Category: AI Infrastructure

  • Nvidia’s $105B Ohio Guarantee Backstops OpenAI’s Lease, Not AI Power Delivery

    Nvidia’s $105B Ohio Guarantee Backstops OpenAI’s Lease, Not AI Power Delivery

    TL;DR · 30-second read

    The Short Version

    Nvidia, which makes the chips behind many AI systems, has promised to cover up to $105 billion if OpenAI, the maker of ChatGPT, cannot pay for a giant computing site in Ohio. Think of a parent co-signing a very large lease.

    The first stage is planned around as much electricity as four large nuclear reactors produce. Nvidia’s promise applies only once the site is ready, so late buildings or late power are largely the developer’s risk.

    If the AI boom cools, this deal shows who is on the hook, and for what.

    Nvidia disclosed in an 8-K filed August 17, 2026 that it has entered residual value guaranties, a form of financial backstop, on leases for approximately 4.25 gigawatts of IT load at SB Energy’s PORTS-Pike Technology Campus in Pike County, Ohio. An OpenAI affiliate will be the tenant, and Nvidia’s aggregate payment obligation for this initial commitment is capped at $105 billion.

    Nvidia’s press release says it will also invest $1.5 billion in SB Energy, will be the exclusive AI compute provider at the site, and holds an option on the remaining 3.75 gigawatts of the planned 8-gigawatt campus. Capacity is expected to come online in phases beginning in 2028.

    Executive Summary

    Nvidia has agreed to stand behind OpenAI’s lease obligations at one of the largest AI campuses announced to date, using its credit to secure land, power and shell capacity, meaning the site, its electricity and the building, for its own computing systems. SB Energy will build, own and operate the data center under a 20-year lease to OpenAI.

    The $105 billion figure is a ceiling on a contingent obligation, not money spent. Nvidia pays only if OpenAI becomes insolvent and defaults or fails to make lease payments, and then generally only the shortfall between a guaranteed minimum lease value and what a replacement lease or sale recovers.

    The structure matters because of what it covers and what it does not. As described in the 8-K, the guarantee addresses tenant credit risk once premises meet ready-for-service conditions. The earlier risks, including whether buildings and power arrive on schedule, sit mainly with SB Energy, SoftBank and whoever finances the build.

    A $105 Billion Cap Is Not $105 Billion in Cash

    Reuters described the commitment as a guarantee of up to $105 billion, CNBC as backing for financing, and Yahoo Finance as an investment of up to that amount. Nvidia’s 8-K defines the number precisely: it is the cumulative cap on Nvidia’s payment obligation for its initial commitment under residual value guaranties. A residual value guaranty promises a landlord that an asset, here a lease, will hold a minimum value. If it does not, the guarantor makes up the difference.

    The cash Nvidia has said it will commit now is the separate $1.5 billion investment in SB Energy. The guarantee is contingent. After a trigger event, Nvidia may assume the lease, require the lessor to seek a new tenant, start a sale process, allow the lease to terminate, or defer those choices for up to one year while paying specified project costs. Its payment is generally the shortfall left after a replacement lease or sale, so the real exposure depends on what the campus is worth to someone else at that moment.

    Dividing the cap by the 4.25 gigawatts it covers gives about $24.7 billion per gigawatt of IT load, the power used by the computing equipment itself. That is a measure of the guarantee ceiling rather than a construction cost, since the guaranteed minimum lease values have not been published. It is still a useful yardstick for the scale of the lease obligations Nvidia is prepared to stand behind.

    Where the Guarantee Stops: Power and Construction Timing

    The most consequential line in the 8-K may be the condition on payment. An agreement generally becomes effective when the applicable lease commences, and Nvidia’s obligations are subject to specified conditions, including that the lessor has met ready-for-service conditions for the relevant premises, expected beginning in 2028. In plain terms, the facility has to be delivered before the guarantee can be called on.

    That points to a split in risk. If OpenAI cannot pay for a finished facility, Nvidia is exposed up to the cap. If the buildings or electricity arrive late, the guarantee as described does not appear to respond, and the delay falls first on the parties responsible for delivery. SB Energy will build, own and operate the data center. SB Energy and SoftBank have also committed to at least 10 gigawatts of new generation and at least $4.2 billion in regional grid infrastructure through a partnership with AEP Ohio, the local utility. Those are large obligations to carry before a guarantee of this kind becomes relevant.

    Two qualifications apply. The press release describes Nvidia’s support as covering the land, power and shell buildout, and the full agreements will not be available until their form is filed with Nvidia’s quarterly report, so details could refine this reading. Nvidia is also not indifferent to delay: its $1.5 billion stake in SB Energy and the planned deployment of its systems both depend on the campus being delivered. But that is commercial exposure, not a payment obligation under the guaranties.

    A Chipmaker’s Balance Sheet Standing In for a Credit Rating

    Among the events that end Nvidia’s obligations is OpenAI achieving a satisfactory credit rating, alongside the 20th anniversary of the lease, OpenAI’s termination of the lease in accordance with its terms, and other customary events. That clause suggests the guarantee is bridging a gap between OpenAI’s current credit standing and the level the agreements treat as sufficient. Nvidia is lending its credit to its customer’s lease, and in return it secures capacity that will exclusively host its compute.

    The arrangement gives Nvidia something concrete. OpenAI will use the 4.25 gigawatts to deploy Nvidia’s full-stack DSX platform, including GPUs, CPUs and networking, subject to limited exceptions. CEO Jensen Huang frames the campus as “long-lived infrastructure” that can be upgraded with each hardware generation, which is the core argument that the site could retain value beyond any single chip cycle.

    The question for investors is how independent the protections are. OpenAI has agreed to reimburse and indemnify Nvidia for amounts paid, but that promise would be tested in exactly the scenario in which OpenAI has failed to pay. A broad fall in AI demand could also weaken the replacement-tenant market and Nvidia’s hardware sales at the same time. This is one deal, and it does not by itself establish an industry pattern, but it does show that chipmaker credit can sit at the center of a gigawatt-scale build.

    Reading the Campus Claims

    Some figures are firmer than others. The 4.25 gigawatts, the $105 billion cap and the payment triggers are set out in a securities filing. The further capacity is conditional: Nvidia can provide credit support for roughly 3.8 gigawatts more, as the 8-K puts it, or 3.75 gigawatts in the press release, exercisable in its sole discretion. The full 8-gigawatt campus therefore depends on a later decision, even though the release names OpenAI as the customer for all of it.

    The release’s 10 gigawatts of generation for 8 gigawatts of IT load implies 1.25 gigawatts of generation per gigawatt of computing load. That leaves room for cooling and other needs, though the companies have not itemized the difference. Community claims are less specific: the $80 million benefits fund is a defined amount, made up of $40 million from SB Energy and $40 million from OpenAI, while “tens of thousands” of jobs and protection for ratepayers are stated without a published breakdown or mechanism.

    Background

    SB Energy is an AI infrastructure developer whose existing investors include SoftBank Group and OpenAI, and Nvidia is now joining them. Its PORTS-Pike Technology Campus is redeveloping the decommissioned Portsmouth Gaseous Diffusion Plant and surrounding land in Appalachian Ohio, across both private and federal property.

    AI data centers depend on more than chips. Developers must also secure land, electricity and the physical building, often shortened to “land, power and shell”, and those long-term commitments can require a tenant with strong credit. Nvidia, best known for its AI processors, is using its own credit in this deal to secure that capacity for its systems.

    Sources

    Source: Nvidia backing $105 billion in financing for OpenAI data center in Ohio, CNBC’s report on Nvidia’s support for OpenAI’s Ohio campus.

    Primary sources: Nvidia 8-K filed 2026-08-17; NVIDIA Guarantees SB Energy’s PORTS-Pike Technology Campus in Ohio to Exclusively Host NVIDIA AI Compute (Exhibit 99.1 to 8-K); Nvidia backing $105 billion in financing for OpenAI data center in Ohio – CNBC; Nvidia to provide up to $105 billion guarantee for OpenAI’s Ohio data center – Reuters; Nvidia to Invest Up to $105 Billion for OpenAI Data Center – Yahoo Finance.

  • Fermi’s 222 MW Lease Shows Why Signed Demand, Not Turbines, Sets Matador’s AI Pace

    Fermi’s 222 MW Lease Shows Why Signed Demand, Not Turbines, Sets Matador’s AI Pace

    TL;DR · 30-second read

    The Short Version

    Fermi, a young company building a huge private power plant and computing campus in the Texas Panhandle for artificial intelligence, saw its shares jump 19%.

