Deals

Announced · 20-year power purchase agreement (Phase 1)

Minimum 200 MW up to 207 MW for Phase 1; power from Vistra's adjacent 1,180 MW Odessa gas plant; expected available Q3 2027; campus planned to scale toward 1.4 GW over time (493-acre site)

Counterparty

Luminant ET Services (Vistra affiliate)

Where

Texas Critical Data Center, Odessa / Permian Basin, Texas, US

Amount

—

MW

207

GPUs

—

Company PR + MarketWatch/RTT Sep 21, 2026. mw set to the contracted maximum (207); minimum is 200. Vistra also gets a 5% non-voting interest in the powered project portion plus ROFR/ROFO on future development. amountUsd null (not disclosed). Full-campus 1.4 GW is planned scale, not contracted Phase 1.

Source

Announced · Long-term AI data center customer / capacity agreement (SYD1 Stage 1)

~8,200 NVIDIA Blackwell Ultra GPUs (HGX B300) expected; initial operations targeted Q4 2026; SYD1 planned as 100 MW+ campus; Stage 1 reuses existing DC building

Counterparty

GreenSquare

Where

SYD1 campus, Norwest Business Park, Western Sydney, NSW, Australia

Amount

—

MW

—

GPUs

8,200

GreenSquare/techpartner/w.media Sep 21, 2026. Financial terms and contracted MW not disclosed — mw/amountUsd left null. Campus pipeline cited as 100 MW+ at SYD1 and >1 GW broader GreenSquare portfolio; those are campus/pipeline figures, not Sharon's contracted allotment.

Source

Announced · Six-year behind-the-meter baseload power supply (incl. balance of plant)

76 MW behind-the-meter; ~40 reciprocating natural-gas generation units; deploy Q4 2026 scaling into Q1 2027; revenue expected from Q1 2027

Counterparty

Kodiak Gas Services

Where

West Texas, US

Amount

—

MW

76

GPUs

—

Kodiak IR PR Sep 21, 2026. Operator, hyperscaler, and GPU-designer guarantor not named. amountUsd null. Kodiak's second long-term primary-power contract to a data center.

Source

Announced · Nuclear uprate subscription (NU-1 tariff / ZECs)

~96 MW new nuclear capacity via EPUs on Georgia Power’s owned share of Vogtle and Hatch; ~$900M projected customer benefits over unit life (not Google payment)

Counterparty

Georgia Power

Where

Plants Vogtle and Hatch, Georgia, US

Amount

—

MW

96

GPUs

—

Georgia Power press release Sep 21, 2026 (PSC filings Docket 44280, 56002). Subject to Georgia PSC approval. Google subscribes to new NU-1 tariff and receives Zero-Emission Credits for the uprate output; structure protects non-participating customers from incremental uprate cost. Hatch 1&2 EPU still needs PSC approval; Vogtle 1&2 EPU was approved in 2025 IRP. NRC license amendments still required. $900M figure is projected customer benefits, not disclosed Google spend — amountUsd left null.

Source

Announced · Data center lease + renewable PPAs (binding MoU)

300 MW first phase (four 75 MW units); explore up to 1 GW; 15-year lease once built; PPC ~€1.2B first-phase build

Counterparty

PPC (Public Power Corporation of Greece / ΔΕΗ)

Where

Agios Dimitrios (Kozani / Western Macedonia), Greece

Amount

—

MW

300

GPUs

—

Binding MoU (regulated Athens Exchange disclosure also filed). AWS leases and runs IT; PPC owns buildings/land/energy infra and supplies renewable PPAs (>10 years). Expansion to 1 GW optional, needs further definitive agreements. First 300 MW targeted operational by end-2028; construction could begin end-2026. €1.2B is PPC first-phase capex, not lease contract value — amountUsd left null.

Source

Announced · Virtual power purchase agreement (VPPA)

433 MW from existing Panther Creek I–III wind farms; part of Oracle’s broader >1.7 GW Texas carbon-free portfolio

Counterparty

RWE Americas (Panther Creek I, II, III wind)

Where

West Texas (Sterling, Howard, Glasscock counties), US — ERCOT; supports Oracle data centers

Amount

—

MW

433

GPUs

—

RWE press release Sep 17, 2026. VPPA supplies Texas grid in support of Oracle data-center operations; Panther Creek farms previously repowered. Oracle cites goal to match 100% of AI data-center electricity with carbon-free power by 2035. amountUsd not disclosed.

