AI compute deals. When will the power arrive?

CoreWeave says most of the planned 22.5 MW of IT capacity at atNorth DEN01 was online on October 27, 2026. The exact online quantity and initial service date are not disclosed. This is a named IT-capacity project, not a utility measurement of energized power. CoreWeave's 1.5 GW active-power and 4.2 GW contracted-power figures are company-wide snapshots from June and August, respectively, and are not added to this site record.

Hardware: NVIDIA Blackwell systems

Meta launched approximately US$35B in public offerings: US$22.5B of Class A common stock under forward sale agreements and US$12.5B of mandatory convertible preferred shares. Meta also plans a separate at-the-market Class C offering of up to US$60B after the class begins trading; the program is not yet established. Meta says the offerings support continued investment in AI infrastructure, but proceeds are for general corporate purposes. Nvidia, AMD, and Broadcom expressed non-binding interest in buying up to US$5B of Class A shares; this is not a committed purchase and is not added. These financing amounts are not project capex, a compute agreement, or MW capacity, and are excluded from financial totals.

SpaceX entered a binding commitment letter on October 24, 2026 for a master finance lease with an Apollo-managed entity. The lessor would acquire and lease GPU pods and related equipment to SpaceX. SpaceX expects approximately US$40B in aggregate principal payments over the lease schedules. The master lease remains subject to closing conditions and definitive documentation; SpaceX expects closing before year-end 2026. No GPU count, equipment price, MW quantity, named data-center site, or energization evidence is disclosed. Principal payments are financing obligations, not project capex or compute-contract value, and are excluded from financial totals.

Hardware: GPU pods and related equipment; manufacturer and quantity not stated in the filing

Crusoe and Araico announced a strategic partnership on October 22, 2026, for AI Cloud infrastructure adjacent to the Itaipu substation. The first deployment is planned for mid-2027. Crusoe describes a path to 106 MW of IT capacity and long-term options to expand to approximately 280 MW total, but it does not state when the 106 MW scale-up will be reached. The 106 MW is not initial capacity; the optional 280 MW is not additive and its basis is not specified. No investment amount or compute contract value is disclosed. This is not Araico's separately announced Iguazu AI Campus, and its campus estimates are not assigned here.

Polar and Crusoe report DRA01 as live, with 12 MW of IT capacity powering Crusoe Cloud. The stated capacity is IT load, not campus or utility power. The October 21 report date does not establish the energization date.

Crusoe and Polar plan DRA02 as a separate phase with an additional 40 MW of critical liquid-cooled IT capacity, expected to go live in late 2027. This is a target, not delivery evidence. It is separate from the 12 MW live at DRA01.

Nebius announced an additional $10 million in compute grants for its AI Builder Program across Europe. The announcement also describes public GPU spot pricing; pricing availability is not a signed customer agreement. Grant value is a program commitment, not a customer contract or infrastructure investment. Recipient allocations and MW are not disclosed.

Positron says General Compute will purchase Atlas systems and has committed to purchasing next-generation Titan systems. Positron states that Atlas will expand customer availability beginning in Q1 2027. System quantities, purchase value, site, MW, and the Titan delivery schedule are undisclosed. Positron's separate $875 million Series C is not this agreement's value or site investment. The Q1 target is customer availability, not dated delivery or proof of energized power.

Hardware: Positron Atlas inference systems; commitment to next-generation Titan systems

Aolani announced non-binding letters of intent for up to 500 MW of data-center capacity across unnamed Malaysian sites, with capacity targeted to start coming online in the second half of 2027. Definitive agreements remain subject to negotiation and execution. Aolani's separate approximately 80 MW already committed in 2026 and its projected 580 MW 2027 total are portfolio figures and are not added to this project record.

Hardware: AI data centers and GPU deployments; quantities not stated

doubleAI says it is building a 1,024-B300 cluster in 128 DGX servers with Bynet Data Communications, MedOne, and NVIDIA. The company describes the system as totaling 2 MW of power but does not define the measurement as IT load or facility power. The company post showed a relative publication age, so the exact post date is not established; the record date is the dated October 19 Wall Street Journal report that identified the chip purchase. The source does not name a site or give a delivery, service, or energization target. Keep the 2 MW out of capacity analytics and do not infer operational capacity. The reported $220 million funding and chip purchase cost are not separated or quantified and are not entered as a contract value or infrastructure investment.

Hardware: 1,024 NVIDIA B300 GPUs in 128 DGX servers

NiSource announced a long-term electricity agreement for Microsoft's previously announced northwest Indiana campus. Microsoft is expected to begin taking service in 2028 under a 20-year term. NiSource says its GenCo-owned shared resource pool will add approximately 1,100 MW of generation resources, and Microsoft will fund the generation and transmission required for its load. The release does not state Microsoft's load in MW, identify the campus site, or report delivered power. The 1,100 MW describes generation resources in a shared pool, not data-center capacity; it is excluded from compute, IT-load, campus, and energized-power totals. NiSource separately disclosed a 600 MW battery contract for shared pool resources. The release does not establish whether that 600 MW is included in or additional to the approximately 1,100 MW portfolio increase, so the figures are not added together. The 2028 service target is not evidence that the campus or all generation assets will be energized by then.

Moore Threads reported that phase one of the Wuxi (Huishan) domestic intelligent-computing center was operating at full load on October 16, 2026. The source identifies Moore Threads full-stack systems and MTT S5000 accelerators, but gives no MW quantity, facility-power basis, total GPU count, or first energization date. The operating statement is delivery evidence; it does not establish metered energized MW.

Hardware: Moore Threads full-stack compute; MTT S5000 GPUs identified in company report

Oracle and Yucca Growth Infrastructure proposed a refinement to Project Jupiter's Bloom Energy fuel-cell microgrid on October 15. Oracle said planned data-center IT capacity remained unchanged. Its October 22 power-plan article states up to 2.45 GW of installed fuel-cell generation capacity. The plan remains subject to state permit review. The 2.45 GW is generation equipment rating, not customer IT load or measured energized power; no commissioning date or delivered MW is stated. Do not add this figure to the separate 4.5 GW compute-capacity amount in the OpenAI/Oracle Stargate record.

Crusoe pledged $50 million in high-performance AI compute credits over three years to the Genesis Mission Consortium. Credits will support workloads and technical assistance for researchers across more than 15 federal agencies. This is an in-kind compute pledge, not a cash contract, infrastructure investment, or disclosed GPU/MW allocation.