    It just signed its first customer, in a 15-year deal worth about $6.5 billion. Yet that deal covers only about one-twentieth of the electricity Fermi hopes to be able to produce in about two and a half years.

    Fermi already has many of the giant engines that make the power. The next job is finding more customers and getting the plant built.

    Yahoo Finance reported that shares of Fermi Inc. (NASDAQ/LSE: FRMI) rose 19% as development of Project Matador accelerated. Project Matador is the company’s planned private power and AI campus on approximately 8,400 acres in the Texas Panhandle.

    In its second-quarter earnings release on August 13, 2026, Fermi said it had met all five objectives in a 90-day plan set in May. It cited a 15-year, 222 MW lease with TensorWave worth approximately $6.5 billion, a roughly 2.6 GW power alliance with Hillcore Energy Capital, three newly arrived Siemens turbines, more than $431 million in convertible notes, and the appointment of Lee McIntire as CEO.

    Executive Summary

    Fermi’s August update turned a largely prospective campus into one with a signed customer. The TensorWave agreement, which Fermi describes as a binding turnkey lease, is the most consequential item because it attaches a named counterparty and a revenue figure to a project that has been defined mainly by its land, permits and equipment.

    The same documents show how much remains conditional. The 222 MW first phase equals about 4.6% of the 4.8 GW Fermi plans within roughly 30 months, and more than half of that 4.8 GW would come from a Hillcore framework whose later power blocks start only when end-user demand is contracted.

    That is the practical reading of the 19% share move. Progress on tenants, power and financing is real, but each advanced by a different degree, and the next phase of the buildout now depends heavily on further signed leases and on construction execution.

    Why Signed Demand, Not Turbines, Now Sets the Pace

    The arithmetic starts with two numbers from Fermi’s own materials. The company reports 1.5 GW of landed power assets, meaning generating equipment that has been delivered, against 222 MW of contracted first-phase load. That is nearly seven times as much delivered equipment as signed demand. Even if TensorWave exercises both expansion options and reaches about 650 MW, it would represent about 13.5% of the 4.8 GW plan. A private grid needs spare capacity, so generation and load do not match one for one, but the gap remains wide.

    The Hillcore terms turn that gap into an operating rule. After a first block of approximately 350 MW, targeted within 24 months of notice to proceed, subsequent blocks are “triggered only by contracted end-user demand.” Fermi also says it is “closely pacing future outlays of capital with commercial commitments.” In practical terms, about 54% of the 4.8 GW plan sits with a partner whose later construction depends on customers who have yet to sign.

    This structure is a sensible way to limit the risk of building capacity nobody has contracted, and Fermi presents it that way. It also means the roughly 30-month figure is best read as a capacity target that depends on leasing pace. The people most affected are contractors sizing their work, Hillcore’s financing partners, and prospective tenants, who would be weighing how much power can be ready and when.

    The Tenant Test: A Real Contract, Concentrated in One Customer

    Of the three tests, tenancy produced the strongest evidence. Approximately $6.5 billion over 15 years averages about $433 million a year, or roughly $163 per kilowatt per month across 222 MW of total facility power, the full electrical load of the site including cooling as well as computing. That is simple division and assumes even payments. It is also revenue, not profit, and a turnkey lease implies that Fermi delivers a finished facility.

    TensorWave is what the industry calls a neocloud, a newer cloud provider focused on AI computing. Fermi’s own forward-looking statements list the ability to secure “creditworthy counterparties” as a risk factor, which makes the durability of this single customer’s payments central to the value of the contract. The two expansion options could triple the footprint, but options give the tenant a right rather than an obligation, so they should not be counted as committed demand.

    The Power Test: Landed Is Not Energized

    Fermi’s physical progress is concrete but early. It has completed excavation for all six power islands in Phase One, the areas that will house six Siemens SGT-800 turbines capable of nearly 300 MW. Phase Two’s three Siemens SGT6-5000F units, rated at up to 728 MW in simple-cycle mode, a fast-starting configuration without a heat-recovery stage, arrived at the Port of Houston in July. The campus itself is in the Panhandle, so arrival at the port is a logistics milestone rather than generating capacity.

    Some claims warrant a closer reading. The investor presentation lists “EPC agreements with Primoris and TSK” on its scorecard, while the earnings release says Fermi has engaged Primoris for balance-of-plant work and “continues to work toward a final EPC agreement” for Phase One. EPC stands for engineering, procurement and construction, the contract that usually sets a project’s scope and price. The release also describes 4.8 GW as on-site generation, while a presentation footnote includes 200 MW from the utility Xcel Energy, along with six SGT-800 turbines expected in 2028.

    The nearest test is Fermi’s target of approximately 200 MW of initial power over the next six months. Unlike a multi-year capacity goal, it can be checked soon. Permitting is further along at the near-term scale, with a 6 GW federal air permit secured and an application for another 5 GW filed, but Fermi cautions that anything beyond approximately 6 GW, including its 17 GW long-term ambition, depends on additional permits, financing and other factors.

    The Financing Test: Capital-Light Does Not Mean Risk-Free

    Fermi says it has invested more than $1.5 billion in site buildout and raised more than $431 million through 5.00% convertible senior notes, which are debt that can turn into shares. The 5% coupon implies roughly $21.5 million a year in interest. The company describes the deal as “shareholder-friendly” and says a capped call, a hedge against dilution from conversion, prevents dilution until the stock doubles to $14.64 from a $7.32 reference price. Those protections have a ceiling, and the conversion feature is part of the true cost of the capital, so a low coupon alone does not capture what the financing costs shareholders.

    The Hillcore structure moves construction capital off Fermi’s balance sheet, with Fermi “committing no capital and issuing no debt for the plant.” It does not remove Fermi’s exposure, because Fermi would serve as anchor buyer of the power under a 20-year agreement. The release says power-purchase terms will match customer leases, yet TensorWave’s first-phase lease runs 15 years, so how the two durations fit together is an important detail. Fermi’s second-quarter 10-Q also lists a Macquarie term loan and several turbine and high-voltage equipment financings, which show that the assets already acquired carry financing commitments of their own.

    Background

    Fermi Inc., operating as Fermi America and listed on Nasdaq and the London Stock Exchange, is an emerging growth company developing a private power grid for AI and advanced computing. Rather than relying mainly on a local utility’s capacity, the model is to generate most of the power on-site, with Fermi citing 220,000 MMBtu per day of firm natural gas supply and five major pipelines within 20 miles. Its risk disclosures also refer to nuclear development and licensing as part of its longer-term plans.

    The company’s pitch rests on “speed to power,” the idea that AI customers will favor sites able to deliver large amounts of electricity quickly. In May 2026, Fermi set a 90-day plan with five objectives under what it now calls “Fermi 2.0,” and its August 13 earnings release reported that all five had been met.

    Sources

    Source: Fermi Inc. (FRMI) Rockets 19% as Project Matador Development Accelerates, Yahoo Finance’s report on the company’s share move.

    Primary sources: Fermi Inc. Form 10-Q for the second quarter of 2026, filed August 14, 2026; Fermi Inc. Form 8-K current report, filed August 14, 2026; Fermi Inc. Form 8-K for its Q2 2026 earnings release, filed August 13, 2026; Exhibit 99.1: Fermi Announces Second Quarter 2026 Results and Delivers on All Five 90-day Objectives; Exhibit 99.2: Fermi Investor Presentation, Second Quarter 2026.

  • Kog’s 3,000 Tokens-per-Second GPU Demo Shows Why Model Size Is the Real Inference Test

    Kog’s 3,000 Tokens-per-Second GPU Demo Shows Why Model Size Is the Real Inference Test

    TL;DR · 30-second read

    The Short Version

    A French startup called Kog claims software alone can make artificial intelligence respond 30 times faster, using chips companies already own.

    Its demonstration produced about 3,000 small chunks of text every second. The catch: it used a deliberately small system, and customers want much bigger ones.

    This matters because these chips are scarce and use a lot of electricity, and people relying on such tools at work can wait hours for results. If the approach works on bigger systems, existing equipment could become far more useful.

    TechCrunch reported on August 14, 2026 that French startup Kog is developing software, the Kog Inference Engine (KIE), to run AI models faster on conventional data center GPUs, the graphics processors that do most AI computing. A May tech preview on AMD MI300X and Nvidia H200 GPUs reached 3,000 tokens per second for a single request, using a purpose-built model of about 2 billion parameters called Laneformer 2B, which is now open source. CEO and solo founder Gaël Delalleau said the preview produced “200 tangible business leads.”