Source

Announced · Renewable energy supply agreements (wind / ERCOT)

Up to 568 MW renewable electricity from ENGIE wind resources serving ERCOT; part of Oracle’s broader >1.7 GW Texas carbon-free portfolio

Counterparty

ENGIE North America

Where

Texas, US (ERCOT); supports Oracle Texas / AI data-center operations

Amount

—

MW

568

GPUs

—

ENGIE North America PR via PR Newswire Sep 17, 2026. Portfolio of wind resources; more than 500 MW / up to 568 MW. Dollar value and contract term not disclosed. Clearway also named in secondary coverage of the >1.7 GW Oracle Texas portfolio but no Clearway-primary MW disclosure found for a separate row.

Source

Announced · Data center lease (first Australia site; inference capacity)

Planned 2.16 GW campus; ~A$32 billion development; online from 2027 (subject to FIRB)

Counterparty

Zerra DC (Singapore-based developer)

Where

Western Downs Digital Park near Dalby, Queensland, Australia (~250 km from Brisbane)

Amount

—

MW

—

GPUs

—

First Anthropic Australia data-centre lease per Reuters (sources). Campus planned by Zerra DC; Anthropic/Zerra declined comment. Subject to Foreign Investment Review Board approval. ABC/Brisbane Times cite ~A$32B campus cost; Reuters cites 2.16 GW total planned capacity. Inference use, not training. Do not treat A$ as USD.

Source

Announced · Multi-year AI cloud compute offtake (training + managed inference)

Dedicated NVIDIA GB300 NVL72 clusters for training; production via Crusoe Managed Inference; MW, GPU count, and dollar value not disclosed

Counterparty

Crusoe

Where

Crusoe Cloud (sites undisclosed in reporting)

Amount

—

MW

—

GPUs

—

Bloomberg Sep 15, 2026 + Crusoe GlobeNewswire same day. Multi-year partnership for full model lifecycle on Crusoe Cloud; Crusoe also adopts Perplexity Enterprise for ~1,800 employees (software side, not capacity). Separate from Jane Street × Crusoe offtake already on tracker.

Source

Announced · 10-year data center services agreement (AI Cloud facility lease)

65.1 MW A202 IT load; energization targeted Q3 2027; with 21.7 MW A201 campus totals 86.8 MW

Counterparty

Unnamed data-center services landlord (same Johor Bahru campus as A201)

Where

Johor Bahru, Malaysia (existing Bitdeer AI campus with A201)

Amount

—

MW

65.1

GPUs

—

Bitdeer AI company PR via GlobeNewswire Sep 14, 2026 (republished on FinancialContent). Largest single SEA capacity add; liquid-cooled rack-scale NVIDIA GB300 NVL72 / Vera Rubin capable. Facility-level DC lease only — no named end-customer offtake for A202 disclosed at announcement. amountUsd left null (pipeline figure is company-wide, not this lease).

Source

Announced · AI data center campus expansion (non-binding MOU)

Pathway to 1.2 GW hub: existing ~300 MW under construction on provincial grid + up to 900 MW additional under Bring Your Own Power (partner gas generation); >$50B / ~$52.5B total capital figure cited including tenant compute and power gen

Counterparty

Government of Saskatchewan

Where

Near Regina (R.M. of Sherwood), Saskatchewan, Canada

Amount

$52.5B

MW

—

GPUs

—

Saskatchewan government news release Sep 14, 2026 (also Bell / CNW coverage Sep 15). Non-binding MOU — expansion subject to customer commitments, commercial agreements, permits, and environmental approvals. amountUsd is the reported ~$52.5B / >$50B prospective total capital (campus + tenant compute + power), not a signed lease NPV. 300 MW phase already under construction; this row captures the expansion-to-1.2 GW announcement.