Applied Digital says an agreement provides access to up to 1 GW of potential power capacity and that it plans an AI Factory campus of that scale. Initial power availability is anticipated beginning in 2028. The company is marketing the site and discussing potential customers; this is not a customer lease, contracted IT load, or delivered capacity. The target is a company forecast, not energization evidence.

CoreWeave plans 240 MW of data-center capacity across three 80 MW buildings at AdaniConneX's Taloja campus. CoreWeave is the sole tenant and plans to deploy Vera Rubin systems. The source does not define the MW basis. The first phase is expected to be available in mid-2028; this is a planned availability target, not delivery or energization evidence. A further 240 MW is an expansion opportunity and is not included.

Hardware: NVIDIA Vera Rubin platform planned

The 20-year agreement covers 890 MW of new nuclear capacity from uprates at 11 Constellation units, with the first uprate expected in 2028. A separate 15-year agreement supports 2,700 MW from existing generation on the PJM grid. Neither amount is a Google data-center allocation or evidence of energized data-center power. Constellation's stated $4.3B investment is not Google's contract value and is excluded.

atNorth announced a new 28.6-hectare site. The first 75 MW phase has secured power; the wider development has a path to power of up to 230 MW. The source does not give a first-capacity date. The approximately €2B site investment excludes customer servers and installation; the separate estimate of up to €10B for customer equipment is not included. No USD investment total is inferred.

Qai will deploy GPU infrastructure in one of MEEZA's Qatar data centres. MEEZA will provide hosting, high-density power, and managed services. The parties did not disclose GPU count, MW, contract value, or a date for first service. This is separate from MEEZA's 4 MW hyperscaler expansion; no site link was stated.

Hardware: Advanced GPU infrastructure; generation and quantity undisclosed

RHAELM and JERA are developing the Chiba project; Dell contributes a standardized AI infrastructure model. The 400 MW figure is power capacity supporting the project, with no critical IT allocation or energized quantity disclosed. More than US$15B covers land, power infrastructure, facility construction and AI compute across all phases; it is project capital, not contract value. Apollo is a planned financing partner. The separate national framework and possible 2030s sites are not additional projects.

Firmus and Meta announced contracted GPU capacity and additional expansion options at Southeast Asian sites. The announcement does not give site names, MW, GPU count, value, or first-capacity timing. Firmus also says Meta already uses its Melbourne AI factory; the source does not tie that existing Australian use to the Southeast Asian agreements or assign it capacity.

General Compute will finance and buy Cerebras systems to offer dedicated inference, initially for agentic coding. Agreement value, site, system count and MW are undisclosed. The separate $400M debt facility is not the purchase price. Q1 2027 is a service target, not dated delivery or energization evidence.

Hardware: Cerebras wafer-scale inference systems

QumulusAI reports a three-year take-or-pay agreement worth US$211.9 million with a returning AI inference platform customer. It says the deal follows that customer's separate US$71.9 million agreement and targets capacity turnover in early 2027. The source combines GPU counts across this and another contract and does not disclose this customer's count. Two optional one-year extensions worth US$70.6 million each are excluded. The associated Atlanta site's up-to 3.75 MW is data-center capacity, not a disclosed customer IT allocation, and its MW basis is unknown.

Hardware: NVIDIA Blackwell B300 GPUs; exact customer allocation undisclosed

QumulusAI reports a three-year agreement worth US$29.1 million with a returning generative-AI inference platform customer. It says the deal follows a separate two-year agreement worth more than US$32 million. Customer GPU quantity and MW are undisclosed; the September announcement gives more than 2,000 B300 GPUs combined across this and another new agreement. A two-year extension option worth US$19.4 million is excluded. December 2026 turnover is a target, not delivery evidence.

Hardware: NVIDIA Blackwell B300 GPUs; exact customer allocation undisclosed

Planned 100 MW deployment across the partnership, not a disclosed customer purchase or energized power measurement. Site, contract value and first-phase MW are undisclosed. Existing private inference traffic does not establish delivery of the planned cloud datacenter. The initial 2026 datacenter target is distinct from 2027 CS-4 access.

Hardware: Cerebras wafer-scale systems with GPUs; CS-4 access expected in 2027

Google campus under construction since June 2025. Connected to Goodnight 1 wind farm (265.5 MW, operational) and Goodnight 2 (265.5 MW, under construction). Wind generation ratings are not campus load or energized IT capacity. Campus MW, agreement value and first-capacity date are undisclosed. Separate from Crusoe’s Abilene campuses.

Project Plans 2 and 3 signed September 18 under the May 5 master agreement. Approximately $11.6B is subject to delivery, availability and termination provisions. Potential additional $9B, equity warrants and supplier capital spending are excluded. Site, MW and dated first service are undisclosed.

Hardware: CPU cloud infrastructure

Alibaba Cloud plans to establish its first Finnish cloud region within 12 months. The company did not name a site, data-center partner, MW quantity, or investment amount. A cloud-region target is not a facility power or energization date.

The $65M figure is an annual inference agreement; the term and MW allocation are not disclosed. Do not treat the annual amount as a full-term contract value or compute it as annual capital expenditure. Crusoe said total Managed Inference contracted annual revenue exceeded $100M; that portfolio figure overlaps this agreement and other customers.

Hardware: Dedicated NVIDIA HGX B200 systems with Quantum-2 InfiniBand

Nscale's September 18 Form S-1 says four Anthropic Services Agreements signed August 25, 2026 provide for aggregate payments of up to approximately US$44.6 billion. CNBC reported that the six-year agreements cover about 460 MW of compute capacity at Monarch and that the campus was expected to come online at the end of 2027. The 460 MW and timing are reported terms, not utility measurements; the filing does not specify whether the MW figure is IT load or total facility power. Do not treat the capacity as energized or delivered. The agreement value is full-term service spend, not annual revenue or infrastructure capital expenditure.

Hardware: NVIDIA Vera Rubin

The consortium signed lease documentation with an Australian Anthropic subsidiary for the first stage of Western Downs Digital Park, subject to customary approvals. The disclosure does not specify leased MW, contract value or delivery date. Consortium funding obligations also remain subject to approvals. Campus estimates are not an Anthropic allocation.