    Delalleau expects software engineering to be the first use case, aimed at professionals frustrated by delays that can stretch to hours for Claude Code users. Kog is now focused on accelerating larger models, because prospective customers are not prepared to fine-tune small ones.

    Executive Summary

    Kog is promising “30x faster LLM inference” through software on GPUs that enterprises already operate. Inference is the stage when a trained AI model generates answers, and an LLM, or large language model, is the type of system behind tools such as chatbots and coding assistants. The appeal is direct: if software can deliver large speed gains, buyers may get more value from hardware they have already paid for.

    The evidence so far is narrower than the promise. Kog’s 3,000 tokens-per-second result came from a small model built for the purpose, while the demand Kog has encountered points to larger models. That gap is the central issue for anyone evaluating the technology.

    The story also matters beyond one startup. Cerebras, which sells purpose-built AI chips, received a warm market reception at its May IPO. Kog represents a different bet: that conventional GPUs still have substantial untapped performance, and that software can release it.

    Model Size Is Where the Speed Claim Gets Tested

    Kog’s headline figure is 3,000 tokens per second for a single request. A token is a small chunk of text, often a word or part of one, so the number describes very fast output. But it came from Laneformer 2B, a purpose-built model with about 2 billion parameters, the numerical values a model uses to turn an input into an output.

    The mechanism behind the challenge is straightforward. Generating each token requires computation across a model’s parameters, and larger models generally demand more arithmetic and more data movement through memory at every step. A result achieved on a small, purpose-built model therefore cannot simply be projected onto a much larger one. Kog has also found that its prospective customers are not prepared to fine-tune small models, a finding that moved the company’s priorities. “That’s why since the launch, we’ve been fully focused on accelerating the development of larger models to meet the demand we’ve seen,” Delalleau said.

    This affects both sides of a potential deal. Enterprises that own GPUs need to know the speed of the model they would actually deploy, not the speed of a demonstration model. For Kog, the next benchmark will say more about the “30x” promise than the first one did, because it would test the engine under the conditions buyers have asked for.

    Faster Requests Do Not Automatically Mean More Capacity

    Software efficiency is an attractive answer to scarce, power-hungry AI compute. Extracting more from installed GPUs would, in principle, ease pressure without new chips, racks or electrical capacity. That is the larger question Kog’s work raises: how far can software stretch hardware that is already deployed?

    Kog’s demonstrated claim, however, concerns single-request speed. Data center operators also care about total output across many simultaneous users. The two measures can pull apart, because serving systems often group requests together to raise overall throughput, which can slow any individual response. An engine could give one user striking speed while total tokens per GPU, or per unit of electricity, rise modestly, stay flat or even fall.

    Fast responses have real value of their own, particularly when a person or an automated workflow is waiting. But shorter waits and capacity relief are different benefits with different buyers. The case for stretching existing compute requires a second set of measurements: aggregate throughput and energy per token, at realistic numbers of concurrent users.

    A Software Bet Beside a Silicon Bet

    The contrast with Cerebras frames the competitive question. Purpose-built chips pursue speed through new hardware, which buyers must acquire and integrate. Kog’s pitch is that “extremely fast single-request decoding is possible on the standard datacenter GPUs enterprises already own.” For organizations with GPUs already installed, software could be an easier decision than a hardware change, if the performance holds on the models they need.

    Kog’s use of both AMD and Nvidia hardware in its demo is also notable. If comparable results carry over to larger models on both, buyers could gain speed without tying that improvement to a single chip supplier. That remains a conditional benefit until larger-model results are available.

    The two approaches need not be mutually exclusive, and one preview does not settle the contest. If Kog delivers at larger scale, GPU owners stand to benefit most. If speed falls sharply as models grow, the argument for specialized inference hardware becomes stronger.

    Demand Signals Are Real but Early

    There are signs that speed has commercial value. Anthropic charges a price multiple for Claude’s Fast Mode, which shows that at least one major AI provider treats speed as something customers will pay extra for. Kog also has design partners whose users generate games and apps from a prompt, and Delalleau said faster results through KIE would mean more revenue for them.

    The 200 leads show interest, and Delalleau’s description of them as “tangible” suggests more than casual curiosity. Still, a lead is an early step rather than a signed contract or recurring revenue. Kog itself recognizes that this market is not yet mature, which is a useful caution when weighing the size of the near-term opportunity.

    Background

    AI inference has become a critical bottleneck on both speed and cost, as more software depends on large models that must respond quickly. GPUs are the workhorse hardware for this computing, but they are scarce and power-hungry, so there is strong interest in getting more from them. Cerebras has taken the hardware route with purpose-built chips, and markets welcomed its IPO debut in May 2026.

    Kog drew attention the same month when its tech preview reached the front page of Hacker News, a widely read technology forum. The preview aimed to show that extremely fast single-request decoding, the process of generating a model’s output, is possible on data center GPUs that enterprises already own.

    Sources

    Source: Kog is going deeper to squeeze more inference out of GPUs, TechCrunch’s August 14, 2026 report on the French startup’s effort to speed AI inference on conventional GPUs.

  • Mistral’s 1 GW Plan Shows Why Sovereign AI Compute Needs Buyers Before Builders

    Mistral’s 1 GW Plan Shows Why Sovereign AI Compute Needs Buyers Before Builders

    TL;DR · 30-second read

    The Short Version

    Mistral, a French artificial intelligence company, says it plans to build up to one gigawatt of computing capacity by 2030. That is about the output of one large nuclear reactor, all feeding data centers that run artificial intelligence.

    The unusual part is how it wants to pay for it. It is lining up big European companies to commit to years of computer time in advance, so the buildings have customers before they go up.

    The goal is artificial intelligence in Europe that does not depend on American tech giants.

    Mistral AI said in an August 11 blog post that it is taking three steps toward what it calls AI sovereignty. First, Mistral Regional Endpoints are now generally available, letting customers choose whether inference (the work of running a trained model to answer requests) happens in Europe or the US. Second, a new Mistral Priority Tier, in public preview, offers custom rate limits backed by an uptime service-level agreement. Third, Mistral’s platform will host third-party open models, starting with Z.ai’s GLM-5.2.

    Mistral is also assembling an anchor group of enterprises whose multi-year commitments will be converted into European Compute Units (ECUs), giving access to Mistral-built infrastructure over several years. The company says it plans to build up to 1 GW of capacity by 2030. The chief executives of Amadeus, ASML, Capgemini, Caisse des Dépôts and CMA CGM are quoted in support.

    Executive Summary

    The headline products are incremental: region selection for inference and an SLA-backed service tier are features enterprise buyers already expect from large cloud providers. The strategically significant piece is the ECU program, which changes the order in which AI capacity gets built. Rather than constructing data centers and then selling time on them, Mistral is gathering multi-year enterprise commitments first and letting that pooled demand decide what is built, where, and for whom.

    That matters because the 1 GW target is large by European AI-infrastructure standards, and capacity at that scale is hard to finance without contracted revenue behind it. The announcement also broadens Mistral from a model developer into a platform that hosts other developers’ open models on its own regional infrastructure, a move that widens the pool of workloads that can fill the capacity it plans to build.

    What the announcement does not yet show is scale of commitment: no ECU volumes, prices, sites or financing structure were disclosed, and the endorsing CEOs describe support rather than signed quantities.

    Why the 1 GW Plan Starts With Buyers, Not Buildings

    AI capacity is expensive long before it earns anything. Sites, grid connections, cooling systems and accelerators (the specialised chips that run models) are paid for years before the revenue matures, and the chips lose value as newer generations arrive. Lenders typically size financing against contracted revenue, and grid operators and utilities plan more readily around connection requests backed by credible demand. A provider that builds speculatively carries the risk of idle capacity on depreciating hardware.

    European Compute Units are Mistral’s answer to that problem. The company says its anchor group’s multi-year commitments “can support infrastructure in Europe at a scale no participant could secure alone,” and that aggregating long-term demand “will help determine what capacity is built, where it is located, and whom it serves.” In effect, this is an offtake model, familiar from power projects where a plant is financed against buyers who have agreed to take its output, applied to AI compute. The commitments come first; the build follows them.