Source

Under construction · GPU services / compute lease (six-year, three tranches)

Up to $13.7B over six years; site planned ~120 MW expandable to ~180 MW; ops targeted early 2027 / Q1 2027

Counterparty

RUM Group (ex-Rumble; Northern Data / Quake AI)

Where

Maysville, Georgia, US (Banks/Jackson counties)

Amount

$13.7B

MW

120

GPUs

—

The Information (Sep 13, 2026) identified Anthropic as the previously unnamed customer in RUM Group's Aug 24, 2026 8-K; neither Anthropic nor RUM had publicly confirmed at TECHi's publish time. Commercial agreement signed Aug 23 for GPU access/services at Maysville in three ~equal tranches (~$13.7B total); third tranche contingent on customer approval of delivery date. Customer also gets warrant for up to ~50.8M RUM Class A shares at $0.01. RUM disclosed it does not yet have financing to build the facility/buy GPUs. MW/acres from Northern Data's Dec 2024 Maysville campus plan (120 MW near-term, expand to 180 MW, 63 acres) as cited by DCD; do not invent GPU count.

Source

Under construction · AI data center campus redesign (5 GW plan redistributed)

5 GW overall plan under redesign; first phase Stargate UAE cited as $30B / 1 GW cluster with first 200 MW due in 2026

Counterparty

OpenAI; Oracle; Nvidia; SoftBank; Cisco (U.S. partners cited)

Where

United Arab Emirates (originally single ~10 sq mi Abu Dhabi campus near Al Dhafra; now likely multi-site network)

Amount

$30B

MW

200

GPUs

—

Reuters exclusive Sep 11, 2026: after Iranian attacks, UAE revising 5 GW AI campus from one Abu Dhabi mega-campus toward a distributed network with hardening (air defenses, possible underground). G42 says work progressing as planned and specifics under continuous review. amountUsd/mw fields reflect the cited first-phase Stargate UAE cluster ($30B / 1 GW / first 200 MW this year), not a newly signed lease. Not a new offtake; redesign of an existing buildout.

Source

Announced · AI factory / data center capacity partnerships (DSX platform)

Up to 2 GW of AI-related capacity by 2027 (partners' sites; Australia's current computing capacity cited at 1.6 GW)

Counterparty

Firmus, Sharon AI, IREN, Megaport, ResetData, CDC, NEXTDC, AirTrunk

Where

Australia

Amount

—

MW

—

GPUs

—

Reuters Sep 10, 2026: Nvidia working with eight Australian cloud/data-centre firms targeting up to 2 GW AI capacity by 2027 using Nvidia DSX. Partner list also reported by The Next Web. No capital commitment, per-site timetable, named end-customer offtake, or generation source disclosed. Not a single lease; ecosystem buildout target.

Source

Under construction · AI data center campus (phase-one site work / worksite permit)

$4B phase one on >1,100 acres; critical IT load and anchor tenant not disclosed; phase-one completion targeted mid-2028

Counterparty

Cameron County (worksite permit); AEP Texas (transmission); East Rio Hondo Water Supply Corporation (water); anchor tenant not disclosed

Where

Unincorporated Cameron County, Texas, US (near Rio Hondo / Valley International Airport)

Amount

$4B

MW

—

GPUs

—

Cameron County worksite permit issued Sep 10, 2026 (clearing/grading/utilities only; building permits still required). Fluidstack and local coverage describe a $4B first phase on more than 1,100 acres with mid-2028 phase-one target and closed-loop cooling. Do not invent MW: Fluidstack has not published critical IT load. No lease, Google backstop, or named offtaker disclosed for this campus.

Source

Announced · AI/data center infrastructure investment + nuclear/wind power offtake

At least €13B ($15.1B) for 2027–2028; 22-year PPA for up to 50% of Loviisa nuclear output; 629 MW new-to-grid onshore wind supported; 94 MW battery near Kajaani (late 2027). Campus IT load MW undisclosed.