Perplexity contracted Crusoe Cloud for model training and managed inference. Contract value, GPU count, MW allocation, sites and first-service date are undisclosed. Crusoe also plans to adopt Perplexity software for its employees; that reciprocal software purchase is not additional compute capacity.

Hardware: NVIDIA GB300 NVL72 clusters / InfiniBand

Crux AI launched as the U.S.-based compute-as-a-service company formed through the Blackstone and Google TPU cloud venture. Crux says Blackstone funds committed $5 billion in equity and Google will provide TPUs, software, and services. Crux plans to bring its first 500 MW of TPU capacity online in 2027 as the initial phase of a multi-gigawatt roadmap. The 500 MW is provider-level planned compute capacity, not a customer allocation; the source does not define critical IT load versus facility power or disclose site allocations. Bloomberg reported delays at proposed sites, including Google's scrapped plan for Crusoe to build in Cheyenne, but did not quantify the affected share or report a revised Crux-wide target. Keep the 2027 target as announced; no dated operating or energization evidence is established. The $5 billion is an equity commitment to the platform, not customer contract value or disclosed construction spend, and is excluded from financial totals.

Hardware: Google Tensor Processing Units (TPUs)

Google announced a €13B two-year Finland program covering digital infrastructure, clean energy, and partnerships. This is not a site cost or contract value. Google named activity in Hamina, Kajaani, Muhos, and Vaala, but disclosed no site-level MW or first-capacity dates. On October 6, 2026, the Finnish Licensing and Supervision Agency asked Tuike Finland Oy to suspend work that significantly changes the environment at the planned Muhos and Kajaani sites by October 23, until the environmental impact assessments are complete. The notice does not cover planning, measurements, soil studies, or other low-impact and readily reversible work. It is a permitting and delivery risk, not evidence of energization or a revised delivery date.

Moore Threads announced a deep partnership with JD Cloud to plan a 100,000-card domestic intelligent-computing cluster on September 9, 2026. The announcement does not identify the site, GPU model, power quantity, IT load, value, or first-service date. The GPU target is not converted to MW or treated as a delivered fleet.

Hardware: 100,000 domestic GPUs; model and site allocation not disclosed

Firmus and OpenAI announced a multi-year agreement for dedicated compute capacity at two Malaysian sites. They did not disclose the OpenAI MW allocation, contract value, or first-capacity date. Firmus's more-than-900 MW company-wide contracted-capacity figure includes multiple customers and is not assigned to this agreement.

Palantir named Nebius its preferred sovereign AI infrastructure partner. Following an integration period, Palantir customers can access Nebius compute and inference. The companies also plan to accelerate capacity deployments, including modular data centers at sites with available power. The announcement gives no contract value, customer allocation, MW quantity, or first-service date.

MEEZA announced a long-term lease with an unnamed global hyperscaler for 8 MW of data centre services at an upcoming campus. MEEZA valued the agreement at more than QAR 1 billion. The customer, contract term, site name, delivery date, and IT-load versus facility-power basis are not disclosed. Keep its capacity and native-currency value out of aggregates. This is a separate announced contract, but the unnamed customer prevents overlap checks against the other MEEZA hyperscaler records.

Initial compute commitment for Figure’s Helix models. Intent to expand above $6B is not an additional signed commitment. Nscale’s separate equity investment is undisclosed and excluded. GPU count is a potential upper limit; customer MW and energization date are undisclosed. Deployment target does not establish operational delivery.

Hardware: Potential deployment of up to 100,000 NVIDIA Vera Rubin GPUs

QumulusAI reports full deployment of 616 GPUs for Runpod in August 2026, ahead of the September 1 target. The agreements have different one- and two-year terms, but value and MW are not disclosed. QumulusAI says the GPUs are served from its existing US data-center footprint; no specific site or utility energization date is given.

Hardware: 616 NVIDIA RTX PRO 6000 Blackwell GPUs in 77 nodes

An unnamed neocloud operator contracted 14 MW within the 37 MW LON-East campus. Company-reported facility delivery does not establish customer occupancy or a utility energization date. Price and term are undisclosed. The campus total is not the customer allocation.

Binding term sheet with an unnamed sovereign AI company based in India for Blackwell cloud GPUs and allied services. Customer identity is undisclosed; do not attribute the order to Sarvam. Site, MW, GPU count and first-service date are undisclosed. Original INR value is retained without currency conversion.

Hardware: NVIDIA Blackwell cloud GPUs; count undisclosed

ChronoScale announced plans for an approximately 50 MW Microsoft deployment on August 27. A ChronoScale filing says the parties entered a two-year strategic partnership on August 6, then extended the service term to three years on August 24. The filing targets projected completion in Q1 2027 and says delivery depends on financing and customary development, construction, and operational conditions. The North American site, contract value, and IT-load versus facility-power basis are not disclosed. This record tracks the planned customer deployment; it does not establish delivered or energized capacity.

Hardware: NVIDIA GB300 NVL72 systems with liquid cooling

SK Telecom plans to spin off its wholly owned SK Broadband data-center business into SK Horizon. A definitive agreement provides for KRW 3.08 trillion in combined equity investment from KKR-managed funds and the IMM Investment-Stonebridge consortium. SKT says SK Horizon's 318 MW portfolio figure includes eight operating data centers and new AIDCs under construction, including Ulsan and Guro. The source gives no site allocation or IT-load versus facility-power basis. The portfolio figure is not incremental capacity and may overlap group-level SK Hyper plans; it is excluded from capacity totals. The equity is financing, not project construction spend, and remains in native currency.

RUM Group's August 24 SEC filing confirms a six-year GPU-services agreement with an unnamed US cloud customer at its Maysville site, then under development. The three initial tranches have an approximately US$13.7 billion total order value, spread evenly across the tranches. The customer must review and accept RUM's proposed delivery date for the third tranche before obligations for that tranche apply. The filing does not name the customer; The Information reports Anthropic, but that attribution is not confirmed by the filing or a company announcement. It states no MW, GPU count, first-delivery date, or utility energization date. RUM said it did not yet have financing for the required development and equipment when it filed; no financing condition applies to its agreement. Expansion agreements are possible but unpriced and are not included.

