facts in new announcements—suggests capacity goals may reflect aspirations rather than committed milestones. The company has not missed a capacity target it acknowledged missing; rather, targets are eclipsed by new ones.

In the next 90 days, observe whether the Horizon 1 delivery materializes by September 30, 2026 (claim 9186 is marked stated, not delayed), and whether Horizons 2–4 launch before year-end. If missed, watch whether IREN publicly revisits timelines or introduces them quietly in the next earnings update.

IREN's capacity announcements have undergone substantial revision and persistent timeline slippage across three key infrastructure classes. Bitcoin mining hashrate targets climbed from 8.3 exahash per second in November 2023 to 10 EH/s (claim 6341), then leapt to 20 EH/s (claim 5689) and further to 50 EH/s by late 2025 (claims 4034, 4021). The installed figure as of March 2026 stands at 46 EH/s (claim 3423), suggesting either capacity forecasting gaps or ongoing delays in deployment. Power delivery timelines have similarly oscillated: a 1400 megawatt Sweetwater 1 target (claim 6523, November 2023) was followed by fractional near-term builds, then a 480 megawatt 2026 goal for Childress (claim 9290) and 1210 megawatt for 2027 (claim 9293). GPU procurement accelerated dramatically from 568 units in February 2024 (claim 6261) to 99,900 units on order by March 2026 (claim 3422), with a 50,000-unit B300 purchase agreement added (claim 11693). The sheer scale expansion raises questions about whether prior forecasts reflected genuine capacity constraints or planning uncertainty.

Delivery timeline adherence has been weak. Early-stage promises for June 2024—80 megawatts of Childress expansion (claims 6495, 6410), T21 miner shipments (claim 6408), and 568 GPU units (claim 6262)—were all marked delayed without public revision dates. December 2024 targets for 50 megawatts of construction (claim 6153), a $43.9 million GPU purchase (claim 5711), and 31 EH/s capacity (claim 5691) also slipped. A March 2026 Dell GPU supply agreement deadline (claim 4161) was flagged as delayed, as were July 2026 deployments at Mackenzie and Childress (claims 9192, 9193).

Recent announcements introduce large partnership values—$9.7 billion with Microsoft (claim 8112), $5.8 billion with Dell (claim 4161), and a $3.4 billion five-year NVIDIA contract (claim 11103)—but these require field validation. The September 2026 Horizon 1 handoff to Microsoft (claim 9186) represents the first near-term milestone under real customer scrutiny. Horizons 2 through 4, totaling 150 megawatts and promised by year-end (claim 8589), would provide immediate evidence of whether execution velocity has improved. Independent verification via utility interconnection queues, permit filings, and power procurement agreements would substantiate the 480 megawatt and 1.2 gigawatt expansion roadmap now in flight.

IREN's capacity announcements have expanded persistently, with deployment velocity failing to align. Bitcoin mining targets escalated from 10 EH/s (November 2023, claim 6522) to 20 EH/s (January 2024, claims 6341–6342) to 31 EH/s promised by Q4 2024 (claim 5691, marked delayed). The most recent disclosed capacity is 46 EH/s installed (March 2026, claim 3423), providing no clarity on whether intermediate targets were achieved or simply superseded. GPU procurement accelerated identically: 568 units announced February 2024 (claim 6261), expanding to 1,896 promised by year-end 2024 (claim 5712, delayed), then 2,100 units (November 2025, claim 4033), culminating in 99,900 units on order (March 2026, claim 3422)—a scale jump lacking disclosed power infrastructure to operationalize it.

Childress campus shows consistent slippage on incremental targets. An 80 MW expansion promised June 2024 (claims 6495, 6410) is delayed. A 50 MW deployment promised December 2024 (claim 6153) is also delayed. GPU retrofit targeting July 2026 (claim 9193) and Mackenzie GPU installation identical (claim 9192) remain delayed as of early September 2026. These represent modest increments against claimed ultimate capacities of 750 MW Childress (claim 8596) and 2,000 MW Sweetwater (claim 8597), up substantially from the original 1,400 MW announcement (November 2023, claim 6523). Failure to execute sub-100 MW quarterly increments undermines confidence in megawatt-scale roadmaps.

Horizon status is ambiguous. Claim 8625 states Horizon 1 "delivered to Microsoft," yet claim 8581 shows that deadline delayed past August 13, 2026, and no operational confirmation appears in records as of September 12. Major partnerships—the $9.7 billion Microsoft contract (claim 8112), Dell arrangements (revised from $5.8 billion to $1.6 billion per claims 4161, 1297), and NVIDIA 5 GW partnership (claim 11077)—lack visible deployment infrastructure or third-party corroboration. No permits, interconnection queue entries, or on-the-ground reporting appear in the record to independently verify announced construction velocity.

Watch Horizon 1 operational handoff confirmation, actual power utilization at existing facilities, and whether the Childress retrofit and Mackenzie GPU installations commence by October 2026.

IREN's capacity announcements have expanded dramatically while execution timelines have compressed and slipped consistently. The company began in late 2023 with modest targets: 10 exahashes of bitcoin mining capacity (claim 6522) and 1,400 megawatts at Sweetwater (claim 6523). By mid-2026, announced capacity has grown to 750 megawatts at Childress and 2,000 megawatts at Sweetwater (claims 8596, 8597), with GPU procurement ballooning from 568 units in February 2024 (claim 6261) to 99,900 units by March 2026 (claim 3422). This twentyfold GPU expansion in 14 months signals either revised business strategy or earlier underestimation.

Near-term delivery commitments have consistently moved. The 80-megawatt Childress expansion promised for June 2024 (claim 6495) remains undelivered. GPU deliveries promised for June 2024 (claim 6262) slipped. The 31-exahash mining capacity target for Q4 2024 (claim 5691) was abandoned. The Dell GPU supply agreement promised for March 2026 (claim 4161) is marked delayed. Most recently, Childress GPU retrofits and Mackenzie installations both promised for July 1, 2026 (claims 9193, 9192) are delayed. Most telling: Horizon 1's 50-megawatt delivery missed its August 13, 2026 deadline (claim 8581) and now carries a September 30, 2026 Microsoft handoff commitment (claim 9186) that arrives in 18 days.

Public commitments have masked slippage through strategic revision. Rather than revising specific claims, IREN has layered larger, longer-dated targets atop missed near-term milestones. The company secured $9.7 billion from Microsoft through 2030 and a 5-gigawatt NVIDIA partnership, both announced November 2025 through May 2026. These long-term agreements provide customer capital flow but do not evidence near-term facility completion.

Independent verification remains limited. Horizon 1's August miss is documented in the claim record itself; no permits, interconnection queue entries, or regional reporting are cited in the data provided. The September 30 Horizon handoff is the first hard test in 90 days. If it slips again, the Microsoft relationship will strain. The Childress Blackwell deployment promised for March 2027 (claim 8993) is the next proof point. Watch quarterly announcements for revised timelines disguised as new initiatives, a pattern observable since 2024.

IREN's delivery trajectory shows escalating capacity ambitions paired with a persistent gap between promise dates and execution. The bitcoin mining operation illustrates the pattern most clearly: November 2023 announced a 10 EH/s target (claim 6522), which expanded to a 20 EH/s medium-term pathway by January 2024 (claim 6342) and then to 50 EH/s by November 2025 (claim 4034). However, March 2026's stated installed capacity of 46 EH/s (claim 3423) fell 8 percent short of that 50 EH/s milestone, marking the first quantified miss on a major hashrate target.

The GPU expansion shows even more dramatic revision without public acknowledgment of the shift. Through late 2024 and early 2025, IREN disclosed specific procurement targets: 568 H100 units by mid-2024 (claim 6262, marked delayed), 1,896 units by year-end 2024 (claim 5712, marked delayed), and 2,100 units by November 2025 (claim 4033). Then in March 2026, the company stated 99,900 GPU units either installed or on order (claim 3422)—a 48-fold expansion of the previously announced figure—with no explanation for the scale shift or earlier target obsolescence.

Facility timelines have consistently slipped. Childress expansions promised for mid-2024 carry delayed markers in December 2023 claims (claims 6410, 6495). The company now targets four 50MW Microsoft Horizon deployments at Childress for full 2026 completion (claims 8564, 8675), with Horizon 1 handoff promised by September 30, 2026 (claim 9186)—a deadline now within weeks. Given three-month slips on earlier 2024 milestones, execution risk on the compressed Q4 timeline appears material.

Independent verification is limited to recent announcement cadence. The $9.7 billion Microsoft contract (claim 8112) and simultaneous Nostrum Group acquisition (completed June 2026, claim 8106) provide commercial anchors, though neither directly confirms facility deployment status. No public permits, interconnection queue entries, or third-party site photographs have been referenced in IREN's claims.

Watch for: Horizon 1 delivery confirmation by September 30, 2026 (claim 9186); Sweetwater 1 substation energization timing (claim 9190); and whether the December 2026 four-unit Horizon deployment (claims 8564, 8675) remains on track or slips into 2027.

IREN's infrastructure trajectory reveals a pattern of capacity inflation followed by strategic pivots, with persistent slippage on near-term delivery milestones. The company's stated power footprint has cycled through multiple frames: a 2023 announcement of 1400 MW West Texas development (claim 6523) gave way to Bitcoin hashrate targeting (8.3 to 31 EH/s by Q4 2024, claims 6493, 5691), then to a GPU-centric model with 2,100 units announced (claim 4033), now anchored to Microsoft and NVIDIA partnerships promising 480 MW AI capacity by year-end (claims 8565, 8579, 8623). These are not incremental adjustments but wholesale reframings of the business—each expansion in stated capacity was paired with a shift in the underlying product and customer base. No independent verification (permitting, grid interconnection queue entries, regional reporting) appears in the public record to validate progress on the promised Bitcoin hashrate, GPU unit deployments, or power distribution upgrades that would support these claims.

Timeline slippage is the clearest signal of delivery risk. The Childress 80 MW expansion promised for June 2024 (claims 6495, 6410) remains marked delayed nearly two years later. GPU delivery commitments originally due June 2024 (claim 6262) migrated to December 2024, then to multiple 2026 tranches, with retrofit installations at Childress and Mackenzie each promised for July 2026 (claims 9193, 9192) already flagged as delayed as of the May 2026 update. Horizon 1, a linchpin of the Microsoft relationship, was promised handoff by September 30, 2026 (claim 9186) for 300 MW; a late August entry (claim 8581) marked its 50 MW component delayed past August 13. The company has not publicly revised dates retrospectively—slips appear in subsequent claim entries rather than explicit restatement of original commitments.

Watch the next 90 days for Microsoft handoff completion (Horizon 1, promised imminent), Horizon 2-4 delivery sequencing (all promised by year-end, claims 8676, 8558), and Childress retrofit GPU deployment (one of three near-term proof points). If the 480 MW target holds, it would mark the first major capacity threshold met in this cycle. Absent third-party grid data or on-site reporting, the risk profile remains opaque.

IREN has cycled through three distinct capacity narratives in under three years, signaling both ambition and execution challenges. The initial 1,400 MW Sweetwater 1 anchor announced November 2023 (claim 6523) paired with Bitcoin hash-rate targets climbing from 10 EH/s to 20 EH/s to 50 EH/s (claims 6341, 6342, 4034). By March 2026, announced installed Bitcoin capacity contracted to 46 EH/s (claim 3423), an 8 percent downward revision stated without explicit acknowledgement. The November 2025 Microsoft contract win for $9.7 billion (claim 8112) then reoriented the entire portfolio toward AI cloud infrastructure, folding the Texas footprint into the Horizon hyperscale program while simultaneously expanding the facility roster to include Mackenzie, Prince George, Canal Flats, and European assets via acquisition.

