TL;DR · 30-second read
The Short Version
Two companies that grew up using warehouses of computers to earn the digital currency Bitcoin are moving on to housing artificial intelligence computers instead.
One of them, Core Scientific, plans to stop by the end of this year. Its customers have signed contracts worth more than $24 billion, over three times what the stock market says the whole company is worth.
The catch: that money arrives over many years, from a handful of customers, and much of it depends on buildings still under construction. Investors have recently grown nervous about some companies in that customer group.
Benzinga and BigGo Finance reported that Core Scientific will exit Bitcoin mining by year-end and become a pure-play AI data center provider. Trending Topics reported that IREN is also ending mining, and that the moves come as shares of neoclouds, the specialist cloud companies that rent out AI computing, sell off sharply.
Both companies filed quarterly results on November 5, 2026. Core Scientific reported third-quarter colocation revenue of $215.4 million, up nearly 60% from the prior quarter, while IREN reported AI Cloud revenue of $200 million out of $228 million in total revenue.
Executive Summary
Core Scientific’s planned exit puts an end date on its move from mining Bitcoin to high-density colocation, in which it builds and operates power-hungry data halls and leases them to customers running AI workloads. Its earnings presentation lists about 1.1 gigawatts (GW) of contracted customer power and more than $24 billion in contracted revenue, against a fully diluted market value of $6.5 billion at the November 4 close.
Mining is already a small part of the business, so the exit matters less for the revenue it removes than for the risk it concentrates. Once mining ends, Core Scientific’s revenue depends on a small group of AI tenants, and on its ability to deliver capacity for them on schedule through 2028.
That concentration is not new, but it is becoming complete at a sensitive moment. The company says more than half of its expected cash flows are backed by investment-grade credit, a rating category for borrowers judged lower risk. The remaining share, and the timing of when the stronger credit starts paying, are what readers should watch.
Mining Is Already the Small Part of the Story
Core Scientific’s results release reports adjusted EBITDA, a measure of operating earnings before items such as interest, tax and depreciation, of $133.6 million at a 57% margin. That implies total revenue of about $234 million, which would make the $215.4 million of colocation revenue roughly 92% of the quarter. Mining and mining hosting therefore appear to account for something near $19 million.
IREN shows a similar pattern with a different business model. Its Q1 FY27 release reports $200 million of AI Cloud revenue within $228 million in total, or about 88%, and its AI Cloud segment earned an 86% gross margin. In both cases, the exit gives up relatively little current revenue.
What goes away is diversification. Mining income follows the Bitcoin market and has little to do with AI tenants’ fortunes, so even a modest mining line is a revenue stream with a different risk. Core Scientific’s release also says it is repurposing its remaining mining facilities “as circumstances allow,” which is a softer formulation than a calendar date and leaves the pace of site conversion an open variable.
Why the $24B Backlog Becomes a Credit Question
The earnings presentation breaks the contracted base into about 590 megawatts (MW) leased to CoreWeave and about 530 MW tied to AMD. Of the AMD-linked total, the Pecos, Hunt and Muskogee sites, about 377 MW, are marked as AMD direct, and the Dalton and Auburn sites, about 152 MW, serve what the deck calls an AMD-backed neocloud. CoreWeave is commonly grouped with neoclouds, so on those figures about two-thirds of contracted power is destined for neocloud tenants, though some of it carries AMD’s support.
The presentation projects average annual recurring revenue of $1.8 billion and $1.6 billion of annual net operating income, or property-level profit, at a blended margin of about 90%. Dividing $24 billion by $1.8 billion suggests roughly 13 years of revenue. That is a long time to depend on any small tenant group, and it is why credit quality carries so much weight in this model. High margins on a lease protect the landlord only as long as the tenant keeps paying.
Timing sharpens the point. Chief Executive Adam Sullivan said the company is nearing full delivery under its first colocation agreement, while construction under its second is underway across five campuses. The five AMD-linked sites match that description, which suggests today’s billing comes mainly from CoreWeave and most AMD-linked income is still ahead. The company’s “over 50%” investment-grade figure is not mapped to specific leases in the presentation, but AMD-linked leases are about 47% of contracted power and carry higher margins, so they could plausibly supply much of it. If so, the stronger credit support arrives later than the mining exit. Lenders, shareholders and equipment suppliers all have exposure to that interval.