Hardware: GPU access and GPU services; model and quantity undisclosed

CoreWeave announced a multi-year agreement with Hudson River Trading for AI research and model development. Its release describes the deal as multi-billion-dollar but gives no exact value, term length, MW allocation, GPU quantity, site, or start date. The multi-billion description is retained as source text and excluded from numeric contract totals.

Hardware: NVIDIA Vera Rubin NVL72 and HGX B200 systems

QumulusAI signed a seven-year colocation agreement for up to 3.75 MW in metropolitan Atlanta, conditional on the current occupant leaving. QumulusAI states a Q4 2026 ready-for-service target and says the site could support up to 2,048 B300-class GPUs. A right of first offer for up to 7 MW is not contracted capacity and is not included. The source does not say whether 3.75 MW is critical IT load or facility power. The later US$211.9 million customer agreement uses this site, but the MW basis remains unclear and the planned site amount is not a delivered customer allocation.

Hardware: Site expected to support up to 2,048 NVIDIA Blackwell B300-class GPUs

Duos signed service orders reserving up to 55 MW of total facility capacity for Axe Compute across multiple U.S. sites. Duos supplies the facilities, power, and colocation services; Axe brings and operates the compute equipment. Initial project readiness was targeted to begin in late 2026 and continue into early 2027, subject to construction, commissioning, testing, and Axe's written acceptance. The agreements are separate from the earlier 10 MW Georgia deployment. The reported hosting payments are retained as source text and excluded from compute-purchase and infrastructure-investment totals.

Program ambition for up to 1 GW by 2030, supported by enterprise compute commitments. Individual sites and incremental capacity beyond earlier Mistral projects are not allocated in this announcement. Excluded from physical capacity totals.

Dedicated IBM Cloud cluster for Together AI open-source inference. Site, MW and GPU count are undisclosed. Expected availability is a target, not delivery evidence. Together AI’s separate financing is excluded.

Hardware: NVIDIA HGX B300 / Spectrum-X

IMC said it would significantly scale its use of CoreWeave, where it had operated several production clusters since 2025. CoreWeave describes additional capacity as immediately available but gives no quantity, value, site, or service-start date. The prior production relationship means this expansion is not a new deployment start.

Three Indonesian facilities with 360 MW of contracted IT power, expected online in 2028. This is the regional deployment allocation, not CoreWeave’s global contracted-power total.

Sharon AI reports a five-year cloud service agreement with an unnamed global AI platform. The initial deployment is expected to use 2,048 NVIDIA B300 GPUs, with revenue expected from Q1 2027. Sharon AI's 132 MW AI Factory capacity and 120 MW contracted across end customers are not allocated to this agreement. Revenue timing does not establish a first-compute or energization date.

Hardware: Initial deployment of 2,048 NVIDIA Blackwell Ultra B300 GPUs

The companies plan to run a Crusoe Spark modular data center on Aalo power in 2027 as a proof of concept. They also describe future Aalo Pods and 50 MWe reactor plants, but do not state a data-center power allocation. Aalo's 10 MWe reactor rating is not entered as delivered campus power. No MW, contract value or energization evidence is disclosed for the data center.

Flow Traders selected CoreWeave as its primary cloud platform for AI model training and secured dedicated compute capacity. The announcement gives no contract value, GPU count, MW allocation, term, site, or dated service start. Flow Traders' plan to launch an AI division is not a compute-delivery date.

The filing reports that a February 2026 lease amendment obligates Nscale to lease an additional 89 MW when available. Ionic expects to energize the increment in the second half of 2027, but the capacity is subject to regulatory approval and is not yet operating. This is incremental to the original 234 MW lease.

QumulusAI first expected capacity for customer use in Q3 2026. It later reported B200 handoff on September 8 and B300 handoff at its Kansas City site on October 1; on October 15 it said all GPUs under the agreement were in production and billing. The agreement reached service after its Q3 target. Customer MW and utility energization are undisclosed. QumulusAI's separate portfolio MW figures are not assigned to this customer.

Hardware: NVIDIA Blackwell B200 and B300 GPUs

Energy Vault began construction on a powered-shell campus for Crusoe Cloud. Phase 1 is contracted for 8 MW and includes ten Crusoe Spark modular data centers. Energy Vault's August 11 filing reports a 1H 2027 commercial-operation target; the July 27 release gave Q1 2027. The release describes expansion potential to 25 MW in Phase 2 and planned site capacity up to 500 MW. Those are not contracted Phase 1 capacity and are excluded from totals. Construction is not energization; no delivered or energized MW is confirmed.

Delta Forge 1: 15-year take-or-pay lease with an unnamed US hyperscaler. Date is the results disclosure, not the undisclosed signing date. Each campus is a separate lease; customer identity is not inferred. Hosting revenue is not included in compute purchase values.

Delta Forge 2: 15-year take-or-pay lease with an unnamed US hyperscaler. Date is the results disclosure, not the undisclosed signing date. Each campus is a separate lease; customer identity is not inferred. Hosting revenue is not included in compute purchase values.

Polaris Forge 3: 15-year take-or-pay lease with an unnamed US hyperscaler. Date is the results disclosure, not the undisclosed signing date. Each campus is a separate lease; customer identity is not inferred. Hosting revenue is not included in compute purchase values.

QumulusAI reports a two-year agreement worth more than US$32 million for dedicated Blackwell B300 clusters serving a generative-AI inference platform. The September 29 US$29.1 million agreement is a separate second commitment by the same customer. The original fall 2026 target remains unverified; the later agreement's December target does not prove delivery of this tranche.

Hardware: NVIDIA Blackwell B300 GPU clusters

SK Telecom established SK Hyper as a wholly owned AI data-center development company. It approved KRW 750 billion through 2030 for foundational development, including sites and substations. The company targets phased opening of 5 GW in 2029 and expansion toward 15 GW total by 2035, subject to market demand. The capacity basis is not stated. In an October 22 Bloomberg interview, SK Hyper's CEO said the first 5 GW phase would require $350 billion including equipment and that the company was discussing financing with potential partners. This is an estimated phase cost, not committed financing, project spend, or contract value. It is excluded from financial totals. The 5 GW is part of the 15 GW plan and is not additive.

QumulusAI reports a two-year take-or-pay agreement worth more than US$18 million with an unnamed GPU marketplace. Its September 10 release says the full Blackwell B300 deployment was handed to the customer on September 1 at the Philadelphia colocation site. GPU count and MW are not disclosed; operational handoff does not give a utility energization measurement.