Timeline performance tells a sharper story. Childress 80 MW expansion promised for June 2024 did not materialize. H200 GPU procurement due December 2024 slipped. Most recently, Mackenzie GPU installation and Childress retrofit delivery both missed 2026-07-01 deadlines (claims 9192, 9193). These misses remain unaddressed in announced revisions, creating a pattern where delays accumulate without public correction. IREN has now committed four 50 MW Horizon modules at Childress by year-end 2026 (claims 8675, 8676), with the critical Horizon 1 handoff to Microsoft arriving 2026-09-30 (claim 9186). Total 2026 capacity target stands at 480 MW (claim 9290), with 1.2 GW promised by end-2027 (claim 8555).

Three signal points define the next 90 days. Horizon 1 handoff is imminent—now three weeks away—and any slip cascades through the Microsoft relationship. Second, the company has not publicly addressed the July GPU delivery misses or clarified how many Horizon units will actually complete by December 31 against the four promised. Third, and most fundamental, IREN has introduced no revised capacity targets following the Bitcoin downgrade from 50 to 46 EH/s, raising questions about whether future announcements reflect realistic execution or aspirational planning. Sweetwater substation energization (claim 9190) should be independently verifiable via ERCOT interconnection queue status; lack of visible progress there would underscore broader pacing concerns.

IREN's infrastructure trajectory reveals systematic patterns of capacity escalation followed by unacknowledged contraction, coupled with pervasive timeline slippage that has persisted across multiple pledge cycles. The company's announced capacity targets have undergone material revision without formal retraction. West Texas facilities, initially positioned at 1400 MW in November 2023 (6523), are now projected to reach 1210 MW by end of 2027 (9293), a 14 percent reduction never publicly reconciled. Bitcoin hashrate underwent multiple upward revisions—10 EH/s announced in late 2023 (6522, 6341), escalated to 20 EH/s by September 2024 (5689), then 31 EH/s targeted for Q4 2024 (5691). By November 2025, claims reached 50 EH/s (4034, 4021), but installed capacity was reported at 46 EH/s in March 2026 (3423), a silent retreat from that threshold with no public explanation.

Timeline delivery has been chronically problematic. Claims from late 2023 promised June 2024 delivery across Childress (6410, 6495), bitcoin hardware (6408), and GPU installation (6262), all now marked delayed. This slippage extended through 2024, with 50 MW Childress construction promised by December (6153, delayed) and GPU procurement deadlines similarly missed (5711, 5712). Two commitments from May 2026 are already overdue: Mackenzie GPU installation and Childress retrofit delivery both promised for July 1, 2026 (9192, 9193, both delayed), now 41 days past target without announced revision. These cascading delays suggest structural execution challenges rather than isolated schedule misses.

A strategic pivot from mining to AI infrastructure occurred during 2025, marked by GPU procurement scaled to 99,900 units (3422 as of March 2026) and partnerships with NVIDIA and Microsoft. However, Horizon 1 capacity descriptions conflict: described with 300 MW for Microsoft handoff by September 30, 2026 (9186), yet recent characterizations place it at 50 MW (8539, 8588), suggesting either scope reduction or conflicting documentation of the same asset.

Independent evidence—ERCOT interconnection queue positions for Childress, BC Hydro queue entries for Mackenzie, local permitting timelines, and actual substation energization events—should be cross-referenced against these claims. Over 90 days, the Horizon 1 Microsoft handoff deadline of September 30 is the most time-critical test case. Resolution of the Mackenzie and Childress retrofit delays, now materially overdue, will indicate whether recent 2026 commitments face the structural challenges that affected 2024 predecessors.

IREN's announced capacity trajectory shows a significant pivot in both scale and focus, with cascading timeline misses that remain largely unacknowledged across consecutive announcements. Starting from a 1,400 MW West Texas ambition in November 2023 (6523), the company shifted emphasis toward bitcoin mining capacity, targeting 10 EH/s by mid-2024 (6341) before escalating to 20 EH/s and then 50 EH/s (6342, 4034). By March 2026, installed mining hashrate had contracted to 46 EH/s (3423), falling short of the prior year's stated target. The company has now reframed capacity in AI compute terms, projecting 480 MW by end-2026 (8633, 8638) and 1.2 GW by 2027 (8634), though this mirrors earlier facility-level claims tied to specific sites like Childress and Sweetwater that lack granular updates.

Promised delivery dates have routinely slipped without explicit acknowledgment. The Childress 80 MW expansion was promised by June 2024 (6495, 6410), then absorbed into larger campus narratives by 2026. GPU deployment timelines—initially June 2024 (6262), then December 2024 (5711), then March 2026 (4161)—have all passed. The Mackenzie GPU installation and Childress retrofit GPU work were due July 1, 2026 (9192, 9193); we are now nine weeks past those dates with no public revision or completion claims. Meanwhile, Horizon 1 delivery to Microsoft remains scheduled for September 30, 2026 (9186)—just three weeks away—suggesting either imminent handoff or further slippage.

Independent verification is sparse. No interconnection queue entries, building permits, or grid operator filings have been cited. The claims conflate power capacity, GPU unit counts, and hashrate across different business lines without clear demarcation of which sites support which workloads. Acquisitions of Nostrum Group (8106, 9294) and Mirantis (8767, 9181) were completed mid-2026, adding execution complexity during a critical delivery window.

Over the next 90 days, watch for: confirmation of Horizon 1 Microsoft handoff by September 30 (imminent), any revised timeline for Horizons 2-4 (promised by end-2026 in claims 9187, 8676, 8675), and clarification of the 480 MW 2026 capacity claim and whether it represents actual grid-connected load or procurement targets. Third-party grid studies or utility filings would help validate claimed power sourcing claims.

IREN's delivery narrative reveals a fundamental strategic pivot masked by project proliferation. The company announced 1,400 MW at Sweetwater and 500 MW additional capacity at Childress in late 2023, framed around bitcoin mining expansion targeting 10 to 31 exahash-per-second—claims that faced serial delays from mid-2024 onward. The 80 MW Childress expansion promised by June 2024 never materialized on schedule (claims 6495, 6410), nor did the September 2024 GPU deliveries (claims 5711, 5712) or the Q4 2024 mining capacity ramp to 31 EH/s (claim 5691). These were never formally withdrawn; instead, IREN announced partnerships with Microsoft in November 2025, pivoting to AI infrastructure, and reframed Childress as a 750 MW AI hub (claim 8580) rather than mining capacity.

Capacity targets have shifted under this repositioning. Sweetwater 1 was reduced from 1,400 MW (claim 6523) to 1,210 MW by end-2027 (claim 9293), a silent downward revision. Meanwhile, near-term 2026 targets accelerated: 480 MW deployed by December 2026 across Childress and other sites (claim 8633), with an aggressive 1.2 GW goal for 2027 (claim 8634). These compressed timelines rest on acquisitions of Nostrum Group and Mirantis completed in June and August 2026.

Recent execution signals are mixed. Two GPU installation promises at Mackenzie and Childress, both pledged for July 2026 (claims 9192, 9193), have slipped six weeks past deadline with no formal acknowledgment. A Dell supply agreement carrying a March 2026 deadline shows as still delayed. Against this sits an imminent Microsoft Horizon 1 handoff due September 30, 2026—nine days from announcement date. IREN has committed to four 50 MW Horizon deployments in 2026 (claim 8564), representing critical near-term proof points.

Watch three items over the next quarter: whether Horizon 1 hands to Microsoft on or before September 30, a hard deadline visible to both parties; whether Horizons 2, 3, and 4 start arrival by year-end, indicating the 480 MW 2026 target remains viable; and whether GPU retrofit work at Childress and Mackenzie completes, resetting credibility on facility timelines. Independent verification through interconnection queue position, permit records, and power purchase agreements would clarify the underlying delivery pace versus announced velocity.

IREN's capacity trajectory shows striking inflation paired with consistent timeline slippage. The company announced 1.4 gigawatts of West Texas capacity in November 2023 (claim 6523) but has since restated targets multiple times. By May 2026, management claimed a 5-gigawatt secured power portfolio (claim 9182) and promised 480 megawatts of AI cloud capacity by year-end 2026 (claims 9290, 8633, 8638) with a 1.2-gigawatt target for 2027 (claims 8634, 8555). The narrative shifted dramatically from Bitcoin mining infrastructure, targeting 31 exahashes per second by Q4 2024 (claim 5691), to GPU cloud and Microsoft AI services anchored by a reported 9.7-billion-dollar five-year contract (claim 8112).

Nearly every major deadline has slipped. The 80-megawatt Childress expansion promised for June 2024 (claims 6495, 6410) and T21 miner shipping (claim 6408) did not materialize. End-of-year 2024 targets—31 exahashes capacity (claim 5691), 1,896 GPU units (claim 5712), and 50-megawatt Childress construction (claim 6153)—all missed. Most recently, Horizon 1, a 50-megawatt Microsoft deployment, was promised for August 13, 2026 (claim 8581) but now carries a delayed tag with revised handoff targeting September 30, 2026 (claim 9186). The Mackenzie and Childress GPU retrofit deployments missed their July 1, 2026 targets (claims 9192, 9193). The Dell GPU supply agreement, promised to complete by March 2026, remains incomplete (claim 4161).

No independent evidence—permit filings, interconnection queue positions, construction verification, or customer deployment confirmations—appears in the record. IREN's pivot from mining to AI cloud, while strategically coherent, compounds execution risk as the company now juggles Bitcoin infrastructure, GPU compute, and Microsoft AI workloads across a growing footprint that includes acquisitions of Nostrum (490 megawatts) and Mirantis.

Watch whether the four 50-megawatt Childress deployments materialize by December 31, 2026 (claim 8675), whether Horizons 2-4 ship as promised (claim 8676), and whether the revised 2027 capacity guidance of 1.2 gigawatts (claim 8634) holds or slips further. Permit records and ERCOT/ISO queue data would clarify whether land, power, and grid interconnection are actually secured.

IREN's expansion narrative reveals a company in constant recalibration, with capacity targets that have scaled dramatically while delivery discipline has lagged. Through 2024, the company announced bitcoin mining targets ranging from 10 to 31 exahashes per second by year-end; actuals reported in March 2026 (claim 3423) showed 46 EH/s, suggesting either earlier targets were conservative or later claims inflated. The GPU fleet trajectory is more troubling. Procurement began modestly with 568 H100 units in early 2024 (claim 6261), grew to 1,896 by late 2024 (claim 5712), then to 2,100 in November 2025 (claim 4033), before jumping to 99,900 units by March 2026 (claim 3422)—a sevenfold increase presented without commensurate explanation of supply chain, power routing, or deployment sites. That gap invites skepticism.

Delivery discipline has been weak. Every major timeline commitment from 2024 slipped: the 80MW Childress expansion (claim 6495), T21 miner shipping (claim 6408), and GPU commissioning (claim 6262) all missed their June 2024 window. The 50MW Childress build (claim 6153) and H200 purchase (claim 5711) missed December 2024. As recently as May 2026, the company promised Childress GPU retrofits and Mackenzie installations by July 2026 (claims 9193, 9192); it is now September.