    The consequences run in several directions. Enterprises trade flexibility for assured access, making a multi-year bet on their own AI usage. For Mistral, pooled commitments are what can turn a target as large as 1 GW, roughly the output of a large nuclear reactor, into phases that can be financed rather than a speculative build. For site developers, utilities and grid operators, one aggregated buyer is easier to plan around than dozens of small ones. Mistral has not said how much has been committed so far, so how much of the 1 GW is already spoken for remains open.

    In-Region Inference, With a Stated Caveat

    Regional Endpoints let customers pin inference to Europe or the US, aligning processing with data-residency rules, sector regulation and latency needs. For European firms handling personal or regulated data, keeping processing in-region simplifies compliance conversations, and Mistral notes that most of its customers already run its models in their own data centers or cloud environments. The endpoints serve those who want Mistral-operated capacity alongside that.

    Mistral is precise about the limits: processing in the chosen region is “subject to limited, safeguarded transfers to sub-processors that may occur outside that region,” as detailed in its Trust Center. That candour is useful, and regulated buyers will want the sub-processor list and the nature of those transfers before treating residency as absolute.

    The Priority Tier adds custom rate limits and an uptime SLA for mission-critical workloads, though the uptime figure and pricing were not published. Mistral’s claim to be the only European AI lab offering both region choice and an SLA-backed tier is scoped to European labs and not independently verifiable from the announcement; large US cloud providers already offer regional, SLA-backed hosting of many models in Europe, which is the comparison many buyers will actually make.

    Third-Party Open Models Help Fill the Capacity

    Mistral’s platform will now run open-weight models from other developers, starting with Z.ai’s GLM-5.2, under the same infrastructure, regional controls and service commitments as Mistral’s own models. Open weights mean the trained model parameters are published, so a customer can inspect, adapt and host the model where it chooses.

    The commercial logic ties directly to the capacity plan. A platform that hosts several models keeps a customer’s workloads, and spend, on Mistral infrastructure even when another developer’s model suits a task better. Factory’s CEO, Matan Grinberg, frames it this way: different workloads need different models, and running them under one set of regional controls eases compliance. Utilisation is what makes data center capacity pay for itself, and a broader catalogue widens the demand that can fill 1 GW.

    There is a nuance buyers will weigh. Z.ai is a Chinese developer. Running its open weights on European infrastructure means prompts and data need not reach the model’s creator, but some public-sector and regulated buyers apply provenance policies to models themselves, not just to where they run. How much that matters will depend on each organisation’s own rules.

    Endorsements Are Not Yet Contracts

    The breadth of names quoted is notable: travel technology (Amadeus), semiconductor equipment (ASML), IT services (Capgemini), French public finance (Caisse des Dépôts) and shipping and logistics (CMA CGM). That spread suggests appetite for European AI capacity well beyond the tech sector.

    The quotes, however, describe ambition and confidence rather than quantities. None specifies a capacity figure, contract length or value, and the announcement does not say which of these organisations hold ECU commitments. ASML is also a significant investor in Mistral, which is relevant context for its endorsement. The most concrete usage claim comes from CMA CGM, which says deployment is already under way among thousands of employees. Caisse des Dépôts describes Mistral Compute as a “European neocloud,” a term for newer providers that specialise in renting AI compute.

    The test between now and 2030 is conversion: whether endorsements become signed ECU volumes large enough to anchor each successive phase of the build.

    Background

    Mistral AI is a Paris-based AI developer founded in 2023 and best known for releasing open-weight models alongside commercial ones. Its product line now spans Studio for building agents and apps, Forge for training and customising models, the Vibe agents, and an AI Cloud business, Mistral Compute, that provides infrastructure for training and inference. Most of its customers today run its models inside their own data centers and cloud environments.

    The announcement lands in a long-running European debate over dependence on US cloud providers for critical digital infrastructure. European organisations operate under data-protection rules such as GDPR and the EU AI Act, and many public bodies and regulated industries want assurance about where data is processed and who controls the underlying systems. Demand for AI capacity has made that question more pressing, because the compute behind modern AI is scarce, capital-intensive and concentrated among a few large providers.

    Sources

    Source: In-region inference, open models, and new European infrastructure for sovereign AI. Mistral AI’s announcement of Regional Endpoints, the Priority Tier, third-party open models and the European Compute Units program.

  • IBM’s $240M Together AI Deal Treats AI Inference Like Reserved Capacity

    IBM’s $240M Together AI Deal Treats AI Inference Like Reserved Capacity

    TL;DR · 30-second read

    The Short Version

    IBM, the century-old computing company, has signed a $240 million, multi-year deal with Together AI, a younger firm that runs freely available artificial intelligence programs for other businesses.

    IBM’s cloud, meaning its rentable computers in data centers, will host a large set of powerful Nvidia chips set aside for one job: answering people’s questions to those programs, the everyday work of a chatbot.

    The surprising part is how it is bought. Instead of paying per use, the computing power is booked years ahead, like leasing a whole building instead of renting hotel rooms by the night.

    IBM and Together AI have signed a $240 million multi-year agreement for an Nvidia-powered AI inference cluster that will run on IBM Cloud, Reuters reported on August 11, 2026. Inference is the stage where a trained AI model answers requests, as opposed to training, where the model is built.

    Together AI operates a cloud platform that serves open-weight AI models, meaning models whose parameters are publicly released, to developers and enterprises. Under the deal, a dedicated cluster of Nvidia accelerators on IBM’s infrastructure is committed to that inference work for several years.

    Executive Summary

    The headline number is $240 million, but the structure matters more. This is not a customer swiping a card for GPU hours. It is a multi-year commitment to a specific, dedicated cluster, which is how data center space and power have long been contracted, and increasingly how AI compute is contracted too.

    For IBM, the deal places IBM Cloud in the business of hosting third-party AI inference at scale, a market where the large hyperscalers and specialist GPU clouds have been most visible. For Together AI, it secures a block of capacity for serving open models at a time when access to Nvidia hardware still depends heavily on who has committed to it in advance.

    The companies have not disclosed the hardware generation, the size of the cluster, where it sits, or how the $240 million is paid over time. Those details determine whether this is a modest capacity reservation or a meaningful new source of inference supply.

    Inference Is Being Booked Like a Lease

    Most people meet AI inference as a metered service: a developer sends a prompt, a provider returns an answer, and the bill is calculated per token, a token being a small chunk of text. That pricing looks elastic, but the hardware underneath it is not. Nvidia accelerator clusters are expensive to buy, take time to deliver and install, and lose value as each new chip generation arrives. Someone has to commit to them before the first token is served.

    A $240 million, multi-year contract for a dedicated cluster is that commitment made explicit. It shifts inference from the on-demand model of cloud computing toward the reserved-capacity model long familiar from colocation and wholesale data center leasing, where a tenant contracts a defined block of infrastructure for a defined term. The provider gets revenue visibility that justifies deploying hardware; the tenant gets guaranteed supply it can resell to its own customers with predictable performance.

    This is one deal, and it does not prove an industry-wide shift on its own. But it shows clearly that open-model inference, often described as a commodity service racing to the lowest per-token price, is being underwritten by the same kind of long-dated capacity contracts as training clusters. The per-token price a developer sees sits on top of a fixed, multi-year obligation somewhere in the stack.

    What IBM Gains by Hosting Someone Else’s Models

    IBM has spent recent years positioning around open and hybrid AI: its Granite models are openly released, and its Red Hat business is built on open-source software. Hosting a platform whose entire proposition is serving open-weight models fits that posture more naturally than competing head-on with the largest clouds for frontier-lab training contracts.

    It also gives IBM Cloud a substantial, recurring AI workload with an anchor tenant. For a cloud provider, a committed customer reduces the risk of buying accelerators that might otherwise sit underused. For enterprise buyers who already run on IBM infrastructure, the arrangement could, in principle, place open-model inference closer to their existing systems, though neither company has described how enterprise customers would access the cluster.

    The Risk Sits in the Utilisation Curve

    Reserved capacity moves risk; it does not remove it. Inference demand can shift quickly as new models arrive, and per-token prices for open models have been falling as providers compete. A multi-year commitment to a fixed pool of hardware exposes whichever party carries it to two pressures: whether the cluster stays busy, and whether the hardware stays competitive as Nvidia releases newer, more efficient generations.

    There is also a physical side. Dense Nvidia clusters draw significant power and increasingly require liquid cooling, which constrains which data halls can host them and how quickly. The value of a contract like this depends on the cluster actually being energised and in service on schedule, and none of those milestones have been published.