Hardware: NVIDIA Blackwell B300 nodes

Sharon AI reports a five-year cloud computing agreement with an unnamed global AI lab. It plans deployments across data center infrastructure in New Zealand and expects revenue to begin in Q1–Q2 2027. That revenue window is not reported as a first compute or power date. Sharon AI's 132 MW AI Factory capacity and 116 MW contracted to customers are company-wide amounts, not this customer's allocation.

Crusoe and Lancium announced a 1 GW campus for an unnamed hyperscale customer. Construction is expected to start in Q3 2026. This is a separate project from IREN’s Childress campus.

China Telecom and Huawei launched an Ascend 910C cluster on July 9, 2026. Shaoguan government reports 11,520 accelerators; China Telecom reports 30 supernodes, 9,000 PFLOPS FP16, and approximately RMB 5.5 billion total project investment. The investment is kept in native currency and excluded from USD contract totals. Accelerator count and compute throughput are not MW, critical IT load, or utility power. The launch establishes operational compute, not a measured power-delivery date.

Hardware: Huawei Ascend 910C accelerators in 30 supernodes; 11,520 cards reported by Shaoguan government

TeraWulf's subsidiary entered a 20-year lease with Anthropic for approximately 401 MW of critical IT load at the Justified Data Campus. Initial leased premises are expected in phases during the second half of 2027; full delivery is expected in early 2028. The lease has two five-year renewal options. TeraWulf states approximately $19 billion of contracted lease revenue over the initial term. That service revenue is not a compute-purchase value or infrastructure investment and is excluded from contract-value analytics. The August 5 filing says the lease had not commenced as of the filing date. This is a customer IT-load allocation, not the campus's gross power availability or proof of energized capacity.

Joint operations company for the existing Narvik development. Its planned 2026 start refers to the operations company, not energized customer capacity. No incremental power allocation or contract value is disclosed.

Colocation agreement across two campuses. CoreWeave stated that initial capacity at Stockholm 4 South was already online. No deal-specific power allocation or contract value is disclosed. The company-wide active and contracted power figures are not assigned to these campuses.

The June 24 MSA covers an initial 20 MW of critical AI compute capacity and gives the customer an option to expand to 52 MW. The September 17 update says operations are targeted to begin in November 2026 and that more than $70 million had been invested as of September 15; neither statement confirms delivery. The approximately 340 MW figure is total potential campus capacity, subject to power, infrastructure, financing, and regulatory approvals, and is not included here. Contract revenue is conditional on the maximum term and is not recorded as compute-purchase value or capital expenditure.

Sharon AI reports a six-year agreement with NVIDIA worth up to US$4.88 billion and enabling 72 MW of new Australian data-center capacity and up to 40,000 GB300 GPUs. The parties use a revenue-sharing and credit-support model. The 72 MW basis is unspecified and may support customer capacity represented in separate Sharon AI agreements. Keep the 72 MW out of IT-load, campus, or energized-power totals, and exclude the maximum collaboration value from contract totals because it may overlap revenue from customer agreements.

Hardware: Up to 40,000 NVIDIA Grace Blackwell GB300 GPUs

QumulusAI reports more than US$124 million in two three-year customer agreements, one with Hyperbolic and one unnamed AI inference platform. The source states that the agreements support 1,280 Blackwell GPUs and include nearly US$21.9 million in combined upfront commitments. It does not allocate contract value or GPUs by customer, so this combined announcement is one record. The figure is counted once; later customer-specific agreements are separate commitments.

Hardware: 1,280 NVIDIA Blackwell GPUs across both agreements

Nebius states approximately £1.7B of investment for UK capacity buildout and three new deployments with 65 MW combined capacity at full ramp in 2027. The source does not define the MW basis or identify individual site allocations. The investment is in pounds and is not a compute contract value. A separate first UK deployment was already operating from November 2025 and is not included as a fourth deployment in the 65 MW figure.

Hardware: NVIDIA infrastructure; model and quantity not stated

SK Telecom plans a gigawatt-scale AI Cloud built on NVIDIA DSX, with the first AI factory planned for 2027. The source does not state an exact MW quantity or IT-load versus facility-power basis. This SKT and NVIDIA plan may overlap SK Hyper's wider 15 GW program and the Ulsan-AWS development; its capacity is excluded from totals pending site and scope reconciliation.

Firmus announced a 12-year wholesale supply agreement for 600 MW of firm electricity. Gunvor will support 1.2 GW of new renewable generation and 1.5 GWh of battery storage by 2032. The agreement supports two planned South Australian campuses; it does not establish campus capacity, IT load, or a first-power date.

Firmus announced a dedicated 360 MW campus with NVIDIA and DayOne. The source says the agreement covers up to 170,000 accelerators through 2027 and 2028 but gives no first-service date. Firmus expects US$25–30 billion in customer offtake receipts over the first six years; that is a company estimate across customer offtake, not a Firmus–NVIDIA purchase value or realized revenue.

Hardware: Up to 170,000 NVIDIA accelerators across Grace-Blackwell, Vera-Rubin, and Vera platforms through 2027 and 2028

Firmus describes 2.7 GW as the planned capacity of the first phase of two campuses. It does not state whether this is IT load or facility power, give a site-level split, or provide a first-capacity date. The announcement places the sites under Firmus's wider Southgate strategy. Their relationship to the earlier 1.6 GW CDC rollout is not clear, so this figure is kept out of capacity subtotals.

The partners describe 328 MW of installed fuel-cell generation capacity for the first US project, expected to be operational in 2026. The release does not name the site or report commissioning, energization, or delivered compute. Keep the generation rating separate from data-center load and actual energized power.

Sharon AI reports an approximately US$950 million five-year take-or-pay cloud infrastructure agreement with an unnamed global technology company. Its August 20 release confirms that the customer accepted an initial deployment, with more NVIDIA GPU clusters expected in the following months. Sharon AI's reported company-wide secured and contracted MW are not allocated to this customer. Customer MW, GPU count, exact site, and utility energization date are unknown.