A structural pivot starting November 2025 reshapes what to scrutinize. IREN shifted emphasis from bitcoin operations to AI cloud services, anchored by a $9.7B Microsoft contract (claim 8112) and major NVIDIA and Dell supply agreements. The Horizon data center complex emerged as the delivery flagship, with Horizon 1 handoff to Microsoft promised for September 30, 2026 (claim 9186)—that deadline is 20 days away. Four additional 50MW deployments (claim 8675) and a 480MW AI cloud capacity target (claims 8579, 8591, 8623) are promised by December 31, 2026.

Over the next 90 days, watch whether Horizon 1 actually transfers to Microsoft on schedule and in the announced 50MW configuration. The company's track record suggests both skepticism and close attention to the gap between promise and payment. Independent permitting or interconnection queue data would help; absent that, only customer acceptance provides external verification.

IREN's capacity announcements have undergone significant reframing rather than straightforward expansion. Early claims centered on Bitcoin hashrate: 10 EH/s (claim 6150), then 20 EH/s aspiration (claim 6342), escalating to 31 EH/s promised for Q4 2024 (claim 5691). By November 2025, hashrate targets stabilized at 50 EH/s (claims 4034, 4021), with 46 EH/s installation claimed by March 2026 (claim 3423). However, starting May 2026, IREN repackaged capacity as "AI Cloud" measured in megawatts: 480 MW by end-2026 and 1.2 GW by 2027 (claims 8623, 8624, 8628, 8629). This metric shift—from hash throughput to power allocation—obscures whether actual compute deployment accelerated or merely reflects how partnerships are labeled.

Timeline accountability shows persistent slippage without formal revision acknowledgment. Childress 80 MW expansion and GPU deliveries promised for June 30, 2024 (claims 6495, 6262) remain delayed. The 50 MW 2024 Childress build (claim 6153) missed its December 2024 deadline. More recently, Childress retrofit GPU delivery (July 1, 2026, claim 9193) and Mackenzie installation (July 1, claim 9192) appear to have lapsed without stated revision. Horizon 1 Microsoft handoff, promised September 30, 2026 (claim 9186), arrives in three weeks with no status update on record here.

Public evidence backing the announced pace remains undocumented in these claims. No permits, grid interconnection queue positions, or third-party site photography appear. The two major acquisitions—Nostrum Group (claim 8106, June 2026) and Mirantis (claim 8767, August 2026 for $40 million)—promise 490 MW and 650 personnel respectively, yet their integration into capacity delivery is unspecified. Microsoft ($9.7 billion, claim 8112) and Dell ($5.8 billion, claim 8577) contracts are headline-grabbing but their actual deployment pacing into IREN's sites depends on commissioning velocity that recent slips suggest may trail announced rates.

Over the next 90 days, watch Horizon 1 handoff completion by September 30 and any narrative around the missed July installations. If the 480 MW AI Cloud target holds by December 31 (claims 8579, 8623), that would represent the first on-time large-scale milestone. Conversely, any revision to 2026 targets or silence on Childress/Mackenzie slips would signal continued compression of delivery against partnership commitments.

IREN's capacity roadmap has undergone a profound strategic pivot that obscures execution reality beneath capacity growth announcements. The company opened 2023 with Bitcoin mining as core business—10 EH/s announced (claim 6522), 20 EH/s pathway articulated (claim 6342)—but shifted dramatically toward AI infrastructure by 2025. Current GPU commitments of 99,900 units (claim 3422) dwarf the 46 EH/s Bitcoin mining capacity installed as of March 2026 (claim 3423), yet this reorientation coincides with systematic timeline deterioration. The Childress campus exemplifies execution lag. In late 2023, IREN promised 80 MW expansion by June 2024 (claim 6495) and 100 MW early works (claim 6496). Neither materialized. By May 2024, the company compressed the narrative to 50 MW by year-end (claim 6153). As of September 2026, Childress retrofit GPU delivery is overdue by two months from a July 2026 promise (claim 9193), and the company is now advertising 480 MW expansion for December 2026 (claim 9290) and eventual 750 MW campus capacity (claim 8580). No permit data, TSO queue positions, or grid connection timelines are publicly available to cross-check feasibility, but the cadence of re-promised timelines suggests capacity announcements precede engineering certainty. Recent partnerships provide some countervailing evidence of acceleration. The November 2025 Microsoft contract worth 9.7 billion dollars (claims 8112, 8563) and May 2026 NVIDIA strategic investment of 2.1 billion dollars (claim 9179) carry mutual accountability. Two acquisitions closed in 2026: Nostrum (490 MW European power) completed June 2026 (claim 8106), Mirantis finalized August 2026 (claim 8767). These moves suggest operational consolidation and capital commitment, yet cannot retroactively explain the 18-month slip on basic infrastructure timelines. Monitor December 2026 carefully. IREN has promised four 50 MW Childress deployments for Microsoft by year-end (claim 8675), Horizon 1 data center handoff (claim 8561), and Horizons 2-4 delivery of 150 MW by the same deadline (claim 8589). Mackenzie GPU installation and Childress retrofit are already late. If end-2026 AI commitments miss again, the company's ability to capitalize on strategic partnerships erodes sharply.

IREN's announced infrastructure trajectory shows sustained ambition coupled with systematic execution slippage. The company began in late 2023 with a focused Bitcoin mining narrative centered on Childress and Sweetwater (claim 6523), progressing from 10 EH/s to 20 EH/s targets through early 2024. The introduction of GPU capacity (claim 6261, January 2024) marked a strategic pivot toward AI cloud services, a shift that accelerated through acquisitions of Nostrum Group and Mirantis (claims 9294, 9295, May 2026) to secure additional power footprint.

Capacity announcements have consistently grown without corresponding timeline compression. A promised 50 MW Childress expansion by end of 2024 (claim 6153) was first delayed, then reframed as part of a 480 MW 2026 target by May 2026 (claim 9290), now further articulated as a 750 MW eventual campus (claim 8580, August 2026). The company has missed delivery dates across three device categories: Bitmain T21 miners (promised June 2024, claim 6408), NVIDIA H100/H200 GPUs (promised December 2024, claims 5711, 5712), and now Mackenzie GPU installation and Childress retrofit deployment (both promised July 2026, claims 9192, 9193—now two months overdue as of September 9, 2026).

The $9.7 billion Microsoft contract (claim 8112, November 2025) appears to be driving the Horizon series ramp at Childress. Horizon 1 handoff is promised by September 30, 2026 (claim 9186)—less than three weeks away—representing the first major test of the company's ability to deliver on current-year commitments. Horizons 2-4 are slated for 150 MW delivery by year-end (claim 8589). Without independent verification of power interconnection progress, substation energization status at Sweetwater (claim 9190), or on-the-ground construction completion, the timeline credibility remains uncertain.

The pattern reveals a company pursuing aggressive scale-up while contending with supply chain and infrastructure execution friction. The near-term bellwether is Horizon 1 completion; any further delay would suggest the 480 MW end-of-2026 and 1.2 GW end-of-2027 targets (claims 8591, 8555) face material risk. Permitting velocity at Sweetwater 1 and actual power delivery from existing renewable partnerships will determine whether the NVIDIA strategic partnership capacity (claim 9289) materializes as planned.

IREN's stated capacity trajectory shows consistent acceleration paired with systematic delivery delays. Bitcoin hashrate targets escalated from 10 EH/s (November 2023, claim 6522) to 20 EH/s (January 2024, claim 6342) to 31 EH/s promised by end-2024 (claim 5691), then 50 EH/s by late 2025 (claims 4034, 4021). Yet March 2026 filings reported only 46 EH/s installed (claim 3423), a contraction from the 50 EH/s capacity announced four months prior (claim 4034). This discrepancy between stated and operating figures persists across timelines.

Delivery discipline deteriorated sharply through 2024. The 80MW Childress expansion promised for June 2024 (claim 6495) missed; so did T21 miner shipments (claim 6408) and GPU delivery of 568 H100 units (claim 6262). A promised 31 EH/s expansion for Q4 2024 did not materialize (claim 5691). By May 2026, two new hard deadlines—50MW Childress retrofit GPU delivery by July 1 and 80MW Mackenzie GPU installation by the same date (claims 9193, 9192)—were marked as delayed before the dates arrived, indicating internal tracking revealed slippage well in advance.

The GPU ledger raises further questions. Claims shifted from 568 H100 units (February 2024) to 1,896 units post-H200 purchase (September 2024) to 2,100 units (November 2025) to 99,900 units on order or installed (March 2026). A 47-fold increase in four months suggests either aggressive forward commitments without near-term fulfillment or inconsistent inventory classification across filings.

The pivot to AI infrastructure—crystallized in the $9.7 billion Microsoft contract (claim 8112) and Nostrum/Mirantis acquisitions (claims 8106, 8767)—creates urgency. Horizon 1's promised handoff to Microsoft by September 30, 2026 (claim 9186) is now 21 days away. Sweetwater substation energization for 200MW is due the same day (claim 9190). Four additional 50MW Horizon deployments are promised by year-end (claim 8675), along with Horizons 2-4 delivery (claim 8558). These commitments are material and to named customers; slippage would signal capability gaps at scale.

Watch whether IREN clears the September 30 Horizon and Sweetwater milestones without invoking force majeure. If those land on time, confidence in the 1.2GW 2027 target rises; if either slides, expect cascading delays through the Q4 Horizon deployments and eventual 2027 expansion claims.

IREN has repositioned from bitcoin mining to AI cloud infrastructure over three years, with announced capacity targets that have consistently outpaced execution. Bitcoin mining targets escalated from 10 EH/s (Nov 2023, claim 6522) through 20 EH/s (Sept 2024, claim 5689) to 50 EH/s (Nov 2025, claim 4034), yet actual installed hashrate was 46 EH/s by March 2026 (claim 3423), suggesting the latest target also receded. A 31 EH/s Q4 2024 expansion (claim 5691) was marked delayed and subsequently displaced from announcements without explicit revision.

Facility timelines show the same pattern. Childress 80 MW expansion promised for June 2024 (claim 6495) remains unfinished; GPU delivery to Childress for June 2024 (claim 6262) slipped. NVIDIA H200 acquisition and GPU fleet expansion to 1,896 units, both promised for December 2024 (claims 5711, 5712), were delayed. Mackenzie and Childress GPU deployments promised for July 2026 (claims 9192, 9193) also appear to have missed dates. Recent announcements from May through August 2026 restate targets without acknowledging these earlier misses.

The pivot to AI cloud infrastructure brings substantial partnerships: a $9.7 billion five-year Microsoft agreement (claim 8112), NVIDIA strategic investment (claim 9179), and Dell GPU supply deals at $1.6 billion (claim 8992). Acquisitions of Nostrum Group (completed June 2026, claim 8106) and Mirantis (completed August 2026, claim 8767) suggest capital deployment and strategic commitment. Announced capacity targets of 480 MW by year-end 2026 (claims 8633, 8638) and 1.2 GW by end 2027 (claims 8634, 8629) represent substantial expansion.