    Background

    IBM has operated a public cloud for more than a decade and in recent years has concentrated its AI strategy on enterprise and hybrid deployments, including its watsonx platform, its openly released Granite models and the open-source software business of Red Hat. IBM Cloud offers Nvidia GPU infrastructure alongside its traditional enterprise services.

    Together AI emerged during the generative AI boom as a platform for running open-weight models, giving developers an alternative to closed models from the largest AI labs. Like other inference providers, it depends on securing large volumes of Nvidia accelerators, whose supply has been shaped by long-term commitments from cloud providers and AI companies.

    Sources

    Source: IBM, Together AI ink $240 million deal for Nvidia-powered AI inference cluster (Reuters): report on a multi-year agreement for a dedicated Nvidia inference cluster on IBM Cloud.

  • Riot Has Delivered 25 MW. Its 191 MW Anthropic Lease Makes AI Execution the Test

    Riot Has Delivered 25 MW. Its 191 MW Anthropic Lease Makes AI Execution the Test

    TL;DR · 30-second read

    The Short Version

    A company best known for mining bitcoin just signed a 20-year deal worth about $9.1 billion to house computers for an artificial intelligence lab, reportedly Anthropic.

    The catch: the new job is more than seven times the size of what it has finished for AI customers so far. The whole company brought in about $174 million last quarter.

    Mining sites have something AI companies want: access to huge amounts of electricity. Now the question is whether this one can build fast enough.

    Riot Platforms disclosed on August 10, 2026, that it has signed a 20-year data center lease for 191 megawatts (MW) of critical IT capacity, the power available to run computing equipment, at its Rockdale campus in central Texas. Riot expects about $9.1 billion in contract revenue over the initial term, which runs through June 2048, and about $16.1 billion if the tenant exercises two five-year extensions.

    CNBC reported that Riot struck a deal with Anthropic. Riot’s earnings release describes the tenant only as “one of the world’s leading frontier AI labs,” meaning a developer of the most advanced AI models.

    Executive Summary

    The lease is Riot’s second at Rockdale, following the Advanced Micro Devices (AMD) agreement announced on January 16, 2026. Riot’s earnings release says the two leases total 241 MW and about $9.8 billion of long-term contracted revenue. An 8-K filed the same day also says Riot has entered a non-binding letter of intent for a proposed lease at its Corsicana Facility.

    The deal matters because of the gap between what Riot has signed and what it has built. Riot reports delivering 25 MW to AMD on time and on budget, while the new lease is more than seven times that size. With a large contract now in hand, the questions move from whether a bitcoin miner can win an AI tenant to whether it can deliver at this scale, and how much protection the contract offers over two decades.

    From 25 MW to 191 MW: Why Execution Is Now the Test

    Riot’s delivery record is real but still small relative to its new obligation. The company says it completed AMD’s initial 25 MW on time and on budget. A further 10 MW is due in November 2026 and 15 MW in May 2027, which would bring AMD to 50 MW. The new 191 MW lease is more than seven times what Riot has delivered and nearly four times AMD’s full contracted capacity.

    The calendar sharpens the point. A 20-year initial term running through June 2048 implies a start around mid-2028, roughly 22 months after the announcement. That inference does not establish how the capacity will be phased. It does suggest that Riot may need to deliver far more capacity, in less time per megawatt, than it has so far.

    Tenants depend on that schedule to bring computing online, lenders and investors depend on it to start the revenue, and suppliers and contractors must be lined up to support it. Riot credits its in-house procurement, engineering and construction capabilities for the AMD record. The Rockdale lease will test whether that model holds at a much larger scale.

    What the $9.1 Billion Figure Does and Does Not Say

    Spread evenly, $9.1 billion over 20 years averages about $455 million a year, or about $2.4 million per MW annually. That is simple arithmetic rather than a payment schedule, and actual revenue may vary with phasing and contract terms. Even so, the scale is striking: Riot reported total second-quarter revenue of $174.2 million, up 14% from a year earlier, including $23.2 million from data centers, about 13% of the total.

    The larger $16.1 billion figure needs more care. Its additional $7.0 billion depends on two five-year extensions that are exercisable at the tenant’s election, so that portion is optional for the tenant rather than committed. Those ten years would average about $700 million annually, well above the initial-term average, though the reason for the difference is not explained in the company’s statement of the headline figures.

    Contract revenue is also not profit. The agreement is a “Data Center Lease and Services Agreement,” so the total may reflect more than rent, and returns will depend on the cost of building and operating the capacity.

    Tenant Credit Becomes Part of the Asset

    A long-term lease is worth only as much as the tenant’s ability to pay and the protections written into the contract. That is especially relevant here because a 20-year term is longer than today’s frontier AI labs have existed. Their spending is substantial, but a contract of this length requires lenders and investors to underwrite a young sector across decades. Credit support, termination rights and remedies for delays would all shape how much of the headline value is dependable.

    The contract book is also concentrated. By Riot’s figures, the new lease accounts for about 79% of contracted capacity (191 of 241 MW) and about 93% of contracted revenue ($9.1 billion of $9.8 billion). Having both AMD and a frontier lab broadens Riot’s customer list, but the financial outcome now rests heavily on one agreement.

    Power Wins the Meeting, but It Does Not Finish the Building

    Riot CEO Jason Les says the platform combines “multi-gigawatt-scale power capacity that is already fully approved and energized” with in-house development expertise and custom engineering. Approved, energized power is a meaningful advantage, and it helps explain why bitcoin miners have become candidates for AI development. They already operate large, power-intensive sites.

    A grid connection is not a finished data hall, however. Critical IT capacity counts only the power available to computing equipment, so a site must also supply cooling and other facility loads. AI facilities also generally demand higher reliability and more complex cooling than mining, which can tolerate interruptions more readily. Riot’s power position may reduce one major development risk, but it leaves construction and commissioning risk in place.

    The deal is one data point rather than proof that every miner can make this transition. Riot’s 10-Q now reports Data Centers as a segment alongside Bitcoin Mining and Engineering, but the Corsicana letter of intent is non-binding and should not be counted alongside signed leases. The measure of the pivot will be capacity commissioned and revenue recognized.

    Background

    Riot Platforms, based in Castle Rock, Colorado, describes itself as a digital infrastructure company specializing in large-scale data centers and bitcoin mining. It operates facilities in central Texas and Kentucky, with engineering and fabrication facilities in Denver and Houston.

    Bitcoin mining uses specialized computers and large amounts of electricity to validate transactions and earn bitcoin. Developers of advanced AI systems also need large amounts of power, which has made mining operators’ sites and grid access relevant to the AI data center market. Riot’s move into leasing began publicly with the AMD agreement in January 2026.

    Sources

    Source: Riot Platforms strikes deal with Anthropic as bitcoin miners shift focus to AI infrastructure – CNBC, on Riot’s deal and the wider shift by bitcoin miners toward AI infrastructure.

    Primary sources: Riot Platforms 10-Q filed 2026-08-10; Riot Platforms 8-K filed 2026-08-10; Exhibit 99.1 to 8-K filed 2026-08-10, Riot Platforms Reports Second Quarter 2026 Financial Results and Strategic Highlights; Exhibit 99.2 to 8-K filed 2026-08-10, Q2 2026 Earnings Deck.

  • Gartner’s $23.3B Inference Forecast Shows AI Cloud Demand Turning Always-On

    Gartner’s $23.3B Inference Forecast Shows AI Cloud Demand Turning Always-On

    TL;DR · 30-second read

    The Short Version

    Businesses are on track to nearly double what they spend renting powerful computers for artificial intelligence this year, to about $42 billion, according to research firm Gartner.

    The bigger change is what that money pays for. For the first time, more will go to running artificial intelligence tools every day — answering customers, handling tasks at work — than to building new ones.

    Building happens in bursts. Running never stops. So the giant computer warehouses behind these tools increasingly have to stay busy, and powered, around the clock.

    Gartner said in a forecast published August 10, 2026 that worldwide spending on AI-optimized infrastructure as a service (IaaS) — rented cloud computing built for artificial intelligence workloads — will grow 96.4% in 2026 to $42.3 billion, up from $21.5 billion in 2025, and reach $66.1 billion in 2027, a further 56.5% increase.

    The research firm also forecast that in 2026 spending on inference (running trained AI models in production) will reach $23.3 billion and surpass spending on training (building those models) at $19 billion. Inference is projected to account for 55% of AI-optimized IaaS spending in 2026 and 59% in 2027.