Hardware: NVIDIA GPU clusters; model and quantity undisclosed

The May 6 announcement covers more than 300 MW at Colossus 1. SpaceX’s June 5 prospectus describes Anthropic service agreements across Colossus 1 and 2, approximately 325,000 GPUs, and a $1.25B monthly fee at that time. Anthropic’s October 26 Form S-1 reports that agreements entered into or amended between May and October provide for up to $120.5B through 2029, with fees of about $3.1B per month after all capacity is delivered, which Anthropic expected in December 2026. The separate agreements have initial non-cancelable periods ending in January and March 2027, after which either party may cancel with 90 days’ notice. The upper-limit spend is not a guaranteed contract value. The filing does not add a new MW allocation or give an energization date. Orbital compute remains an expression of interest.

Hardware: 220,000+ NVIDIA GPUs

Nscale announced an agreement with Microsoft and Start Campus for 66,000+ Rubin GPUs starting in late 2027. The 200 MW figure is for a second campus building; the full campus is permitted for 1.2 GW. Neither site figure is an allocation to this Microsoft deployment. Nscale separately reported more than 12,600 Blackwell Ultra GPUs at the first building; those GPUs are not added to this expansion.

Hardware: 66,000+ NVIDIA Rubin GPUs

Nscale states a second 200 MW building and €465M investment in that building, plus €230M for shared infrastructure. The figures describe campus build and investment, not Microsoft's compute allocation or contract value. The campus is permitted for 1.2 GW; the full-campus figure is not added to this building. No energization date is stated for the building.

Expansion separate from the earlier 5,120-GPU programme. April disclosure valued this project at approximately $2.8B; July 21 described Yotta 2 at approximately $2.5B without reconciling the difference. September 8 reports equipment deliveries across the wider Yotta programme, not operational acceptance of this tranche. Project value remains outside compute-purchase totals. No MW conversion is inferred from GPU count.

Hardware: 20,736 NVIDIA B300 GPUs

Executed agreement with an unnamed US AI service provider specialising in large-scale model training. Committed value covers GPU cloud computing and allied services for six months. Customer identity, site, GPU count, MW and service commencement are undisclosed.

Anthropic committed more than $100B over ten years to AWS technologies for up to 5 GW of new Trainium and related capacity. Nearly 1 GW is expected before the end of 2026. Amazon also announced a $5B equity investment, with up to $20B more later; those equity figures are omitted from the dollar column.

Hardware: Trainium2–4

Jane Street committed approximately US$6B to use CoreWeave's AI cloud platform across multiple facilities. The agreement expands an existing relationship and includes access to next-generation compute, including NVIDIA Vera Rubin technology. The announcement does not state the term, MW allocation, sites, or first-service date. Jane Street separately invested US$1B in CoreWeave equity; that financing is excluded from the compute purchase amount and financial totals.

Hardware: NVIDIA Vera Rubin technology included; quantity not stated

Expanded Microsoft agreement adds more than 30,000 Rubin GPUs at Narvik in 2027. The 230 MW figure describes the campus, not the power allocated to this expansion.

Hardware: 30,000+ NVIDIA Rubin GPUs

Multi-year agreement to support Claude development and deployment, with capacity starting later in 2026. No exact dollar or megawatt figure is disclosed.

Expanded agreement for approximately $21B through December 2032. Includes new capacity and exercise of an option under a previous order. Read with the earlier agreement rather than treating both as independent projects.

The April 6 announcement is for multiple gigawatts of next-generation TPU capacity starting in 2027. Anthropic’s May 6 recap states a 5 GW agreement with Google and Broadcom. No public dollar figure appears on either Anthropic page. Builds on a prior October 2025 TPU-capacity increase.

Hardware: Google TPUs

China Telecom and Alibaba Cloud jointly built the first Zhenwu ten-thousand-card-class cluster in the Greater Bay Area. The cluster was reported online on April 3, 2026. The report gives a class-size description, not an exact card count. It reports no MW, site-specific capital amount, or utility energization date. Treat operational compute as evidenced; keep actual energized power unknown.

Hardware: Alibaba Zhenwu AI accelerators; ten-thousand-card-class cluster

Site securing and development agreement for the Sievari industrial area. Fortum supports grid design and permitting; preliminary land-sale work is underway. Power capacity remains subject to design and permitting. No first-capacity date or contract value is disclosed.

New 900 MW campus serving Microsoft, next to the existing Abilene development. Includes two buildings and onsite generation. The combined 2.1 GW Abilene figure includes the existing campus.

The campus already operates four Crusoe Spark modular data centers on a 12 MW / 63 MWh microgrid. The announced expansion increases the deployment to 24 data centers and describes work with Redwood Energy at 20 MW, but does not state whether 20 MW is incremental or the final power-system total. Do not add this figure to power subtotals until the basis is reconciled. The source reports 99.2% operating availability for the existing system; that is not a new delivery milestone for the expansion.

Includes $12B of dedicated capacity and up to $15B of additional capacity that remains after sales to other customers. The $27B ceiling is not all dedicated capacity.

Hardware: NVIDIA Vera Rubin

Letter of intent for up to 1.35 GW at Monarch. The campus’s potential power expansion above 8 GW is not the Microsoft allocation. A letter of intent is not a completed supply contract.

Hardware: NVIDIA Vera Rubin NVL72

Nscale announced its agreement to acquire AIPCorp and Monarch Compute Campus. Nscale states initial onsite microgrid power capacity of 2 GW is expected online by 1H 2028, with expansion to approximately 8 GW planned for 2031. The 2 GW is site power capacity, not a customer compute allocation or energized power. Microsoft's separate up-to-1.35 GW letter of intent at Monarch is linked and must not be added to the campus figure.

An additional $100B over eight years expands the earlier $38B AWS agreement. Approximately 2 GW of Trainium capacity is included. Amazon’s equity investment is separate.

Hardware: Trainium3 / Trainium4

Four-year NVIDIA DGX Cloud engagement within Yotta's Blackwell Ultra supercluster. April says NVIDIA is expected to account for roughly half the tranche offtake. The full 20,736 GPUs, host facility MW and over $2B build budget are not NVIDIA allocations or additional contract value. The April deployment completion target is September 30, 2026. No customer service acceptance was verified.

Long-term €1.2B infrastructure partnership at Borlänge. EcoDataCenter will build and operate the facility. The disclosed euro amount is retained without a dollar conversion.

Hardware: NVIDIA Vera Rubin

OpenAI announced 750 MW of inference compute to be introduced in phases through 2028. The announcement does not disclose a contract price or individual sites.