Yet execution risk persists. Horizon 1 Microsoft handoff is scheduled for September 30 (claim 9186)—less than three weeks away—with no independent grid or construction evidence yet public. Four 50 MW Horizon deployments at Childress must complete by year-end 2026 (claims 8564, 8675) to achieve the annual target. The 99,900 GPU order announced March 2026 (claim 3422) represents concentration risk against NVIDIA allocation constraints.

Watch for Horizon 1 completion confirmation by late September, Sweetwater 1 substation energization status (claim 9190), and visible progress on 2026 capacity milestones by November, which would distinguish genuine operational acceleration from optimistic restatement.

IREN has executed a consistent upward trajectory in announced bitcoin mining capacity from late 2023 through early 2026, scaling from a 10 EH/s target (claim 6522) through 20 EH/s installed (claim 5689) to 50 EH/s stated capacity (claims 4034, 4021). The progression indicates operating deployment against targets, with claimed hashrate rising from 10 EH/s in May 2024 (claim 6150) to 20 EH/s by September 2024 (claim 5689). Bitcoin mining has been substantially eclipsed by a strategic GPU infrastructure pivot beginning in late 2024. GPU unit commitments escalated from 568 H100 units promised by June 2024 (claim 6261) to 1,896 by year-end 2024 (claim 5712), then 2,100 (claim 4033), and dramatically to 99,900 units by March 2026 (claim 3422), reflecting a pivot toward AI cloud services under the November 2025 $9.7 billion Microsoft contract (claim 8112).

The facility timeline record reveals systematic slippage paired with unrevised capacity guidance. Childress 80MW expansion missed June 2024 (claims 6495, 6410). GPU delivery to Childress and Mackenzie missed July 2026 deadlines (claims 9193, 9192). The Dell GPU supply agreement missed March 2026 (claim 4161). H200 procurement missed December 2024 (claims 5711, 5712). Most recently, Horizon 1 delivery commitments shifted from September 30 (claim 9186, May 2026) to December 31 (claim 8588, August 2026) without public acknowledgement of the revision. Capacity guidance announced in May 2026—480 MW by end 2026, 1.2 GW by end 2027—remained unchanged through August (claims 8633, 8634), despite these accruing delays.

No independent verification exists in the record: grid interconnection status, permitting progress, or physical site advancement are absent. August 2026 announcements commit four Horizon deployments (200 MW aggregate) to Microsoft at Childress by December 31 (claims 8564, 8588, 8675, 8676).

In the next 90 days, monitor whether the December 31 Horizon delivery target holds (claims 8558, 8676, 8588) and whether Sweetwater 1 substation energization (claim 9190) proceeds. Dell supply fulfillment for Blackwell GPUs (claim 8992) and third-party construction reporting at Childress would provide essential independent counterweight to the pattern of timeline slippage with unrevised capacity targets.

IREN's infrastructure trajectory reveals fundamental strategic realignment masked by repetitive capacity claims. Bitcoin mining dominated announcements through 2024—promising 10 EH/s by Q1 2024 (6522) and later 31 EH/s by Q4 2024 (5691)—but hit consistent delivery slips on the 80 MW Childress expansion (promised June 2024, marked delayed; claims 6495, 6410) and T21 miner acquisitions (claim 6408). By March 2026, IREN stated only 46 EH/s installed (3423), abandoning mining as its core narrative without acknowledging the pivot. The company's capacity announcements contracted from 50 EH/s (claims 4034, 4021 from late 2025) to this lower figure.

The pivot to GPU-heavy AI infrastructure crystallized suddenly between September 2024 and November 2025, coinciding with the $9.7B Microsoft contract (claim 8112). GPU counts escalated dramatically: 1,896 units promised by December 2024 (5712), then 2,100 by November 2025 (4033), then 99,900 by March 2026 (3422)—suggesting rescoped or acquired capacity rather than linear growth. The Dell supply agreement ($5.8B per claim 8577, originally promised March 2026 per 4161 but marked delayed) anchors Blackwell deployments at Childress through 2027.

Recent claims exhibit troubling redundancy. From May through August 2026, identical capacity targets appear repeatedly across multiple claims: 480 MW by December 2026 appears in claims 8638, 8633, 8628, 8623, 8591, 8579, 8565, and 8554. This pattern suggests either placeholder messaging or inability to differentiate delivery milestones. The Horizon 1 handoff to Microsoft is promised for September 30 (claim 9186)—now 23 days away—representing the first concrete test of execution. Meanwhile, Childress GPU delivery (50 MW, claim 9193) and Mackenzie deployment (80 MW, claim 9192) were promised July 1, 2026, now 67 days past with no reclassification to delayed status.

Independent corroboration remains sparse. Sweetwater 1 substation energization (200 MW target, claim 9190) carries no timeline. The 1.2 GW AI capacity target by end-2027 (claim 8634) depends on clearing the 480 MW intermediate gate by December 31. With Horizon 1's September 30 deadline imminent and the December 31 cluster of capacity commitments (claims 8638, 8675, 8676) approaching, IREN faces its clearest performance accountability test since pivoting from mining to cloud infrastructure. Watch whether missed July dates get acknowledged and whether Microsoft's Horizon 1 acceptance proceeds on schedule.

IREN's capacity roadmap reveals substantial upward revision paired with serial date slippage. The company began 2024 targeting 10 exahashes of bitcoin mining hashrate (claim 6341), which reached 20 EH/s claimed by September (claim 5689) before promising 31 EH/s by December 2024 (claim 5691). As of March 2026, installed hashrate stands at 46 EH/s (claim 3423), suggesting the December 2024 target slipped but growth continued. Concurrently, IREN pivoted sharply toward GPU-based AI capacity, starting with 568 H100 units promised by mid-2024 (claim 6262, marked delayed), escalating to 1,896 units by year-end 2024 (claim 5712, also delayed), and exploding to 99,900 total units on order or installed by March 2026 (claim 3422). This 176-fold scale-up in GPU commitments in less than a year suggests either aggressive portfolio acceleration or tracking imprecision.

The Microsoft relationship, formalized in November 2025 with a $9.7 billion five-year contract (claim 8112), has become the primary growth narrative. IREN committed 480 MW of AI cloud capacity by end-2026 (claims 8623, 8633, 8565, 8579) and 1.2 GW by end-2027 (claims 8566, 8624, 8639). The Horizon project at Childress carries conflicting metrics: Horizon 1 is stated as 50 MW (claim 8539) yet Microsoft is scheduled to receive a 300 MW handoff by September 30 (claim 9186). Four separate 50 MW deployments are also promised by year-end 2026 (claim 8675), raising questions about whether these tranches comprise the larger 300 MW or represent additional capacity. Earlier Childress expansion claims from 2023–2024 (80 MW by June 2024, claim 6410; 100 MW early works, claim 6496) remain unfulfilled and are now subsumed into the Horizon narrative.

Regional expansion claims include Mackenzie at 80 MW GPU by July 1, 2026 (claim 9192, now past) and Childress retrofit at 50 MW by the same date (claim 9193). These July commitments appear to have missed their windows with no revision announced. Watch for confirmation of Horizon 1 Microsoft acceptance by September 30 as a validation gate. Securing the full 480 MW by December 31 would support 2027 target credibility. Dell GPU agreement details remain undated post-delay (claim 4161 was March 2026). Horizon 2-4 delivery by year-end (claim 8676) carries commitment but lacks capacity specificity. Independent evidence of substation energization at Sweetwater 1 (claim 9190, no deadline) and Canadian site execution will strengthen assertions about distributed expansion pace.

IREN has undergone a remarkable strategic pivot from bitcoin mining to GPU cloud infrastructure, with announced capacity scaling dramatically since late 2023. The bitcoin mining roadmap started modestly—10 EH/s in November 2023 (6522)—before jumping to a 20 EH/s pathway by January 2024 (6342). By September 2024, claims had escalated to 20 EH/s installed with a Q4 2024 target of 31 EH/s (5689, 5691), but that deadline appears to have slipped. By November 2025, the company reset to 50 EH/s (4034, 4021), with March 2026 reporting 46 EH/s installed (3423)—suggesting earlier expansion targets were not met on their announced schedule.

The GPU buildout has been far more volatile. February 2024 disclosed 568 NVIDIA H100s (6261). September 2024 raised this to 1,896 units (5712). A staggering revision came in March 2026: 99,900 GPU units on order or installed (3422), a 52-fold increase from the prior figure. This pivot coincided with blockbuster partnership announcements including a $9.7 billion Microsoft five-year contract (8112, 8563) and a $5.8 billion Dell GPU supply agreement (4161).

Timeline discipline has been consistently poor. The 80 MW Childress expansion promised for June 2024 (6495, 6410) is marked delayed. GPU deliveries promised for June 2024 (6262) and NVIDIA H200 purchases due December 2024 (5711, 5712) both slipped. The 31 EH/s expansion promised for Q4 2024 (5691) is flagged delayed. Even the most recent assertions—Horizon 1 delivery by September 30 (9186), four 50 MW deployments by year-end, 480 MW capacity by December 2026 (8565, 8554)—carry execution risk given the pattern.

Independent validation is absent. No FERC interconnection queue entries, grid studies, or on-the-ground reporting appear in the record to confirm commissioning pacing. The narrative relies on announced partnerships and customer prepayments (45 percent of GPU capex funded this way, claim 8595), which can prove contingent on delivery milestones IREN fails to meet.

Over the next 90 days, Horizon 1 handoff to Microsoft by September 30 is the nearest hard milestone. Then track the four 50 MW Horizon deployments (8675) and Horizons 2-4 delivery (8676) both due by December 31. These dates will reveal whether execution discipline has tightened or whether 2026 joins the catalog of missed timelines.

IREN's announced capacity trajectory reflects a fundamental business pivot rather than conventional scaling. From 2023-2024, the company made discrete claims around bitcoin mining infrastructure, targeting 10 EH/s operational capacity (claim 6150, May 2024) and eventually 50 EH/s (claims 4034, 4021, late 2025). Simultaneously, claims for data center power grew from 1400 MW at Sweetwater 1 (claim 6523, November 2023) toward fragmented regional targets across Childress, British Columbia, and Australia. Beginning in May 2026, this shifted dramatically: bitcoin hashrate was reframed as secondary (46 EH/s installed per claim 3423, March 2026) while GPU procurement exploded to 99,900 units on order (claim 3422). AI cloud capacity targets jumped to 480 MW by end-2026 (claims 8633, 8638) and 1.2 GW by 2027 (claim 8639).

Timeline adherence has been consistently poor through 2024-2025. The 80 MW Childress expansion promised for June 2024 (claims 6495, 6410) and 50 MW for December 2024 (claim 6153) are marked delayed. T21 miner shipments (claim 6408) and H200 GPU purchases (claim 5711) both slipped from their 2024-Q4 targets. However, the pattern shifted in August 2026 with claims that Horizon 1 (50 MW to Microsoft) was both delayed past its August 13 target (claim 8581) and delivered (claim 685). This contradiction is critical: if Horizon 1 slipped, the four remaining 50 MW Microsoft deployments promised by December 2026 (claim 8675) face compressed timelines with only 115 days remaining from today.

Independent verification is limited in the provided claims. No permits, interconnection queue data, or construction permitting is cited. The company's Nostrum Group acquisition (claim 9188, announced May 2026 for 490 MW European platform) was claimed completed by June 2026 (claim 8106), but this recent acquisition's operational status and contribution to capacity targets remains opaque. The Microsoft partnership announced May 2026 ($9.7 billion five-year contract, claim 8112) and Dell GPU supply agreement iterations ($5.8 billion, then $1.6 billion) suggest real capital commitments, yet the claims themselves do not substantiate delivery timelines or actual on-the-ground progress.