    Executive Summary

    The headline number is the near-doubling of rented AI cloud spending, but the more consequential figure is the crossover: in Gartner’s forecast, 2026 is the year inference overtakes training as the larger use of AI-optimized cloud capacity. Hardeep Singh, Sr Principal Research Analyst at Gartner, attributes the shift to fine-tuned and domain-specific models being built into customer-facing and operational systems that require “continuous, real-time execution rather than periodic training.”

    That distinction matters for anyone who builds, powers or connects AI infrastructure. Training demand arrives as large, time-bound projects; inference demand scales with everyday usage and persists. A market increasingly weighted toward inference behaves less like a series of construction jobs and more like a utility load — steady, growing with adoption, and tied to where users and applications are.

    Gartner’s numbers also put the AI boom in proportion: AI-optimized services are forecast at 14.7% of total IaaS spending in 2026, and conventional cloud infrastructure keeps growing at roughly 20% a year alongside it. Data center operators face rising demand on both fronts, not a wholesale swap of one for the other.

    Inference Overtakes Training, and Demand Stops Arriving in Bursts

    Training is the process of building an AI model by feeding it vast amounts of data; inference is the work of using that finished model to answer a question, draft a document or complete a task. Gartner forecasts inference spending of $23.3 billion in 2026 against $19 billion for training, with inference’s share of AI-optimized IaaS rising from 55% in 2026 to 59% in 2027. Singh ties the shift to models moving into production systems that require “continuous, real-time execution rather than periodic training.”

    The mechanism is the shape of the demand. A training run is a project: a large block of accelerators — the specialized chips, usually GPUs, that do AI math — is reserved, runs for weeks, and is then released. Inference consumption instead scales with use: every customer query and every step an AI agent takes draws compute. Gartner specifically flags agentic AI, software that carries out multistep tasks on its own, as amplifying compute intensity. Demand that tracks usage behaves like a utility load — it persists, rises with adoption and follows the rhythm of the businesses it serves.

    That changes the planning problem for several groups. Cloud providers get fleets that are more consistently busy, which improves the economics of expensive accelerators. Data center operators and utilities see more load that draws close to its contracted power much of the time, rather than in project-driven peaks. And because inference feeding customer-facing systems has to reach those customers quickly, well-connected sites near users and networks tend to gain value relative to remote campuses chosen purely for cheap power.

    The Percentages Are Slowing. The Dollars Are Not.

    On growth rates alone, AI-optimized IaaS looks like it is cooling: 180% in 2025, 96.4% in 2026, 56.5% in 2027. But the absolute increments rise each year — about $20.7 billion of new annual spending in 2026 and roughly $23.9 billion in 2027. For people who build capacity, the increment is what counts: under Gartner’s forecast, each year requires more new AI-ready capacity than the year before, not less.

    The forecast also keeps AI in proportion. AI-optimized services are about 9.7% of total IaaS spending in 2025, 14.7% in 2026 and 18.4% in 2027. They account for roughly a third of the market’s growth in each forecast year. The remainder — general-purpose rented infrastructure — still grows by about 22% in 2026 and about 20% in 2027, on Gartner’s figures. AI is not crowding conventional cloud out of these numbers; builders face demand for both high-density AI halls and standard cloud capacity, which have very different power and cooling requirements.

    Rented AI Capacity Is Not the Same as Built Capacity

    Gartner’s figures measure what customers pay to rent AI-optimized infrastructure, not what cloud providers spend building it or what enterprises spend on hardware for their own facilities. The forecast therefore shows strong and rising demand for rented AI capacity, but on its own it cannot show AI investment shifting away from ownership toward the cloud, because it does not measure the ownership side.

    What it does indicate is where enterprise AI is being consumed as it moves into production. Singh says the shift toward deployment is “accelerating the cloud consumption patterns.” Rental spending is the revenue that providers weigh their construction against, and a forecast in which that revenue nearly doubles and tilts toward always-on inference strengthens the case for facilities designed for sustained, high-density load — heavier power delivery and the advanced cooling dense AI racks often require. The caveat is that this is a forecast: whether chips, power and finished data center space arrive fast enough to meet it is a separate question.

    Background

    Gartner, Inc. (NYSE: IT) is a Stamford, Connecticut-based research and advisory firm whose market spending forecasts are widely used by technology buyers, vendors and investors. Infrastructure as a service is the part of cloud computing in which customers rent raw computing, storage and networking from providers such as Amazon Web Services, Microsoft Azure and Google Cloud, paying for what they use instead of running their own hardware.

    The generative AI boom created a distinct segment within that market: cloud capacity built around accelerator chips and the high-speed networking needed to train and run large language models. Gartner’s figures show that segment grew 180% in 2025 to about $21.5 billion, and the firm now expects spending to shift from building models toward running them in everyday business applications.

    Sources

    Source: Gartner Forecasts Worldwide AI-Optimized IaaS Spending to Grow 96% in 2026 – Gartner — Gartner’s August 10, 2026 forecast of AI-optimized cloud infrastructure spending, including its inference-versus-training split.

  • BCE’s $1.3B AI Capex Jump Shows Sovereign AI Is Paid For With Telecom Cash Flow

    BCE’s $1.3B AI Capex Jump Shows Sovereign AI Is Paid For With Telecom Cash Flow

    TL;DR · 30-second read

    The Short Version

    Bell, one of Canada’s biggest phone and internet companies, is spending an extra $1.3 billion this year, mostly to build a giant computer center for artificial intelligence in Saskatchewan.

    The site is planned at 300 megawatts, roughly the electricity a small city uses. The idea is that Canadian businesses and governments can run artificial intelligence on computers kept inside Canada.

    The catch is that Bell pays from its own pocket. In the three months to June, more cash came in, but less was left over after building costs.

    Pulse 2.0 reported on August 9, 2026 that BCE, the parent of Bell Canada, expects its 2026 capital expenditures to rise by $1.3 billion from 2025, primarily because of construction of Bell AI Fabric’s 300 MW data center in Saskatchewan. BCE expects 2026 capital intensity of approximately 20% and free cash flow of $2.1 billion to $2.3 billion.

    In the second quarter, capital expenditures rose 41.5% year over year to $1.08 billion from $763 million, including $163 million of U.S. investment in Ziply Fiber’s fiber-to-the-premises expansion. Free cash flow fell 9.5% to $1.042 billion even as cash flow from operating activities rose 11% to $2.162 billion. Revenue rose 1.5% to $6.176 billion and adjusted EBITDA rose 1% to $2.702 billion.

    Executive Summary

    BCE is turning a slow-growing telecom business into the funding base for one of Canada’s larger AI infrastructure projects. The headline figure is a $1.3 billion year-over-year increase in 2026 capital spending, directed mainly at a single 300 MW data center campus in Saskatchewan under the Bell AI Fabric banner, alongside continued work on an expansion in Merritt, British Columbia.

    The second-quarter numbers show how that bet is financed. Operating cash flow grew 11%, but capital spending grew 41.5%, so free cash flow, the cash left after investment, fell 9.5%. Revenue and adjusted EBITDA grew only 1.5% and 1%. In other words, the AI build is being carried by an established cash engine whose earnings are growing far more slowly than its investment program.

    BCE frames the spending as part of a sovereign AI strategy, meaning compute capacity located and controlled in Canada, and points to 29% combined growth at its Ateko and Bell Cyber enterprise units as evidence of demand for AI-powered services. Whether the 300 MW campus earns its keep will depend on details BCE has not yet laid out, chiefly who has committed to use it and when.

    The Bill Arrives in Free Cash Flow

    The clearest way to see how Bell AI Fabric is being funded is to compare two lines in BCE’s second quarter. Cash flow from operating activities, the cash the existing business generates, rose 11% to $2.162 billion. Capital expenditures, money spent building and upgrading physical assets, rose 41.5% to $1.08 billion, an increase of about $317 million from $763 million a year earlier. The result was a 9.5% decline in free cash flow to $1.042 billion. The operating business improved; the build program absorbed the improvement and then some.

    The full-year guidance extends that pattern. BCE expects 2026 capital spending to be $1.3 billion higher than in 2025 and capital intensity, capital spending as a share of revenue, to reach approximately 20%. Second-quarter capex of $1.08 billion against $6.176 billion of revenue works out to about 17.5%, which suggests spending is expected to run heavier in the back half of the year as Saskatchewan construction advances. BCE’s free cash flow guidance of $2.1 billion to $2.3 billion is the company’s own acknowledgment that the AI investment weighs on near-term cash generation.