Counterparty

Fortum (Loviisa nuclear PPA); Valorem and Suomen Hyötytuuli (onshore wind PPAs); Fingrid partnership

Where

Finland (Hamina expansion; new sites in Kajaani, Muhos, Vaala; Loviisa nuclear)

Amount

$15.1B

MW

—

GPUs

—

Google Sep 9, 2026 press + Reuters: largest single Google investment in Europe. Three new northern data centers plus Hamina expansion. First Google nuclear PPA outside the U.S. Fortum: PPA supports Loviisa life extension through 2050. MoU with Fortum to explore new nuclear/flexibility. Wind total 629 MW is Google-supported new-to-grid capacity, not data-center IT load. Battery is 94 MW grid flexibility near Kajaani.

Source

Announced · AI compute offtake / anchor customer (multi-year)

Dedicated capacity from two Malaysian AI Factory sites; Firmus total contracted capacity across all customers now >900 MW (OpenAI share undisclosed)

Counterparty

Firmus (Nvidia-backed AI infrastructure)

Where

Malaysia (two Firmus AI Factory sites)

Amount

—

MW

—

GPUs

—

Firmus + OpenAI multi-year partnership Sep 8, 2026: OpenAI becomes anchor customer contracting dedicated AI compute from two Firmus AI Factory sites in Malaysia, with Vera Rubin / NVIDIA DSX planned. Firmus said the deal lifted its total contracted capacity across all customers to more than 900 MW; it declined to disclose OpenAI contract value or OpenAI-specific MW. Five Firmus sites (incl. Malaysia) are under development; two operational in Australia and Singapore. Also covered on Firmus newsroom.

Source

Announced · AI data center campus (build / capacity announcement)

Up to 1 GW AI campus; 264 acres; up to Rs 70,000 crore (~$7.41B) investment

Counterparty

Government of Telangana; TPG (strategic partner); other unnamed partners

Where

Hyderabad / Future City, Telangana, India

Amount

$7.4B

MW

—

GPUs

—

Reuters Sep 5, 2026 (and TCS HyperVault press): HyperVault + partners to invest up to 700 billion rupees (~$7.41B) for a phased up-to-1GW AI campus on 264 acres in Hyderabad. Aimed at AI companies and hyperscalers; no named anchor tenant or signed PPA in the announcement. CM set a June 2, 2028 inauguration target (state Formation Day).

Source

Announced · AI cloud compute offtake (GPU clusters)

~$13B five-year cloud/GPU commitment for training and inference; MW and GPU counts not disclosed

Counterparty

Crusoe

Where

Crusoe cloud (sites undisclosed in reporting)

Amount

$13B

MW

—

GPUs

—

Bloomberg Sep 3, 2026 (people familiar); neither company confirmed on the record. Featured in Sep 4 AI Today morning brief. Jane Street also previously committed ~$6B to CoreWeave (not double-counted here).

Source

Announced · AI cloud compute offtake / preferred-provider partnership

Initial $3.5B compute commitment; up to 100,000 Nvidia Vera Rubin GPUs; initial deployment targeted H2 2027; intent to scale past $6B

Counterparty

Nscale (UK neocloud); Nvidia Vera Rubin platform

Where

Barstow, Texas, US

Amount

$3.5B

MW

—

GPUs

100,000

Joint company announcements Sep 3, 2026 (Figure + Nscale). GPU figure is a ceiling ('up to 100,000'); MW not stated. Nscale also takes an undisclosed stake in Figure and becomes preferred compute provider.

Source

Announced · AI server capacity reservation / backlog

$95B AI backlog; $60.9B AI server orders in fiscal Q2

Counterparty

AI server customers (6,500+ AI customers cited)

Where

Global (Round Rock, Texas HQ)

Amount

$95B

MW

—

GPUs

—

Earnings print, not a single bilateral lease: record $60.9B AI server orders in quarter ended July 31; $16.4B AI-optimized server revenue; exited with $95B AI backlog. Labeled as capacity reservation / server backlog, not MW or site power.

Source

Announced · Data center / power lease

~350 MW Texas campus; $35B cloud-computing commitment

Counterparty

Lambda (Nvidia-backed); Hut 8 developing campus; Nvidia reportedly holds lease

Where

Nueces County, Texas, US

Amount

$35B

MW

350

GPUs

—

Reuters (source) and WSJ first report: Anthropic signed a $35B cloud deal with Lambda for a Hut 8 Texas DC at about 350 MW; WSJ said Nvidia would hold the lease. Parties did not immediately comment.