Watch for: (1) Whether Horizon 1 and Horizons 2-4 hand off to Microsoft on the claimed December 2026 schedule; (2) any revision to the 480 MW end-2026 AI cloud target; (3) updates on Nostrum Group's European power integration and whether it contributes to 2027 capacity; (4) public permitting or utility interconnection announcements for the remaining three Microsoft 50 MW builds at Childress. The next 90 days will reveal whether the May-August 2026 capacity and timeline claims represent realistic execution or continue the pattern of historical delays.

IREN has aggressively reoriented its asset portfolio from Bitcoin mining toward AI cloud services over the past nine months, with capacity claims inflating across both vectors before the company narrowed messaging. The Bitcoin expansion path shows a sharp three-stage climb: 10 EH/s (claim 6522, Nov 2023), 20 EH/s (claim 5689, Sept 2024), then 50 EH/s (claims 4034, 4021, late 2025). However, March 2026 disclosures report only 46 EH/s installed (claim 3423), a gap of 4 EH/s, and recent August announcements contain no Bitcoin capacity figures despite the asset class remaining operationally material. This silence warrants monitoring.

GPU procurement claims have scaled dramatically without matching disclosure of deployment velocity. Initial procurement targets of 568 H100s (claim 6261, Feb 2024) and 1,896 H200s (claim 5712, Sept 2024) represented single-digit percentages of the 99,900-unit inventory claimed by March 2026 (claim 3422). Facility capacity claims have also inflated: Sweetwater started at 1,400 MW (claim 6523, Nov 2023) and now targets 2,000 MW (claim 8597, Aug 2026). The Horizon project cluster—Microsoft-directed deployments—claims 200 MW combined load at Childress (claim 8578), with individual units of 50 MW each (claims 8539, 8588). August announcements claim Horizon 1 as delivered (claim 8625) while simultaneously promising 2026-12-31 handoff (claim 8588), a contradiction requiring clarification.

Timeline slippage is chronic. The H200 acquisition ($43.9B, claim 5711) missed its 2024-12-31 deadline, as did 50 MW Childress construction (claim 6153) and T21 miner shipments (claim 6408). The Dell GPU agreement ($5.8B, claim 4161) missed a 2026-03-01 promised date. Yet August 2026 claims present Horizons 2–4 (150 MW, claim 8589) and the 480 MW AI cloud target (claims 8565, 8554) as deliverable by year-end with no acknowledgement of historical pattern. Independent evidence—permit filings, grid interconnection queue position, third-party site reports—is absent from the record, limiting verification of construction pace or grid readiness for claimed rollouts.

The next 90 days determine credibility: whether Horizons 2–4 are materially deployed, whether Sweetwater or other announced sites show measurable progress, and whether August's missing Bitcoin context reflects strategic repositioning or merely temporary omission.

IREN's capacity narrative has undergone substantial reorientation and scaling since late 2023. The company began with blockchain mining focus—promising 10 EH/s expansion (6522) in November 2023, escalating to a 20 EH/s pathway (6342) by January 2024, then 31 EH/s target for Q4 2024 (5691). Current operating claims show 46–50 EH/s (4021, 4034, 4023), though the gap between promised and delivered figures reflects persistent execution challenges. GPU infrastructure has become the company's primary growth vector. H100 procurement was promised for June 2024 delivery (6262); a $43.9M H200 acquisition was due December 2024 (5711). Both slipped without formal revision. GPU inventory claims then exploded from 2,100 units claimed in November 2025 (4033) to 99,900 units by March 2026 (3422)—a 47-fold expansion within months that likely reflects aggressive forward commitments rather than completed supply.

Schedule adherence has deteriorated consistently. Childress 80MW expansion promised for June 2024 (6495, 6410) remains outstanding 26 months into slippage. H100 delivery (6262) missed June 2024. The 31 EH/s Q4 2024 target (5691) never materialized. Horizon 1 data center delivery was scheduled for August 13, 2026 (8581)—today's date—and is already marked delayed (8581). This represents real-time miss of a flagship Microsoft commitment.

Capacity targets have expanded sharply. Sweetwater is projected at 2,000 MW (8597); Childress at 750 MW (8596). The 480 MW AI cloud target for end-2026 (8623, 8633) and 1.2 GW for end-2027 (8592, 8624) depend entirely on executing Horizon deployments (8675, 8676, 8589) that have already begun slipping. Microsoft represents the anchor customer across these projects, with contracts weighted at 4.0-year average terms (8108).

Over the next 90 days, track: Horizon 1 acceptance by Microsoft despite the August 13 delay (8561); delivery of Horizons 2-4 within 137 days remaining in 2026 (8558, 8676); and resolution of the Dell GPU supply agreement originally due March 2026 (4161). The 480 MW year-end capacity target assumes flawless execution of all Horizon cohorts, a trajectory now demonstrably under pressure given Horizon 1 has already missed its deadline before any units shipped.

IREN's capacity narratives have undergone radical expansion over three years, yet delivery patterns reveal systemic timeline slippage masked by revised claims. In late 2023, the company announced 1,400 MW for West Texas development (6523); by August 2026, it claimed eventual capacity of 750 MW at Childress and 2,000 MW at Sweetwater (8596, 8597)—a consolidation rather than growth despite adding AI infrastructure targets. Early Bitcoin mining commitments illustrate the drift: Q4 2024 targets for 31 EH/s expansion (5691) appear abandoned, with current capacity stated at 46–50 EH/s (3423, 4021, 4034) without acknowledged revision. GPU procurement figures jumped inexplicably from 2,100 units in late 2025 (4033) to 99,900 by March 2026 (3422), yet supporting infrastructure claims remained flat. The Horizon program presents the clearest delivery friction. Claims from August 13, 2026 state Horizon 1 was delivered to Microsoft at Childress (8625, 8561), yet the same date carries a 'delayed' marking for Horizon 1 (8581) with promised delivery on August 13 itself—suggesting post-hoc revision. Earlier GPU delivery commitments for June 30, 2024 (6262) and H200 procurement for December 31, 2024 (5711) both slipped, now merged into a broader AI cloud target of 480 MW by year-end 2026 (364, 8633) and 1.2 GW by end of 2027 (8634). The Dell supply agreement ($4.161B, claim 4161) was delayed from March 1, 2026 with no subsequent promised date. Multiple identical capacity and timeline claims from a single day (August 13) suggest bulk database updates rather than discrete milestones. Over 90 days, watch whether Horizons 2–4 ship by December 31 (8676, 8589) and whether 480 MW AI cloud actually materializes or faces revision. Track any acknowledgment of the Horizon 1 discrepancy and whether Bitcoin mining targets are formally abandoned or re-baselined. Verify whether the 99,900-unit GPU figure represents actual purchase commitments or modeling overflow.

IREN's announced capacity has undergone significant restatement and directional pivot since late 2023. Bitcoin mining hashrate targets climbed from 10 EH/s (claim 6522, Nov 2023) to 31 EH/s by Q4 2024 (claim 5691, Sept 2024), yet achieved 46 EH/s by March 2026 (claim 3423) before stabilizing at 50 EH/s by late 2025 (claims 4034, 4021). GPU capacity, by contrast, underwent explosive upward revision: from 568 H100s announced in February 2024 (claim 6261) to 99,900 units by March 2026 (claim 3422), reflecting a strategic shift toward AI infrastructure rather than cryptocurrency mining. Sweetwater's planned capacity expanded from 1,400 MW (claim 6523, Nov 2023) to 2,000 MW (claim 8597, Aug 2026). Childress evolved from a 80 MW expansion promised by June 2024 (claims 6495, 6410) to an eventual 750 MW footprint (claims 8596, 8580). Multiple early deadlines were substantially missed: Childress 80 MW was promised June 2024 but remains undelivered as of September 2026; T21 miner shipments promised June 2024 (claim 6408) slipped; GPU commissioning promised June 2024 (claim 6262) slipped; and 31 EH/s target for Q4 2024 (claim 5691) was not achieved. These delays were not explicitly acknowledged as revisions in subsequent filings; instead, later announcements simply restated different targets. The August 2026 announcements (most claims dated 2026-08-13) mark a consolidation phase: Horizon 1, a 50 MW AI deployment, is stated as delivered to Microsoft (claims 685, 8625), with Horizons 2–4 (150 MW total) promised for delivery by year-end 2026 (claims 8675, 8589). A $9.7 billion five-year Microsoft contract (claim 686) and a strategic NVIDIA partnership covering 5 GW (claim 8642) provide anchor demand. IREN projects 480 MW AI capacity by end 2026 (claims 8638, 8633) and 1.2 GW by end 2027 (claims 8639, 8624). The claims dataset contains no independent verification: no permit filings, interconnection queue evidence, power purchase agreements, or on-site reporting substantiate the delivery pace. Over the next 90 days, critical tests include whether Horizons 2–4 materialize on schedule before December 2026, whether the 480 MW target is approached by year-end, and whether capacity utilization data or customer acceptance milestones emerge. Supply-chain documentation for Dell ($1.6 billion deal, claim 1297) and NVIDIA component procurement will be essential to validate feasibility.

IREN's trajectory reveals a dramatic strategic pivot from Bitcoin mining to AI infrastructure, shadowed by persistent execution gaps and an absence of independent corroboration. The company began 2023 targeting 10 exahash Bitcoin capacity and 1,400 MW at Sweetwater, then escalated Bitcoin ambitions to 31 exahash by end-2024 and claimed 50 exahash capacity by 2025. Yet multiple near-term deadlines slipped without revision: 80 MW Childress expansion promised by mid-2024 (claims 6410, 6495) delayed; 568 H100 GPUs and GPU delivery by June 2024 (claim 6262) delayed; 50 MW Childress by end-2024 (claim 6153) delayed. By March 2026, IREN reported only 46 exahash installed—a gap versus announced 50 exahash targets—suggesting modest underperformance on the core Bitcoin business.

The 2025–2026 pivot to AI, anchored by a $9.7 billion Microsoft contract, introduced new capacity claims: 480 MW AI cloud by end-2026 and 1.2 GW by end-2027. Notably, IREN published thirty-plus near-identical claims on a single date (2026-08-13), all reiterating these targets. Horizon 1 (50 MW) is stated delivered; Horizons 2-4 (150 MW) are promised by December 2026. However, no independent evidence appears in the record—no interconnection queue entries, no utility coordination filings, no building permits or regulatory sign-offs supporting a 1.2 GW deployment at Childress or Sweetwater by 2027. The June 2026 Nostrum Group acquisition (claim 8106) appears to consolidate capacity rather than grow it, suggesting IREN may be absorbing underutilized or stranded assets to meet targets.

The Microsoft and Dell partnerships provide some visibility: customer prepayments represent 45 percent of GPU capex (claim 8595), indicating committed funding. Yet prepayments alone do not guarantee delivery velocity. The Dell agreement was also revised downward in May 2026 from $5.8 billion to $1.6 billion (claims 4161 vs. 1297), a material contraction signaling either scope reduction or renegotiation pressure.