    This matters because it defines who is carrying the risk. BCE has not described project-level financing, partner equity or a joint venture for the Saskatchewan campus, so on the disclosed figures the build is funded through the corporate balance sheet and paid for, in effect, by Bell’s telecom customers’ monthly bills. With revenue up 1.5% and adjusted EBITDA up 1%, the funding source is stable but not growing quickly. That places the burden of proof on Bell AI Fabric itself: the capacity has to attract paying tenants fast enough to restore free cash flow growth, which is exactly the outcome BCE’s CEO says the company is focused on.

    Why a Telecom Is Building 300 MW of AI Capacity

    Sovereign AI, meaning AI computing that sits within a country’s borders and under domestic control, has become a policy and procurement priority for governments and regulated industries that do not want sensitive data or models processed abroad. A national carrier has natural advantages in that market: an existing fiber backbone to connect data centers to customers, long-standing relationships with government and enterprise buyers, and a brand that already signals domestic ownership. BCE is explicitly positioning Bell AI Fabric in that role.

    The scale is notable. Data centers built for AI are sized in megawatts because dense graphics-processor servers draw and shed enormous amounts of power; 300 MW is a large campus by any standard, not an incremental add to an existing colocation footprint. Siting it in Saskatchewan, alongside the Merritt, B.C. expansion, puts the capacity away from the country’s most congested metropolitan markets, where power and land are harder to secure, though BCE has not detailed the power arrangements behind the site.

    The demand signal BCE cites is its enterprise technology business: combined Ateko and Bell Cyber revenue rose 29% year over year. That is strong growth, but BCE did not disclose the dollar base, so it is not possible to gauge how much of a 300 MW facility that business could plausibly fill. The growth rate supports the direction of the strategy more than it supports the size of this particular bet.

    Two Build Programs Competing for One Cash Engine

    Bell AI Fabric is not the only capital-hungry project on BCE’s books. Second-quarter spending included $163 million of U.S. investment supporting Ziply Fiber’s fiber-to-the-premises expansion, which runs fiber directly to homes and businesses. BCE is therefore financing a Canadian AI data center build and a U.S. residential fiber build from the same slow-growing earnings base at the same time.

    The two programs have very different economics. Fiber-to-the-premises is a familiar telecom model with long asset lives and subscriber-driven returns. AI data centers depend on a smaller number of large customers, rapid hardware cycles and power availability. Running both concurrently raises capital intensity and leaves less room to absorb delays or cost overruns in either. For investors, the practical question is sequencing: which program yields returns first, and how much flexibility BCE retains to slow one if the other disappoints.

    For the wider market, BCE’s approach is a useful test case. If a national carrier can fill a 300 MW sovereign AI campus with domestic demand, other telecoms with similar assets have a template. If lease-up is slow, the cost shows up plainly in free cash flow, as it already has in the second quarter.

    Background

    BCE is the parent company of Bell Canada, one of the country’s largest communications providers, with wireless, internet, media and enterprise businesses. It also owns Ziply Fiber, a U.S. fiber provider, whose fiber-to-the-premises buildout is part of BCE’s capital program.

    BCE introduced Bell AI Fabric in 2025 as a network of Canadian data centers designed for AI workloads, positioned around sovereign AI: keeping compute capacity and sensitive data within Canada. Alongside it, BCE’s enterprise technology and cybersecurity units, Ateko and Bell Cyber, sell AI-powered services to business and public-sector customers.

    Sources

    Source: BCE: $1.3 Billion AI Data Center Spending Increase Supports 300 MW Bell AI Fabric Expansion (Pulse 2.0), on BCE’s higher 2026 capital spending plan and second-quarter results.

  • Anthropic Joins 5 Custom-Chip Builders: Nvidia’s Exposure Is Inference, Not Training

    Anthropic Joins 5 Custom-Chip Builders: Nvidia’s Exposure Is Inference, Not Training

    TL;DR · 30-second read

    The Short Version

    Anthropic, the company behind the Claude chatbot, is starting to design its own computer chips for running its AI. Samsung is reportedly in talks to build them.

    Today most AI runs on chips from Nvidia, which are expensive and use a lot of electricity. Google, Amazon, Meta, Microsoft and OpenAI already make their own chips.

    What most people miss is that these custom chips are built to answer people’s questions, not to create new AI models. Creating new models still mostly happens on Nvidia hardware. Nvidia’s rivals are winning ground on the everyday work of answering requests.

    Anthropic is building an in-house chip development team to co-design custom ASICs (application-specific integrated circuits, chips built for one kind of job) to run its Claude models. The company told Business Insider it is co-designing the chips so that Claude can run faster and more efficiently at the scale its customers need. It is hiring engineers now, and the job listing stresses delivering the design on schedule. Tom’s Hardware reported the development on August 7, 2026. It added that The Information had reported in July that Anthropic was in talks with Samsung to manufacture the chips.

    Anthropic has not named its design partner. No chip specifications, production volumes or deployment dates have been disclosed.

    Executive Summary

    Anthropic is the latest frontier AI developer to pursue its own silicon. Google, Amazon, Meta, Microsoft and OpenAI already have custom chip programs. The chips are aimed at inference, which is the work of running a trained model to answer requests. They are not aimed at training, the far more compute-intensive process of building the model in the first place.

    That distinction matters more than the headline rivalry with Nvidia. Training advanced models still depends heavily on Nvidia GPUs. Inference can run on a much wider range of hardware, and its cost recurs with every prompt a customer sends. For a company whose usage is growing quickly, especially from agentic AI that consumes many more tokens per task than a single chat exchange, inference is the line item that scales. It is therefore the one worth engineering down.

    The reported Samsung link adds a supply-chain dimension. Samsung is a small player in custom chip co-design, but it brings manufacturing capacity and memory. Memory is one of the scarcest inputs in AI hardware right now.

    Why Nvidia’s Exposure Sits in Inference, Not Training

    Every custom AI chip program disclosed so far aims first at running models, not building them. The economics explain why. Training a frontier model is an episodic, enormous job, and Nvidia’s GPUs and software ecosystem remain the default for it. Inference is continuous. Each question a user asks, and each step an AI agent takes, consumes compute. As usage grows, inference spending grows with it. Agentic workloads multiply the token count per task, so the growth is faster still.

    Inference also tolerates specialization. A chip that only needs to execute a known family of models efficiently can drop much of a general-purpose GPU’s flexibility. In exchange it gains lower power draw and lower cost per request. Estimates put custom silicon’s total cost of ownership advantage at up to 65% compared with Nvidia GPUs. That figure is a ceiling across the industry, not a disclosed number for Anthropic’s chip, which has no published specifications. Anthropic’s own stated goal is narrower and consistent with this reasoning: making Claude run faster and more efficiently at customer scale.

    For Nvidia, this means the pressure is concentrated in inference deployments. Each custom chip installed for inference displaces a GPU that would otherwise have done that work. Training, where Nvidia’s position is strongest, is not what these programs target. The large AI labs are also unlikely to become fully self-sufficient. Demand is growing too fast, and Nvidia retains strong control over access to the supply chain. The realistic outcome is mixed fleets, with GPUs for training and flexible workloads and in-house chips for high-volume serving.

    The Real Cost Is the Software, and the Years

    Designing a chip is only the visible part of the effort. A custom accelerator needs a full software stack, including compilers, runtimes and kernels, before it can run production models. The models themselves need tuning to exploit the hardware. That burden is why most companies sensibly buy general-purpose GPUs. Custom silicon only pays back at very large scale, where small per-request savings multiply across billions of requests.

    The reference point is sobering. Google has built its Tensor Processing Units for 12 years, with Broadcom as co-design partner on each generation. Amazon runs parallel Trainium and Inferentia lines. Meta has announced several MTIA designs for deployment through 2027. Anthropic is at the hiring stage, and its first chip will arrive into a market where rivals are several generations in. The job listing’s emphasis on schedule suggests the company understands this, but no delivery target has been disclosed.