Source

Under construction · Power-flexible AI factory / data center

Nearly 100 MW Vera Rubin power-flexible AI factory

Counterparty

Digital Realty; Nvidia; tested with EPRI, Dominion, PJM

Where

Manassas, Virginia, US

Amount

—

MW

100

GPUs

—

Company announcement via VentureBeat (Aug 25 round context; Closing Time Aug 28): Manassas project with Digital Realty and Nvidia, nearly 100 MW, slated later this year. Equity round ($150M Series A) is separate; this row is the site/power project.

Source

Announced · GPU lease (neocloud financing)

~$1B short-dated private debt to buy Nvidia GPUs Microsoft will lease

Counterparty

Lambda (Nvidia-backed); JPMorgan-arranged debt

Where

Undisclosed

Amount

$1B

MW

—

GPUs

—

Bloomberg / TechCrunch: Lambda raised about $1B short-dated private debt arranged by JPMorgan to buy Nvidia GPUs that Microsoft will lease; separate ~$926M term loan that week. GPU count not printed.

Source

Announced · GPU capacity expansion

2 million Nvidia GPUs across 2027–2028 (on top of 1M agreed March)

Counterparty

Nvidia

Where

AWS regions (undisclosed)

Amount

—

MW

—

GPUs

2,000,000

TechCrunch on Nvidia earnings call: AWS will add two million Nvidia GPUs across 2027 and 2028, atop one million agreed in March. Terms undisclosed.

Source

Announced · AI cloud compute rental

~460 MW; Nvidia Vera Rubin; ~6-year commitment; start ~end-2027

Counterparty

Nscale (UK neocloud)

Where

West Virginia, US

Amount

$45B

MW

460

GPUs

—

Bloomberg first, CNBC confirmed: ~$45B to rent capacity at Nscale’s flagship West Virginia development; ~460 MW; Vera Rubin hardware; people said six years and end-of-2027 start.

Source

Announced · GPU systems order / gigawatt compute plan

9,000 Nvidia Vera Rubin systems; $8B plan for ~1 GW compute

Counterparty

Nvidia (systems order); Greenko-backed

Where

Southern India

Amount

$8B

MW

—

GPUs

9,000

Bloomberg: Greenko-backed AM Intelligence placed a binding order for 9,000 Vera Rubin systems under an $8B plan to stand up a gigawatt of compute; servers slated next year in southern India.

Source

Announced · Power / site investment (Stargate campus)

Stargate campus power landlord stake; dollar size only described as billions

Counterparty

Lancium (Stargate power developer); OpenAI explored stake then walked

Where

Abilene, Texas, US

Amount

—

MW

—

GPUs

—

The Information: OpenAI explored investing in or acquiring Lancium (power developer behind Stargate in Abilene) last quarter before Nvidia put billions into the same firm. Exact check size beyond 'billions' not printed.

Source

Announced · Power-and-site developer partnership / minority stake

Site, power, cooling, compute decisions via Nvidia DSX; stake size undisclosed

Counterparty

Cloverleaf Infrastructure

Where

US (utilities ↔ data-center developers)

Amount

—

MW

—

GPUs

—

TechCrunch: Nvidia partners with Cloverleaf (2024 power-and-site developer that raised $300M that year) and takes a minority stake; size undisclosed (other reports: hundreds of millions). Extends Nvidia’s infrastructure bets after $1.5B SB Energy stake tied to OpenAI’s Ohio campus.

Source

Under construction · Multi-campus data center investment / lease campuses

Three campuses; planned investment raised to $18B (from $12B in Feb); MW not disclosed in Amazon release

Counterparty

STACK Infrastructure (developer/owner cited in coverage); SWEPCO (power)

Where

Caddo & Bossier Parishes / Shreveport, Louisiana, US

Amount

$18B

MW

—

GPUs

—

Amazon Aug 18, 2026 update of Feb $12B plan: third Shreveport campus brings NW Louisiana planned investment to $18B. No IT-load MW in Amazon’s own release.