Over 90 days, watch for: (1) whether Horizons 2-4 activate by year-end, validated by power interconnection data or satellite imagery showing rack density at Childress; (2) any public filing, permit, or utility agreement confirming capacity buildout pace; (3) whether prepayment or customer contract terms slip; (4) whether Microsoft or Dell issue any caution on deployment timelines. The lack of independent infrastructure documentation so far is the most material risk signal.

IREN's announced capacity trajectory reveals a company that pivoted from bitcoin mining to AI cloud infrastructure while carrying forward a string of delayed commitments. From November 2023 through early 2024, IREN announced bitcoin hashrate targets of 10 EH/s (claim 6522), then 20 EH/s pathway (claim 6342), backed by Bitmain T21 miner acquisitions. Nearly every near-term deadline from this period slipped: the 80 MW Childress expansion promised for June 2024 (claims 6410, 6495) remains delayed, as do T21 shipments (claim 6408) and GPU commissioning (claim 6262). By September 2024, IREN claimed 31 EH/s capacity by Q4 2024 (claim 5691, delayed); by March 2026, only 46 EH/s was installed (claim 3423), falling short of the 50 EH/s stated target announced four months prior (claims 4034, 4021).

The $9.7 billion Microsoft AI cloud contract signed in November 2025 (claim 8112) marked a sharp strategic reorientation. Capacity announcements thereafter emphasize AI infrastructure rather than hashrate. IREN claims Horizon 1—a 50 MW deployment—was already delivered to Microsoft as of mid-August 2026 (claim 8581). The company now targets 480 MW AI cloud capacity by December 2026 and 1.2 GW by end-2027 (claims 8638, 8639), with stated facility endstates of 750 MW at Childress and 2,000 MW at Sweetwater (claims 8596, 8597). GPU inventory claims surged from 1,896 units in September 2024 (claim 5712) to 99,900 units on order by March 2026 (claim 3422), suggesting deep customer prepayments—45 percent cited in claim 8595.

The 480 MW year-end target requires roughly 430 MW delivered in 120 days. This is plausible only if Horizons 2-4 (claimed at 150 MW total in 2026, claim 8589) approach completion. IREN's stated partnerships—NVIDIA's 5 GW pipeline (claim 8642) and Dell's $5.8 billion agreement (claim 8577)—signal serious customer backing. However, the company's history of date slippage through 2024 and absence of independent verification (permits, interconnection queue entries, third-party site reporting) counsel caution. Watch for public acceptance notices from Microsoft for Horizons 2-4 by November, and any interim guidance revisions. Any delay beyond December would indicate execution risk persisting into 2027 and cast doubt on the 1.2 GW trajectory.

IREN's infrastructure narrative has undergone dramatic recalibration over eighteen months, marked by capacity scaling that obscures unmet Bitcoin milestones and a pivot toward AI workloads backed by headline-grabbing contracts. The Bitcoin mining roadmap illustrates the pattern. In January 2024, IREN announced a 10 exahash per second expansion target (claim 6341) and a pathway to 20 EH/s (claim 6342). By May 2024, the company claimed 10 EH/s operational (claim 6150). Four months later, September 2024, a 20 EH/s installed baseline appeared (claim 5689), alongside a promise to reach 31 EH/s by Q4 2024 (claim 5691)—a target that was never delivered. The next public capacity claim, in November 2025, shifted to 50 EH/s (claim 4034). By March 2026, that announced 50 EH/s had quietly become 46 EH/s installed (claim 3423), a 8 percent downward revision presented without qualification. The Childress 80 megawatt expansion, promised twice for June 2024 (claims 6495, 6410), was never completed; instead, IREN's August 2026 announcements cite a 750 megawatt eventual Childress capacity (claim 8596) and a 2,000 megawatt Sweetwater capacity (claim 8597)—targets, not delivered facts.

The AI contract announcements distract from execution risk. A Dell GPU supply agreement announced at $5.8 billion in November 2025 (claim 4161), delayed to March 2026, reappeared six weeks later as a $1.6 billion deal (claim 1297)—a 72 percent reduction offered without explanation. GPU unit claims have been equally unstable: 568 units in February 2024, 1,896 by September 2024, then 2,100 in November 2025, before jumping to 99,900 units claimed on order (claim 3422) and 150,000 units targeted (claim 8317) in August 2026. The Microsoft contract for $9.7 billion over five years (claim 8590), announced November 2025, is significant but remains unvalidated by independent interconnection queue filings or permitting records. IREN delivered Horizon 1, a 50-megawatt data center, by August 13, 2026 (claim 685), but the company now promises Horizons 2–4 (an additional 150 megawatts) by year-end 2026 (claim 8626)—four months to deliver what the Bitcoin roadmap failed to deliver across two years. Watch whether those December 2026 completions hold. Monitor Childress grid interconnection filings and whether the 480-megawatt 2026 AI capacity target (claim 8623) materializes, as all prior infrastructure timelines have slipped without public acknowledgement. Track Dell and NVIDIA supply stability following the unexplained magnitude shift in the GPU agreement.

IREN has undergone a dramatic strategic pivot from bitcoin mining to AI cloud infrastructure, with capacity claims scaling upward but delivery timelines showing persistent strain. Through 2023-2024, the company pursued a crypto-mining trajectory, targeting 10-31 EH/s with promised milestones clustered around mid-2024. Claims 6495 and 6410 both promised 80 MW at Childress by June 2024, marked delayed; claim 6262 targeted 568 H100 GPUs by the same date, also delayed; claim 5691 projected 31 EH/s by December 2024, flagged delayed. This pattern of announced capacity without timely delivery suggests either underestimated build complexity or resource constraints that persisted through 2024.

Starting November 2025, IREN announced major partnerships reshaping its trajectory: a $9.7B five-year Microsoft contract (claim 8112) and $5.8B Dell GPU supply agreement (claim 8577), followed by a $1.6B Dell amendment (claim 1297). The shift reflects capacity repurposing: bitcoin hashrate claims remained at 50 EH/s (claims 4034, 4021), but GPU inventory ballooned from 1,896 units (claim 5712, delayed from December 2024) to 99,900 units on order (claim 3422) by March 2026. Most recent claims, concentrated in August 2026, project 480 MW AI capacity by year-end (claims 364, 8591) and 1.2 GW by end-2027 (claims 365, 8592), with Horizon 1 delivery to Microsoft claimed complete and Horizons 2-4 targeting 150 MW delivery in Q4 2026 (claim 8589).

Capacity expansion claims have scaled aggressively: Sweetwater now targets 2,000 MW (claim 8597), Childress 750 MW (claim 8596), and a South Australia campus of 800 MW operational by January 2028 (claim 8286). However, independent verification of delivery pace remains absent from available disclosures. The August 2026 announcement bundle (claims 8588, 8589, 8564, 8318) stacks multiple near-term commitments—four 50 MW Horizon modules, 150,000 GPUs by year-end—against a company that consistently missed 2024 milestones. Watch Q4 2026 Microsoft acceptance metrics, Dell GPU logistics flow, and whether the 480 MW target materializes; watch also for the South Australia permitting and construction pace, as the 2028 deadline will test whether this represents genuine capacity or announcement-driven positioning.

IREN's capacity claims have undergone a dramatic transformation alongside a strategic pivot from bitcoin mining to AI cloud services. Bitcoin hashrate targets escalated from 10 EH/s (Nov 2023) to 50 EH/s (Nov 2025), though actual installed capacity stood at 46 EH/s as of March 2026—a consistent shortfall suggesting execution friction. More remarkably, GPU capacity expanded from 568 units (Feb 2024) to a claimed 150,000 units by August 2026, with interim claims reporting 1,896 units and then 99,900 units on order or installed, indicating a volatile trajectory or serial revisions to deployment plans. The August 2026 announcement package reflects aggressive ambition: 480 MW of AI cloud capacity by year-end 2026 and 1.2 GW by end-2027, anchored by a $9.7 billion five-year Microsoft contract and a new $3.4 billion NVIDIA agreement. However, multiple facility timelines have slipped. Key 2024 deadlines—including 80 MW Childress expansion (June), GPU delivery of 568 units (June), and 31 EH/s hashrate delivery (December)—all missed their stated windows. A $5.8 billion Dell GPU supply agreement also slipped from March 2026. On the ground, only Horizon 1 (50 MW, Childress) has reportedly been accepted by Microsoft; Horizons 2–4 are promised for 2026 but remain unconfirmed. Geographic expansion claims are ambitious but unmoored: South Australia (800 MW, 2028), Spain (490 MW), and Oklahoma (1.6 GW) lack clear timelines or interconnection evidence. No permits, grid queue entries, or third-party facility reports corroborate the deployment velocity implied by August's announcements. The company's pattern suggests serial capacity upgrades tied to contract wins rather than a unified deployment roadmap. In the next 90 days, watch for: concrete confirmation of Horizons 2–4 handoffs, actual AI cloud MW online by November, and any revision to the end-2026 480 MW target, which is now only weeks away. Track Dell GPU supply shipment data and any public power-on announcements for international sites.

IREN's announced capacity has experienced sustained, unacknowledged revision across its core workloads. Bitcoin mining capacity started at 10 EH/s (claims 6522, 6341), scaled to 20-21 EH/s (claims 6342, 5690), then jumped to 50 EH/s (claims 4034, 4021) without documented explanation for the gap. As of March 2026, installed hashrate sits at 46 EH/s (claim 3423)—below the stated 50 EH/s—widening the delta between announced and operational capacity.

GPU deployments show even more dramatic variance. The trajectory moved from 568 units (claim 6261) to 1,896 units (claim 5712) to 2,100 units (claim 4033) to 99,900 units "installed or on order" (claim 3422) to a 150,000-unit expansion target (claim 8317). Without clarification on what "on order" means versus deployed, these figures lack transparency.

Timeline slippage has been pervasive and persistent. The Childress 80MW expansion and 100MW early works, both promised for June 2024 (claims 6495, 6410, 6496), are flagged as delayed with no revised completion date published. T21 miner shipping (claim 6408) and GPU delivery (claim 6262), also June 2024 targets, remain unresolved. The 31 EH/s expansion slated for Q4 2024 (claim 5691) has not materialized; no August 2026 claim updates its status.

IREN announced it delivered and accepted Horizon 1 (50 MW, claim 723) to Microsoft at Childress by August 2026, and claims NVIDIA Exemplar Cloud certification (claim 7576). These assertions merit independent verification through Microsoft infrastructure reports or NVIDIA's portfolio statements, as neither has been confirmed via third-party sources available as of publication.

The most recent capacity targets—480 MW AI cloud by end-2026 (claims 8554, 7914) and 1.2 GW by 2027 (claims 8102, 7915)—arrive with less than four months to the first deadline. Geographic expansion into South Australia (800 MW by 2028, claim 8286), Spain (490 MW, claim 7909), and Oklahoma (1.6 GW, claim 7908) was announced in August 2026, leaving 16 months for South Australia ramp. The pattern suggests IREN favors forward-looking announcements over retrospective reporting of achieved milestones.

Watch for: (1) Horizon 2-4 deployment status by year-end (claims 8558, 7571); (2) third-party confirmation of Horizon 1 Microsoft handoff; (3) published GPU allocation between on-order and deployed units; (4) whether Childress expansions resolve 2024 slippage before 2027 targets emerge.