Hardware: Cerebras wafer-scale systems

15 MW deployment at Verne’s Icelandic campus. The power basis is not specified as critical IT load, so this figure remains outside the IT power subtotal. Verne reported that the first 7.5 MW phase was deployed on April 23, 2026; this does not confirm the remaining 7.5 MW.

Hardware: Approximately 4,600 NVIDIA Blackwell Ultra GPUs

Anthropic committed to purchase $30B of Azure compute capacity and to contract additional capacity up to 1 GW, initially on NVIDIA Grace Blackwell and Vera Rubin systems. NVIDIA and Microsoft also committed to invest up to $10B and $5B in Anthropic; those equity figures are omitted from the dollar column. Amazon remains Anthropic’s stated primary cloud and training partner.

Hardware: Grace Blackwell / Vera Rubin

Anthropic announced a $50B investment in U.S. computing infrastructure, building data centers with Fluidstack in Texas and New York. The announcement says Fluidstack can deliver gigawatts of power but does not state a megawatt figure. This is an infrastructure build, not a cloud offtake.

The results recap a $6.3B NVIDIA arrangement; the September 9 order gives NVIDIA access to residual unsold capacity. This backstop can overlap capacity sold to other customers and is excluded from aggregate purchase values. The date identifies the cited results disclosure.

Approximately $9.7B over five years, with 20% prepayment. Horizon 1–4 support 200 MW of critical IT load within the 750 MW Childress campus; the full campus is not assigned to this contract.

Hardware: NVIDIA GB300

Multibillion-dollar agreement for tens of thousands of NVIDIA GPUs. No exact contract price or megawatt allocation is disclosed in the announcement.

Hardware: NVIDIA GPUs, including GB300 NVL72

Microsoft disclosed an incremental $250B Azure services purchase by OpenAI. The equity ownership and prior investment are separate from this services commitment.

MEEZA announced a lease with an unnamed global hyperscaler for 6 MW of data centre services as phase one, with capacity intended for cloud and AI workloads. The company stated a value above QAR 750 million. Customer, site, contract term, delivery date, and IT-load versus facility-power basis are not stated. Do not add this capacity or native-currency value to USD totals. The unnamed customer prevents overlap checks against MEEZA's other hyperscaler contracts.

Vantage, OpenAI, and Oracle announced this Stargate site on October 22, 2025, with completion scheduled for 2028. Vantage's current campus page states 902 MW of critical IT load across four buildings. This is the full site figure, not an OpenAI or Oracle allocation; utility power requirements use a separate basis. The Wisconsin PSC revoked the first transmission application's completeness determination on September 10, 2026. The replacement application is docket 137-CE-229 and remains under review. Oracle's September 16 statement says its commitment to completion is unchanged but gives no first-service date. Construction completion is not a first-capacity target, and no energized-power evidence is recorded.

The October 2025 plan targeted up to 1.6 GW of AI factories through 2028 and cited A$4.5 billion for its initial stage and up to A$73.3 billion at program scale. Firmus later said it and CDC had mutually agreed not to proceed with the proposed development partnership. This ended partnership is not Firmus's current total capacity plan. The reported 42 MW deployment is not assigned to IT load or energized power, and its deployment date is undisclosed. Do not include the original program budget or capacity in current totals.

This is the initial Microsoft deployment at the first SINES building. Nscale's September 18, 2026 Form S-1 says more than 12,600 Grace Blackwell Ultra GPUs were deployed there for Microsoft. The filing confirms deployment but does not date the start of customer service or disclose Microsoft MW allocation or utility energization. This is separate from the 66,000+ Rubin expansion announced in May 2026 for late 2027.

Hardware: Approximately 12,600 NVIDIA GB300 GPUs

This is a site-level deployment under Nscale's approximately 200,000-GPU Microsoft agreement across four countries. The GPUs are not added to a global GPU total. Nscale's October 1, 2026 update says Ward County was live and delivering thousands of GPUs, but it does not identify them as Microsoft capacity. No dated evidence that Microsoft compute service began was verified by October 26, 2026. Ionic Digital's separate 234 MW lease is facility power, not Microsoft's compute allocation.

Hardware: Approximately 104,000 NVIDIA GB300 GPUs

Ionic leased its full 234 MW Ward County capacity to Nscale for an AI/HPC campus. Its filing says the powered campus has operated since 2023 and monthly lease payments began in August 2026. Nscale's October 1 update says the Ward site is live and delivering thousands of GPUs, but it does not identify those GPUs as Microsoft capacity. The 234 MW is site power, not customer IT load. The campus announcement's approximately 240 MW figure is not added as a second record.

Filed September 30 for a September 25 order. Up to approximately $14.2B through December 2031, subject to delivery and availability terms. An option to expand through 2032 is separate.

September expansion only. CoreWeave also disclosed a $4B May expansion and approximately $22.4B in cumulative OpenAI contracts. Do not add the cumulative figure again.

September disclosure of the July Oracle agreement: over $300B across five years and up to 4.5 GW of additional Stargate capacity. This is part of the Stargate program, not capacity additional to its headline total. For Project Jupiter, Oracle proposed an Energy Service Agreement with El Paso Electric on October 13 to support late-stage construction, initial commissioning, emergency backup, and worker life safety. On October 15, Oracle said its revised fuel-cell permit reduced permitted emissions by 35%; on October 22, it described the private microgrid plan and stated up to 2.45 GW of installed fuel-cell generation capacity. Oracle's October 13 update says campus scope and planned capacity did not change. The 2.45 GW power-system rating is recorded separately from this compute agreement. These releases do not confirm energization or a revised customer delivery date. Oracle's reported force-majeure notice and debt-market concerns remain unverified in company or regulatory filings; they are not treated as changed contract terms. The proposed utility agreement does not disclose an MW allocation; its expected $10M annual power payment is not compute-contract value or project investment. A separate proposed ratepayer credit is not campus capacity or customer compute.