    Samsung, Memory and a Concentrated Co-Design Market

    Custom AI chips are rarely designed alone. Broadcom and Marvell account for roughly 95% of the ASIC co-design market. Broadcom works with Google, OpenAI and Meta, as well as ByteDance and Fujitsu. It claims a $73 billion order backlog and expects more than $100 billion in annual AI chip revenue by the end of 2027. Marvell supports Amazon’s Trainium and Microsoft’s Maia and is expected to earn upward of $11 billion from co-design work in 2026. Whichever partner Anthropic has chosen for design, that partner is serving a crowded order book.

    Samsung’s reported role is manufacturing, and its appeal is not mainly design share. Samsung would bring fabrication capacity and chip design expertise. It would also bring access to memory, which is in short global supply. Modern AI accelerators depend on high-bandwidth memory (HBM) stacked alongside the processor. That memory, along with advanced packaging such as TSMC’s CoWoS, has been a practical limit on how many AI chips the industry can ship. A manufacturing partner that also makes memory could ease one of the tightest constraints a new chip program faces. Neither company has confirmed such an arrangement.

    What Changes for Data Center Operators

    For the facilities that host this hardware, custom inference chips add variety to planning. GPUs are power-hungry, and inference ASICs are built specifically to cut power per request. A fleet that mixes Nvidia GPUs with several in-house accelerators could bring different rack densities, cooling needs and networking layouts to the same campus. Operators serving the large AI labs should expect less standardization over the next few hardware cycles, not more.

    Background

    Anthropic develops the Claude family of AI models and sells access to consumers, businesses and government customers. It has grown quickly over the past year, and agentic AI, where models carry out multi-step tasks, has sharply increased the volume of computation each customer uses. Like other frontier labs, Anthropic runs its models on hardware from several suppliers, with Nvidia GPUs the industry’s default.

    Custom AI silicon has become standard practice among the largest AI companies and cloud providers, often called hyperscalers because of the scale of their data center fleets. The market for designing these chips is dominated by Broadcom and Marvell. Much of the manufacturing and advanced packaging runs through TSMC, and memory supply has become a key constraint.

    Sources

    Source: Anthropic co-designing custom AI inference chips to bypass costly Nvidia GPUs — Samsung reported as manufacturing partner for Claude maker (Tom’s Hardware): Tom’s Hardware’s report on Anthropic’s in-house inference chip team and its reported manufacturing talks with Samsung.

  • AMD’s Buy of 2023-Founded Taalas Shows AI Inference Is Getting Its Own Silicon

    AMD’s Buy of 2023-Founded Taalas Shows AI Inference Is Getting Its Own Silicon

    TL;DR · 30-second read

    The Short Version

    Chipmaker AMD is buying Taalas, a Toronto startup founded only in 2023. Taalas designs chips shaped around the artificial intelligence model they will run, instead of chips built to run anything.

    Every time someone asks a chatbot a question, a computer in a data center does the work, and that happens constantly. Flexible, do-everything chips waste time and electricity on that job. AMD is betting that purpose-built chips can answer faster and use less power.

    The price and the first products have not been announced.

    AMD announced on August 6, 2026 that it has reached a definitive agreement to acquire Taalas, a Toronto-based company founded in 2023 that develops specialized silicon for AI inference, the work of running a trained model to answer requests. AMD said it plans to integrate Taalas’ technology into its accelerator roadmap and build system-level products that pair it with AMD Instinct GPUs.

    Financial terms were not disclosed. The deal is subject to customary closing conditions and regulatory approvals, and AMD gave no expected closing date.

    Executive Summary

    AMD, one of two major suppliers of general-purpose GPUs for AI data centers, is buying a company whose stated purpose is to get around the limits of general-purpose hardware. AMD’s own announcement describes Taalas’ technology as reducing “compute and memory bottlenecks associated with general-purpose architectures.” Taalas CEO and co-founder Ljubisa Bajic described the company’s approach as “building the hardware around the model.”

    The significance is directional. AMD is not abandoning GPUs. It frames Taalas as a complement to Instinct accelerators, EPYC CPUs, ROCm software and Helios rack-scale systems. But a leading GPU vendor is treating inference as a workload that justifies dedicated silicon on its own roadmap. That matters for anyone planning data center capacity, power budgets or AI hardware purchases over the next several years.

    What the announcement does not yet provide is proof. There is no price, no performance figure behind the “breakthrough” claim, no named customer and no product timeline. The deal is best read as a strategic commitment whose payoff remains to be demonstrated.

    A GPU Maker Buying Non-GPU Silicon Is the Signal

    AMD’s flagship AI product, the Instinct GPU, is a general-purpose accelerator. It can train models, run them, and handle scientific computing. So it is notable that AMD’s release justifies this acquisition by pointing to the bottlenecks of general-purpose architectures. AMD says it will integrate Taalas’ technology into its accelerator roadmap and pair it with Instinct GPUs in system-level solutions. In other words, AMD is adding inference-specific hardware alongside the GPU rather than asking the GPU to do everything.

    The mechanism behind that bet comes from how the two main AI workloads differ. Training teaches a model and happens periodically, at enormous scale, on hardware that benefits from flexibility. Inference runs the finished model every time a user sends a request. It is repetitive: the same model performs the same operations, over and over, at high volume. Inference is often limited less by raw arithmetic than by moving the model’s data between memory and the compute units, the “memory bottleneck” AMD’s release refers to. Taalas says it optimizes inference dataflows, the path data takes through the chip, for the model being served. That is what Bajic means by building the hardware around the model, an approach sometimes described as hard-wiring a model into silicon.

    If the approach delivers, the people most affected are those who pay for inference at scale: cloud providers, AI service companies and the data center operators who host them. Inference runs continuously, so efficiency per answer compounds into power, cooling and floor-space requirements. More answers per unit of electricity would let operators serve more demand within a fixed grid connection, and grid connections are increasingly the constraint on new AI capacity.

    The Price of Specialization Is Flexibility

    Hardware tuned to a model is efficient for as long as that model is in use. AI models are revised and replaced quickly, while chips take years to design, validate and manufacture. A chip fitted too tightly to today’s model risks being stranded when customers move on. This likely explains why AMD positions Taalas as a complement rather than a replacement. Boppana’s framing, the “right compute solutions for every AI workload,” implies a division of labor. GPUs would handle training and fast-changing models, and specialized silicon would take stable, high-volume inference.

    For infrastructure buyers, that points toward mixed fleets: racks combining different accelerator types, each doing the job it is best at. Mixed fleets complicate capacity planning, procurement and software. Workloads have to be routed to the right hardware, and utilization has to stay high across all of it. How Taalas’ technology is exposed through ROCm, AMD’s software platform for programming its accelerators, will largely decide whether customers can adopt it without rewriting their deployment stacks.

    What AMD Is Actually Buying

    Taalas was founded in 2023, making it roughly three years old at signing. The release names no Taalas product, customer or performance figure. Instead, it leans heavily on the “world-class engineering team” and on retaining and growing Canadian talent. That suggests AMD is acquiring a design approach and the people behind it as much as any finished product line. Acquiring engineering teams to accelerate a chip roadmap is a well-established move in semiconductors, and it is a reasonable way to enter a design space quickly.

    Read even-handedly, the claims in the release fall into two groups. Some are substantiated by the fact of the deal itself: AMD has committed to acquiring the company and has stated where the technology will go. Others remain assertions until products ship, including “breakthrough inference performance and efficiency” and “significantly reducing” bottlenecks. The competitive logic is clear enough. AMD is building a full-stack AI platform of CPUs, GPUs, networking, software and rack-scale systems, and inference-specific silicon would fill a gap in that stack. Whether it closes that gap depends on execution the market cannot yet evaluate.

    Background

    AMD, based in Santa Clara, California, sells EPYC server CPUs and Instinct GPU accelerators for AI and high-performance computing. It also sells ROCm, its software platform for programming those accelerators, and Helios, a rack-scale system that packages them for data centers. The company has used acquisitions to broaden its data center portfolio before, most notably its purchase of programmable-chip maker Xilinx. It positions itself as a full-stack alternative in AI infrastructure.

    Taalas was founded in 2023 in Toronto to design silicon specifically for AI inference. Inference is the stage where trained models serve real users, and AMD calls it one of the fastest-growing segments of the AI market. As inference volumes rise, the cost and power consumed per answer have become central concerns for cloud providers and data center operators. That has drawn interest in specialized chips designed to do that one job more efficiently than general-purpose hardware.

    Sources

    Source: AMD Acquires Taalas to Advance Compute Solutions for Rapidly Growing AI Inference Market. AMD’s announcement of its definitive agreement to acquire Toronto-based AI inference silicon developer Taalas.