IREN's announced capacity trajectory reveals a dramatic strategic pivot from Bitcoin mining to enterprise AI infrastructure, accompanied by escalating targets that outpace historical delivery. Through 2023-2024, the company committed to Bitcoin mining capacity—1400 MW at Sweetwater 1, expansion to 20-31 EH/s hashrate, and Childress campus buildout with multiple tranches totaling 230+ MW promised by mid-2024. Claims 6495, 6410, 6262, and 5711 marked these deadlines as delayed with no transparent revision; by May 2024 (claim 6150), only 10 EH/s materialized despite targets of 20-31 EH/s. The company did announce 46 EH/s installed by March 2026 (claim 3423), suggesting some capacity materialization, though the gap between promised 31 EH/s in Q4 2024 and stated 46 EH/s by 2026 lacks intermediate milestones.

Since July 2026, IREN has announced AI cloud expansion targets at 480 MW by end-2026 (claim 364) and 1200 MW by 2027 (claim 365), paired with simultaneous capacity claims of 150,000 GPU units (claim 8317), 5 GW secured power globally (claim 727), and new regional campuses: 800 MW South Australia, 490 MW Spain, 1.6 GW Oklahoma. The August 2026 announcements cluster thirteen claims in five days, including partnerships worth $9.7B (Microsoft), $3.4B (NVIDIA), and $2.8B with unnamed AI infrastructure companies.

Independent support exists for Horizon 1: Microsoft's acceptance by August 2026 (claims 346, 723) and delivery of 50 MW (claim 685) suggests at least partial operational delivery. However, claims of Horizons 2-4 completion by end-2026 (claims 353, 8274, 7571) remain unverified with less than four months remaining. The Dell GPU supply agreement (claim 4161) was originally promised by March 1, 2026, now listed as delayed, with a revised $1.6B agreement appearing only in May 2026 (claim 1297). Nostrum Group acquisition (claim 8106) and Mirantis purchase (claim 8279) announced without disclosed capacity contribution or integration timeline.

Watch through December 2026 for: (1) evidence of 480 MW AI cloud operational capacity; (2) completion and acceptance of Horizons 2-4 by Microsoft; (3) GPU delivery against 150,000-unit target; (4) Dell agreement resolution and equipment arrival; (5) Nostrum/Mirantis integration progress. The eight-month compressed timeline from first major AI announcement to claimed 1.2 GW target suggests either aggressive execution or optimistic forecasting that mirrors 2024 Bitcoin delays.

IREN's delivery trajectory shows a pattern of early-stage delays followed by a pivot toward AI infrastructure with claims of rapid scaling. Multiple bitcoin mining timelines slipped in 2024—the 80-megawatt Childress expansion (claim 6495) promised for mid-2024 and the 50-megawatt 2024 build-out (claim 6153) both deferred without public acknowledgment of revised dates. Similarly, GPU procurement timelines missed targets, with the H200 acquisition (claim 5711) and H100 delivery (claim 6262) both pushed into 2025. By September 2024, operating hashrate stood at 20 exahashes despite earlier ambitions to reach 31 exahashes by year-end (claim 5691, delayed).

The announced capacity has scaled dramatically since late 2025. Bitcoin mining targets climbed from 10 exahashes to 50 exahashes (claims 6341 to 4034), while GPU inventory claims jumped from 568 units to 99,900 on order within eighteen months (claims 6261 to 3422). Most notably, IREN pivoted toward AI cloud infrastructure: Horizon 1, a 50-megawatt facility, was delivered to and accepted by Microsoft (claims 685, 723), providing independent verification of at least one facility. The subsequent announcements—480 megawatts by December 2026, 1.2 gigawatts by end of 2027, plus greenfield campuses in Spain, Oklahoma, and South Australia totaling 2.89 gigawatts—represent a twentyfold expansion within two years (claims 364, 365, 7914, 7915, 7908, 7909, 8285).

The gap between committed capacity and installed hardware remains substantial. As of March 2026, installed bitcoin hashrate was 46 exahashes despite claims of 50-exahash capacity (claim 3423). The Dell GPU supply agreement (claim 4161) slipped past its March 2026 deadline. Independent verification beyond Horizon 1 acceptance is limited; the breadth of simultaneous megaproject announcements in August warrants scrutiny about sequencing and capital deployment.

Over the next ninety days, track whether Horizons 2-4 delivery occurs before year-end (claim 353), Dell and NVIDIA supply chains produce announced volumes, and whether South Australia or Oklahoma groundbreaking proceeds. Execution risk rises sharply as IREN juggles five geographies and multiple customer commitments.

IREN's claimed capacity has undergone radical revision and strategic repositioning over 2.5 years, tracking a company searching for its operating model. From late 2023 through mid-2024, the narrative centered on Bitcoin mining and modular data centers: claim 6522 promised 10 EH/s hashrate, escalating by January 2024 to 20 EH/s (claim 6342). By September 2024, the company reported 20 EH/s actually installed (claim 5689), suggesting on-track delivery. But the gap widened sharply thereafter. By late 2025, IREN claimed 50 EH/s capacity (claims 4034, 4021), yet by March 2026 disclosed only 46 EH/s actually operating (claim 3423)—a meaningful gap between announced and realized. More telling, GPU commitments exploded: from 568 NVIDIA H100s promised by June 2024 (claim 6262, marked delayed) to 2100 units by November 2025 (claim 4033) to 99,900 units by March 2026 (claim 3422). This 170-fold escalation in 14 months suggests either aggressive deal-making or repositioning away from miners. The Childress facility exemplifies timeline slippage. Claim 6410 promised an 80 MW expansion by June 2024 (marked delayed even then); this morphed into a 50 MW promise by December 2024 (claim 6153, also delayed). The facility appears repeatedly renamed or refactored—Horizon 1 emerged in August 2026 claiming 50 MW delivery to Microsoft, yet this announcement appeared within weeks of the claimed acceptance date (claims 723, 779), leaving no window for independent verification through permits, grid interconnection records, or satellite imagery. More recently, capacity targets have inflated dramatically: claim 8102 projects 1.2 GW AI cloud capacity by end of 2027, with intermediate 480 MW claimed by year-end 2026 (claims 364, 7914). Geographic expansion accelerated sharply—Spain (490 MW, claim 7917) and Oklahoma (1.6 GW, claim 7908) announced August 2026 alongside major contracts with Microsoft and NVIDIA. The Dell GPU agreement, promised complete by March 2026 (claim 4161), has already slipped past deadline. Watch: whether Horizons 2–4 actually deploy by December 2026 (claim 353); external validation of the Horizon 1 delivery and Microsoft acceptance; whether the Dell GPU agreement materializes; and whether geographic expansion announcements (Spain, Oklahoma) produce tangible progress over the next quarter.

IREN has undergone a dramatic strategic pivot from bitcoin mining to AI infrastructure, yet its execution track record raises concerns about delivery timelines. Early bitcoin expansion targets—80 MW of Childress capacity and 31 EH/s hashrate by end of 2024 (claims 6495, 5691)—failed to materialize on schedule, marked repeatedly as "delayed." The company claims 46 EH/s actual installed capacity as of March 2026 (claim 3423), but this falls short of its concurrent promise of 50 EH/s capacity (claims 4034, 4021). The gap between target and delivered hashrate suggests the early bitcoin roadmap unraveled.

Capacity targets, by contrast, have escalated substantially. IREN announced AI cloud capacity of 480 MW by end-2026 and 1.2 GW by end-2027 (claims 364, 8102), alongside GPU procurement of 99,900 units claimed in March 2026 (claim 3422). This represents a shift from distributed bitcoin mining to centralized AI cloud serving hyperscalers. The $9.7 billion Microsoft contract and $3.4 billion NVIDIA partnership (claims 778, 781) provide revenue visibility, though both were announced only in August 2026 with compressed deployment timelines.

Independent verification remains absent from the record. IREN claims Horizon 1—a 50 MW deployment at Childress—was delivered and accepted by Microsoft (claims 685, 499), but with no external confirmation, this rests entirely on company disclosure. Prior misses on H100/H200 delivery (claim 6262) and Childress expansion (claim 6153) create credibility friction. The March 2026 hashrate figure of 46 EH/s—three months of deterioration against promised targets—suggests internal resource constraints.

Over the next 90 days, watch whether 480 MW of AI cloud capacity enters service by December 31, 2026 (claims 364, 7914) and whether Horizons 2-4 deployment progresses as stated (claim 7571). Missing these targets would extend the company's delay pattern into a third year. Monitor for revised capacity guidance in Q4 earnings, joint statements from Microsoft or NVIDIA confirming deployment progress, and whether the Warriors sponsorship due October 1 (claim 7895) signals brand-building or engineering distraction. Any admission of slippage on the AI cloud targets should be treated as a leading indicator of broader infrastructure execution risk.

IREN's announced capacity reflects substantial revision and reprioritization across three distinct periods. In cryptocurrency mining, claims started at 8.3 EH/s operating hashrate in November 2023 (6493), advancing to 10 EH/s by May 2024 (6150), 20 EH/s by September 2024 (5689), and a 31 EH/s expansion target for Q4 2024 (5691). The company disclosed 46 EH/s as installed capacity as of March 2026 (3423), neither matching the Q4 2024 promise nor clarifying the gap from interim targets. GPU allocations followed similar expansion without integration: 568 H100 units in February 2024 (6261), 2,100 units by late 2025 (4033), and 99,900 units by March 2026 (3422).

Early facility timelines consistently slipped. The Childress 80MW expansion promised for June 2024 (6410) and Bitmain T21 miner shipment promised for June 2024 (6408) both registered as delayed. GPU delivery by June 2024 (6262) also missed. Starting in late 2025, IREN announced substantially larger partnerships—Microsoft ($9.7B five-year contract, 8112) and NVIDIA ($3.4B five-year contract, 1031)—shifting from cryptocurrency toward AI cloud infrastructure. Horizon 1, a 50MW facility, was reported delivered in August 2026 (685) with Microsoft acceptance noted (346), marking the first documented facility completion in this record.

Current targets call for 480 MW of AI cloud capacity by end of 2026 (364) and 1.2 GW by end of 2027 (8102). No permits, interconnection filings, or independent construction verification appear in the claims. Horizon 1's completion and acceptance support credibility for AI execution, yet the Horizons 2-4 deployment schedule (promised by 2026-12-31 per 353) remains unvalidated.

Over the next 90 days, monitor Horizons 2-4 deployment status by 2026-12-31 (353), whether 480 MW is installed (7572), closure of the Warriors jersey sponsorship deal by 2026-10-01 (7895), and any independent evidence of construction or equipment procurement at Spain (490 MW, 7909) and Oklahoma (1.6 GW, 7908) sites.

IREN's announced capacity targets have inflated dramatically while execution timelines contracted. Bitcoin mining capacity escalated from a stated 10 EH/s target (claim 6522, November 2023) to 20 EH/s (claim 5689, September 2024) to 50 EH/s (claims 4034, 4021, December 2025–March 2026) without acknowledgement of the earlier targets or explanation for the acceleration. Similarly, GPU procurement jumped from 568 H100 units (claim 6261, February 2024) to 1,896 H200 units (claim 5712, September 2024) to 99,900 total units on order (claim 3422, March 2026)—a 175-fold increase in stated capacity in under two years.