Nscale's January 2025 UK investment announcement said the Loughton site was scheduled to be live in Q4 2026. Its September 2025 Microsoft deployment announcement stated 50 MW of AI capacity, scalable to 90 MW, with 23,040 GB300 GPUs targeted from Q1 2027. These were separate site-readiness and customer-service targets. On October 19, 2026, Reuters reported Microsoft's spokesperson confirmed that its decision to terminate the Loughton agreement related to this single site. Nscale said it remains committed to the site and UK expansion, is exploring alternative power sources, and has a strong long-term relationship with Microsoft for other projects. The 50 MW is retained as the historical announced customer allocation, not a current Microsoft commitment. The Microsoft agreement ended before first compute; the 2027 customer target no longer applies. Nscale has not announced a replacement customer or revised site delivery date. Reported grid delays into the 2030s are not a confirmed energization date. The 90 MW figure remains a possible site power scale, not IT load or contracted Microsoft capacity. The separate UK-wide investment plan and Stargate UK framework are not added to this site record.

Hardware: 23,040 NVIDIA GB300 GPUs initially

Approximately $17.4B for dedicated GPU capacity, subject to deployment and availability. Additional services could increase the value to approximately $19.4B. Microsoft can terminate delayed tranches under the filing’s conditions.

Third lease dated August 28 adds 150 MW to the initial 250 MW, bringing CoreWeave’s campus allocation to 400 MW. The record contains only the incremental allocation. Reported campus delivery remains attached to the original record to avoid counting it twice.

Two ten-year hosting agreements. Approximately $3.7B initial-term revenue; $8.7B includes optional extensions. Google’s $1.8B backstop and warrants are financing support, not another compute purchase. The 160 MW expansion is separate. The August 2026 filing reports partial lease commencement by June 30, with later phases extending into early 2027. Its 102 MW energized figure covers Lake Mariner across customers, not a Fluidstack allocation.

Nscale and Aker announced a 230 MW initial campus with ambition for another 290 MW. Nscale later disclosed a possible 115 MW Narvik expansion backed by an uncommitted financing feature. OpenAI welcomed the opportunity to be an initial offtaker; the announcement does not state a signed OpenAI capacity allocation. OpenAI's page describes a 500 MW data center. These figures may describe the Narvik development also covered by Microsoft/Nscale announcements, so this record is excluded from capacity totals. Nscale and Aker said they committed about $1B to an initial 20 MW phase; this is project financing/investment, not an OpenAI contract value. The end-2026 target is a GPU deployment target, not proof of first energization.

Hardware: 100,000 NVIDIA GPUs targeted initially

CoreWeave described its intent to commit up to $6B to equip a new Lancaster data center. The source states an initial 100 MW facility with potential expansion to 300 MW, but does not define the MW basis or give a first-capacity date. The amount is an infrastructure commitment, not a signed compute purchase or annual capital expenditure. The 300 MW expansion is potential and is not added to the stated 100 MW initial capacity.

MEEZA announced a long-term agreement with an unnamed global hyperscaler to use a 4 MW data centre expansion. MEEZA valued the contract at more than QAR 350 million over more than ten years. On July 1, 2026, MEEZA said the expansion was handed over on June 21 after signing a ready-for-service certificate. The source does not name the customer or define the MW basis; retain the native currency and exclude 4 MW from power aggregates until the basis is known.

SK Group and AWS signed a 15-year strategic partnership to build a data center in Ulsan for a new AWS AI Zone. SK Group leads construction; AWS will establish the AI Zone and provide cloud capabilities. SK Telecom's July 3, 2025 update retains the 2027 initial-operations target and says a combined total above 300 MW by 2030 includes Ulsan and the Guro data center; it does not give Ulsan's site capacity or MW basis. The 300 MW is not assigned to this record. The source gives no contract value or later delivery confirmation. The Ulsan project is related to broader SK data-center plans, so no group-level capacity is assigned to this record.

Initial leases cover 250 MW for approximately 15 years, with approximately $7B expected revenue. A separate 150 MW option was subsequently exercised, taking the campus to 400 MW under contract. The initial allocation is not the whole campus. Hosting revenue is excluded from compute purchase totals.

The 60 MW announcement is the planned total site capacity, not an incremental 60 MW above the existing 30 MW. The Q2 2025 go-live target has elapsed. Nscale's current site page, checked October 25, 2026, states 30 MW operational capacity, expandable to 60 MW; it gives no dated confirmation that the expansion reached 60 MW. Bloomberg reported on October 15, 2026 that construction was six months behind schedule and that 6.4 MW of the planned 30 MW was operating in early October. This reported operating amount conflicts with Nscale's 30 MW operational-capacity statement and has no published utility or company measurement cited here. Keep the source claims separate; do not replace the announced 60 MW site capacity or count either claim as an independently verified energized-power total. Bloomberg also reported delay penalties owed to customers, citing people familiar with the matter, without a public amount or filing that confirms the liability.

Initial compute agreement. Later expansions are separate records; the September cumulative figure is not another contract.

Stargate is a new company that intends to invest $500B over four years in U.S. AI infrastructure for OpenAI, beginning with $100B. Initial equity funders: SoftBank, OpenAI, Oracle, and MGX. Technology partners: Arm, Microsoft, NVIDIA, Oracle, and OpenAI. This is a project-company build, not a bilateral cloud offtake. The $100B figure is the initial deployment; the $500B figure is the stated intention.

xAI reported a 100,000-Hopper-GPU Colossus training system in use. The $6B funding round is not a compute contract price. This record describes a self-build milestone.

Hardware: 100,000 NVIDIA Hopper GPUs

Government announcement of support for a Canadian AI data center serving Cohere. The up-to-C$240M government contribution is not the total project cost or a compute purchase price. Builders are not named in this source.

Yotta reports Sarvam 1 was trained on Shakti Cloud in India. Uses the October 24, 2024 announcement date in the body; the page header says November 24. Price, reserved GPU count and MW are undisclosed. Sarvam released 30B and 105B models on March 6, 2026, trained in India using IndiaAI Mission compute. That later release does not identify Yotta as their provider and is not a new compute contract.

Shaoguan Daily reported that Tencent and Shaoguan signed an investment agreement on May 8, 2024 for a Greater Bay Area computing center. The municipal source states RMB 5 billion total project investment, about 30,000 standard racks, and phase-one operating capability for 200,000 servers. It gives no MW, rack power, GPU quantity, or energized-power measurement. A July 20, 2026 city update said it would push Tencent and GDS projects into production before year-end; this is a government target, not confirmation that the Tencent site is live. Later Shaoguan cluster totals and the October 25 six-cluster report are not attributed to this site because their project boundaries are unclear.

Hardware: AI, large-model, cloud-computing and IoT workloads; phase-one server capability is not a GPU count