Source

Under construction · Data center lease / AI factory campus (PORTS-Pike)

~8 GW-IT under 20-year lease; initial 4.25 IT-GW with Nvidia credit support; first ~800 MW targeted 2028

Counterparty

SB Energy (SoftBank-controlled); Nvidia (exclusive compute + credit support)

Where

Pike County, Ohio, US (PORTS-Pike / former Portsmouth site)

Amount

—

MW

—

GPUs

—

OpenAI Aug 17, 2026: ~8 GW-IT at PORTS-Pike; SB Energy builds/owns/operates under 20-year lease; Nvidia exclusive AI compute, $1.5B SB Energy investment, credit support on initial 4.25 IT-GW (SEC residual-value guarantee capped at $105B cited elsewhere — not campus lease rent). First 800 MW expected 2028. Later Daily/CNBC covered SB Energy S-1 dependency on OpenAI as tenant.

Source

Announced · Data center expansion / contracted IT power

360 MW contracted IT power across three facilities; expected online 2028

Counterparty

Indonesia facilities (own-and-operate; three sites)

Where

Indonesia (Asia-Pacific; first CoreWeave APAC DCs)

Amount

—

MW

360

GPUs

—

Reuters / CoreWeave Aug 4, 2026: three Indonesia facilities totaling 360 MW contracted IT power, own-and-operate, targeted 2028. Capex not disclosed in the Reuters brief.

Source

Announced · AI/HPC campus on former enrichment site + dedicated power

Up to ~1.8 GW utility / >1.2 GW compute stated; $100B investment figure in coalition release

Counterparty

NextEra Energy (dedicated generation); U.S. DOE Paducah Site

Where

McCracken County, Kentucky, US (Paducah American Energy Hub)

Amount

$100B

MW

1200

GPUs

—

Coalition Jul 29, 2026 via Axis: Brookfield+NextEra on DOE Paducah Site. DOE Aug 4 account differs on completion year (2031 vs 2032) and phrasing of capacity. $100B is stated investment, not a signed tenant offtake. PSC approval still required for power service agreement.

Source

Under construction · Data center JV / lease of entire campus

1 GW compute capacity campus; ~$14B development costs; Meta leases entire campus; online beginning ~2028

Counterparty

BlackRock-managed funds (80% venture interest); Meta 20%

Where

El Paso, Texas, US

Amount

$14B

MW

—

GPUs

—

Meta/BlackRock Jul 28, 2026: 80/20 venture; ~$14B development costs; 1 GW compute; Meta sole initial occupant via lease. RVG aggregate ~$13B cited in Meta release.

Source

Announced · Self-build AI data center campus + power contract (Project Camellia)

3.2 GW contracted power phased 2028–2032; county framed ~$20B minimum investment

Counterparty

Georgia Power (power offtake); self-develop campus

Where

Effingham County, Georgia, US (Savannah Gateway Industrial Hub)

Amount

$20B

MW

—

GPUs

—

OpenAI Jul 22, 2026: designing/developing Camellia; 3.2 GW from Georgia Power phased 2028–2032. $20B is the county/incentive framing (Bloomberg cited >$30B at full build — not used as amountUsd). Acres (~1,400) from local coverage of four-building campus. Power figure is contracted utility delivery, not disclosed critical IT load.

Source

Under construction · Data center lease (15-year phases; campus commercialization)

704 MW IT across two 352 MW leases; 1 GW utility campus; $19.6B base-term contract value

Counterparty

Hut 8 (Beacon Point campus)

Where

Nueces County, Texas, US (Beacon Point)

Amount

$19.6B

MW

704

GPUs

—

Hut 8 Jul 20, 2026: second 15-year $9.8B / 352 MW lease doubles same IG tenant to 704 MW IT; campus base-term $19.6B (renewals could reach ~$50.2B). FT later named Nvidia as tenant; Hut 8 release did not name the counterparty. Related but distinct from Anthropic’s later Lambda cloud offtake on the same campus.