Timeline performance deteriorated through 2024. An 80 MW Childress expansion promised for June 2024 (claims 6495, 6410) was marked delayed. GPU delivery timelines (claim 6262) and miner shipments (claim 6408) both slipped past June 2024 deadlines. The Q4 2024 expansion to 31 EH/s (claim 5691) was similarly delayed. A $5.8 billion Dell GPU agreement promised by March 2026 (claim 4161) remains unfulfilled as of September 2026.

In August 2026, IREN announced a strategic pivot toward AI cloud services with three major partnership claims: a $9.7 billion five-year Microsoft contract (claims 686, 778), a $3.4 billion NVIDIA deal (claims 351, 781), and a new $5 billion partnership with NVIDIA (claim 362). These announcements introduced aggressive new capacity targets: 480 MW AI cloud capacity by end-2026 (claims 364, 500, 7914), expanding to 1.2 GW by end-2027 (claims 365, 7915), plus a claimed 1.6 GW Oklahoma campus (claims 7908, 7916) and 490 MW Spain facility (claims 7917, 7909).

However, a critical inconsistency emerged immediately: claim 7061 marks Horizon 1's approval and operational status as "delayed" on August 21, 2026, yet claims 499, 723, and 779—filed the same day—assert the facility was already "delivered and accepted" by Microsoft. No independent permit filings, interconnection queue positions, or third-party infrastructure reports have surfaced to corroborate either status or the underlying Microsoft acceptance.

Over the next 90 days, monitor: (1) actual Horizon 1 operational status and customer acceptance confirmation beyond IREN statements; (2) Dell GPU agreement finalization and delivery commencement (now overdue); (3) evidence of Oklahoma and Spain facility permitting or construction activity; (4) NVIDIA Exemplar Cloud certification validation; (5) monthly capacity utilization reports for Horizon 1 if operational.

IREN's infrastructure narrative has undergone significant reshaping in 2026, shifting from Bitcoin mining as the primary capacity story to large-scale AI cloud deployments. Through late 2023 and 2024, the company announced Bitcoin mining targets: 10 EH/s by early 2024 (claim 6341), escalating to 20 EH/s by mid-2024 (claim 5689) and 31 EH/s promised for Q4 2024 (claim 5691). By November 2025, claimed capacity jumped to 50 EH/s (claims 4034, 4021). Yet recent announcements from August 2026 prioritize Microsoft and NVIDIA partnerships and AI cloud deployments measured in megawatts—480 MW for 2026 year-end (claims 364, 7914) and 1.2 GW by end 2027 (claims 365, 7915). Bitcoin capacity claims have largely vanished from public messaging.

Execution credibility signals caution. Multiple facility timelines promised for mid-2024—including Childress 80 MW expansion (claim 6495), GPU delivery (claim 6262), and T21 miner shipping (claim 6408)—all carry delayed flags. The Dell GPU supply agreement of $5.8 billion was promised for March 2026 (claim 4161) and remains delayed. Most strikingly, Horizon 1's operational approval was promised for August 21, 2026 (claim 7061, marked delayed), the same date IREN claims the facility was delivered and accepted by Microsoft (claims 499, 681, 723, 779). This timing collision merits scrutiny: if accepted by the customer, why is operational status delayed?

The company has also expanded geographically without consistent narrative. Original focus centered on Childress and Sweetwater, Texas; recent announcements reference Oklahoma (1.6 GW, claim 7908) and Spain (490 MW, claim 7917), alongside earlier British Columbia sites with unclear status. The Mirantis acquisition (claim 7899) signals vertical integration into cloud infrastructure software, yet capacity implications remain unstated.

Critical 90-day milestones include the 480 MW AI cloud target for December 2026 (claims 364, 7914), Horizons 2–4 deployment completion also promised by year-end (claims 353, 7571), and Warriors jersey activation by October 1 (claim 7895). Watch for: official acknowledgment of the six-plus delayed timelines; third-party permits or interconnection filings for Oklahoma and Spain; actual Horizon 1 utilization versus committed Microsoft capacity; and any revised 2027 guidance in light of execution headwinds.

IREN's delivery timeline reveals systematic divergence between announced targets and demonstrated completion. Bitcoin mining capacity promises from late 2023—an initial 10 EH/s target (claim 6522) escalating to 31 EH/s by end-2024 (claim 5691)—are flagged [delayed]. The company promised Childress expansion phases at 80 MW by mid-2024 (claims 6495, 6410) and GPU deliveries on the same date (claim 6262), both [delayed]. When deadlines passed without formal revision, IREN continued announcing: by November 2025, claims of 50 EH/s installed capacity (claims 4034, 4021) displaced discussion of the earlier 31 EH/s miss. A Dell GPU supply agreement due March 2026 (claim 4161) is [delayed]. Most acutely, Horizon 1's promised operational status of August 21, 2026 (claim 7061) is [delayed]—now fifteen days overdue.

Capacity scaling accelerated in August 2026 following a strategic reorientation from bitcoin to AI cloud. IREN announced a $9.7 billion five-year Microsoft contract (claims 498, 778), a $3.4 billion NVIDIA partnership (claim 351), and claimed Horizon 1 delivery (50 MW, claims 345, 499, 723). Simultaneously, the company stated 480 MW AI capacity by year-end 2026 (claims 7914, 7906) with 1.2 GW by end-2027 (claims 365, 7915). New geographic announcements—Oklahoma's 1.6 GW campus (claim 7908), Spain's 490 MW site (claim 7909)—carry no specific timelines. The $465 million Mirantis acquisition (claim 7899) suggests internal recognition that hardware commitments require software infrastructure.

No independent evidence validates these claims: no grid interconnection filings, no permit records, no third-party audits appear in this record. The pattern—delayed hardware targets, unrevised public narratives, sudden new announcements—suggests capacity statements reflect intentions rather than committed schedules.

Watch the next 90 days for: actual delivery of 480 MW AI capacity by 2026-12-31 (claims 7914, 7906); Horizons 2-4 deployment by 2026-12-31 (claim 7571); and validation that the $9.7B Microsoft contract represents genuine capacity purchase and payment. A single missed deadline, or failure to provide independent evidence of Horizon 1 operation, would indicate August 2026 announcements materially misrepresent IREN's deployment capability.

IREN's capacity announcements reveal a dramatic strategic pivot and consistent pattern of missed infrastructure timelines. Through late 2024, the company pursued Bitcoin mining scale—targeting 10 EH/s (6522, Nov 2023), revised to 20 EH/s pathway (6342, Jan 2024), then 31 EH/s by Q4 2024 (5691, Sept 2024). The Childress 80MW facility promised for June 2024 delivery (6495, 6410) remains undelivered as of September 2026—a 27-month slip. Parallel GPU procurement exploded from 568 H100s (6261, Feb 2024) to 1,896 units (5712, Sept 2024) to 99,900 units on order (3422, Mar 2026) without corresponding timeline acceleration. GPU delivery targets for June 2024 (6262) and equipment acquisitions for December 2024 (5711) both slipped without public revision acknowledgment.

Mid-2026 brought a wholesale reorientation toward enterprise AI cloud, with Microsoft ($9.7 billion, 498, 681) and NVIDIA (3.4 billion, 351) anchoring the new strategy. Horizon 1 reportedly delivered August 2026 (499, 723, 779), providing 50MW. However, IREN simultaneously committed to 480MW AI capacity by December 2026 (364, 7906, 7914) and 1,200MW by end-2027 (7915), requiring eight-fold acceleration over current delivery pace. The Dell GPU supply agreement (4161, 5.8 billion USD) promised for March 2026 appears abandoned in favor of a smaller 1.6 billion agreement (1297, May 2026), suggesting material scope reduction. Geographic expansion announcements—Oklahoma 1.6GW, Spain 490MW (7916-7909)—carry no dates or milestones.

Independent verification remains absent. No permits, interconnection queue entries, or third-party capacity assessments appear in the record. Bitcoin mining capacity claims fell from 50 EH/s target (4034, Nov 2025) to 46 EH/s installed (3423, Mar 2026), indicating 8% shortfall against stated goals within three months. Watch for: (1) Horizons 2-4 delivery against December 2026 deadline (353, 7571)—critical for 480MW target credibility; (2) clarification on Dell partnership scope following delayed 5.8 billion agreement; (3) any public revision of 2026 capacity goals without announcement; (4) timeline materialization for Oklahoma and Spain facilities. The company's 27-month slip on Childress and redirection mid-program suggest aggressive targets outpacing execution capacity.

IREN's datacenter ambitions have undergone radical expansion and repeated deadline misses since late 2023. Capacity promises have scaled from approximately 1.4 gigawatts at Sweetwater 1 (claim 6523, Nov 2023) to 5 gigawatts of "secured power globally" by August 2026 (claim 783). The pivot from Bitcoin mining (claims 6522, 6341, targeting 10–31 EH/s through 2024) to AI cloud infrastructure accelerated sharply in 2025, with GPU procurement promises growing from 568 units (Feb 2024, claim 6261) to 99,900 units on order (Mar 2026, claim 3422) and 150,000 units claimed deployed (Aug 2022, claim 1027).

Promised delivery dates have consistently slipped without explicit revision notices. June 2024 milestones for Childress expansion (claims 6495, 6410) were marked delayed by December 2023. A May 2024 promise of 50 megawatts by year-end 2024 (claim 6153) carried delayed status immediately upon announcement. September 2024's claim of 31 exahash capacity by December 2024 (claim 5691) bore the same marker. A November 2025 Dell GPU agreement worth $5.8 billion was promised by March 2026 (claim 4161, delayed); this reappeared in May 2026 as a $1.6 billion deal (claim 1297) with no public reconciliation of the discrepancy.

Horizon 1's delivery claims illustrate this pattern. Multiple August 2026 announcements state it has been "delivered and accepted" (claims 499, 723, 779) with Microsoft approval secured (claims 346, 1029). Yet no independent evidence surfaces: no operational metrics, no interconnection queue confirmation, no third-party verification of Microsoft's 50-megawatt acceptance. The Sweetwater 1 facility, announced as operational in August 2026 (claim 1025) at 1.4 gigawatts, similarly lacks permit or utility interconnection documentation.

IREN's August 2026 announcements of 480 megawatts by December 31, 2026 (claims 364, 500) and 1,200 megawatts by end 2027 (claims 365, 7573) rest on Horizons 2–4 deployment in the next 117 days (claim 353). Watch whether these ships on schedule, whether Sweetwater 2's promised 600 megawatts materializes (claims 1026, 6805), and whether Microsoft or NVIDIA publish independent acceptance certificates or capacity metrics. Absence of interconnection queue movements or power purchase agreement filings would signal execution risk.

Insufficient data to produce trajectory analysis. All claims date to 2026-08-26 (four days prior), with no historical precedent, prior announcements, or revised timelines to evaluate. Claims 10, 12, 14 reference Mirantis acquisition; claims 11, 13 reference Nostrum Group; these appear to be duplicate entries rather than discrete announcements. Claim 18 cites 'data center geographic expansion' without specifying capacity targets, timeline, or geography. No independent evidence provided: no permit records, interconnection queue positions, construction progress, or third-party reporting to validate announced pace. To produce meaningful analysis of delivery trajectory, require: (1) historical claim sequence showing how capacity or timeline evolved across prior announcements; (2) specific megawatt or rack capacity targets and go-live dates; (3) independent corroboration (regulatory filings, interconnection status, site photography); (4) disclosure of whether these acquisitions are infrastructure-related or represent other business lines. Current dataset reflects a single announcement moment with no trajectory.