Delivery Is the Other Half of the Bet
Core Scientific was billing for 485 MW at September 30, representing more than $700 million in average annualized colocation revenue. That is under half of the contracted power and about 40% of the $1.8 billion target. Contracted revenue only becomes earned revenue when halls are delivered.
The AMD-linked schedule runs in stages: Pecos and Auburn in the first half of 2027, Dalton and Muskogee in the second half of 2027, and Hunt County in the first half of 2028, with the full 530 MW due by the end of 2028. The company’s forward-looking statements name financing and completing construction, sourcing electricity, and obtaining long-lead equipment among the factors that could change results. With mining gone, any slippage would have less other revenue around it.
Growth beyond the current contracts depends heavily on power that is not yet in hand. The presentation’s AMD reservation includes about 385 MW secured through existing grid connections, about 600 MW still in utility load studies, and about 940 MW of behind-the-meter pathways, meaning on-site generation, which is being evaluated alongside grid supply. These are pathways rather than a simple sum of future capacity. Calvert City and Grand Forks add about 170 MW of uncommitted leasable power, and more than 2 GW of new-site opportunities remain in due diligence.
Reading the Sell-Off and the $1.2 Billion Profit
Core Scientific’s headline net income of $1.2 billion needs careful reading. The company attributes it largely to a non-cash change in the fair value of its warrants, which are rights to buy shares at a set price, and says the swing reflects moves in its stock price. Gains of this kind generally arise when a share price falls, so the figure says little about operating performance. Adjusted EBITDA, which more than tripled from the prior quarter, is the better guide to the business.
A sell-off in neocloud shares does not by itself change a lease, and a lower stock price is not a missed payment. Core Scientific’s billed capacity and revenue both rose during the quarter. The transmission mechanism is indirect: neoclouds rely on outside capital to fund expansion, so a harder financing market could matter to their landlords over a contract of this length. It could also affect what Core Scientific itself pays for capital.
The two companies carry that risk differently. Core Scientific acts mainly as a landlord, while IREN’s 10-Q reports both AI cloud services and AI cloud lease revenue and lists GPU hardware among its assets. That means IREN holds more direct exposure to the chips and to customer demand for computing, whereas Core Scientific’s exposure runs chiefly through its tenants’ ability to pay.
Background
Core Scientific designs, builds and operates large data centers in Alabama, Georgia, Kentucky, North Carolina, North Dakota, Oklahoma and Texas. Most of its revenue now comes from high-density colocation serving AI workloads, with the remainder from mining digital assets for its own account and hosting mining equipment for others. Its 10-Q also records the Hunt County and Polaris DS acquisitions and senior secured notes due 2031.
Bitcoin miners became candidates for AI data center work because mining requires what AI buildouts also need: land, large electrical connections and experience running power-intensive sites. AI tenants, however, generally need more demanding cooling, networking and reliability than mining does, so conversion requires substantial construction. IREN, whose fiscal year ends June 30, has made a parallel shift, while also operating an AI cloud business of its own. Source: Core Scientific and Iren End Bitcoin Mining and Go All In on AI Data Centers as Neocloud Stocks Crash (Trending Topics), on both companies’ moves away from mining, with further coverage in Core Scientific to Exit Bitcoin Mining by Year-End and Become a Pure-Play AI Data Center Provider (Benzinga) and Core Scientific to Exit Bitcoin Mining by Year-End, Pivots to Pure-Play AI Data Centers (BigGo Finance). Primary sources: Core Scientific 10-Q filed 2026-11-05; Core Scientific 8-K filed 2026-11-05; Core Scientific Announces Third Quarter 2026 Results (Exhibit 99.1 to 8-K); Core Scientific Third Quarter 2026 Earnings Call presentation (Exhibit 99.2 to 8-K); IREN 10-Q filed 2026-11-05; IREN 8-K filed 2026-11-05; IREN Reports Q1 FY27 Results (Exhibit 99.1 to 8-K).Sources