Source

Announced · Data center lease (20-year NNN)

175 MW critical IT; ~$6.6B contracted revenue initial term (up to ~$11.6B with extensions); deliveries Q4 2027

Counterparty

CleanSpark (Sandersville campus)

Where

Sandersville, Georgia, US

Amount

$6.6B

MW

175

GPUs

—

CleanSpark Jul 14, 2026 announced 20-year NNN lease with unnamed high-investment-grade tenant ($6.6B initial term; up to ~$11.6B with extensions; rent from ~Nov 30, 2027). Sep 17, 2026 disclosure in investor materials/SEC context: tenant Anviran LLC is a wholly owned Meta subsidiary; Meta guarantees rent and opex. Concurrent $2.227B senior secured notes plan (CSDC Finance I) to fund remaining construction — offering not closed. LOI/exclusivity over CleanSpark Texas portfolio (up to 885 MW) noted but not counted as a separate closed lease. Meta reveal coverage: https://thedeepdive.ca/cleanspark-reveals-meta-behind-6-6-billion-ai-lease-requires-2-23-billion-financing/

Source

Under construction · Hyperscaler AI campus expansion (self-build / JV)

Expanded to 5 GW compute capacity; >$50B regional investment; ~2 GW targeted by 2030, full ~2032

Counterparty

Blue Owl / Beignet Investors JV; Entergy Louisiana (power)

Where

Richland Parish, Louisiana, US (Hyperion)

Amount

$50B

MW

—

GPUs

—

CNBC on Meta Jul 13, 2026: Hyperion raised from prior ~2 GW / $27B JV framing to 5 GW and >$50B. Meta did not separate critical IT load from site power in the announcement; amount is investment figure, not a bilateral lease rent.

Source

Under construction · AI-optimized data center campus (self-build)

1 GW AI-optimized campus; >CAD $13B investment (USD equivalent not converted here)

Counterparty

Self-build; Alberta grid partners (Greenlight LP, AltaLink, Capital Power, AESO cited)

Where

Sturgeon County, Alberta, Canada

Amount

—

MW

—

GPUs

—

Meta Jul 8, 2026: groundbreaking on 1GW AI-optimized DC in Sturgeon County, Alberta — first Canada site; >CAD $13B investment. amountUsd null (CAD not converted). IT vs gross capacity not separated.

Source

Announced · Data center lease (20-year)

~401 MW critical IT load; phased H2 2027 → full early 2028; ~$19B contracted lease revenue

Counterparty

TeraWulf (Raylan Data LLC)

Where

Hawesville / Hancock County, Kentucky, US (Justified Data campus)

Amount

$19B

MW

401

GPUs

—

TeraWulf Jul 6, 2026 press / 8-K: 20-year lease for ~401 MW critical IT; ~$19B contracted revenue over initial term; IG credit support expected. Separate Abernathy JV sale to Fluidstack-led group is not this row.

Source

Live · AI data center campus (Fairwater) — first facility live

First Fairwater facility fully operational; MW undisclosed; second facility targeted 2028; Microsoft cites ~$4.7B statewide WI hyperscale construction spend 2024–2028 (not used as deal amount)

Counterparty

Self-build (Village of Mount Pleasant / Racine County)

Where

Mount Pleasant, Wisconsin, US

Amount

—

MW

—

GPUs

—

Microsoft Jun 23, 2026: first Mount Pleasant Fairwater DC complete/operational after April startup. No IT-load MW in release. Dollar figure in coverage is statewide construction-spend estimate, so amountUsd is null. Acres (315) from campus plan coverage of original site.

Source

Under construction · Data center campus expansion

$1.5B investment across 2026–2027; site MW undisclosed (300 MW cited as contracted regional generation, not campus IT load)

Counterparty

Self-build expansion; TVA region power partners

Where

Jackson County, Alabama, US

Amount

$1.5B

MW

—

GPUs

—

Google Jun 15, 2026: $1.5B investment across 2026–2027 to expand Jackson County campus (former Widows Creek coal plant). Critical IT load not disclosed; >300 MW cited as contracted regional generation for TVA area, not campus IT load.

Source

Live · Cloud / GPU offtake (Colossus I & II)

~$1.25B/month through May 2029 (~$15B/yr); ~325,000 Nvidia GPUs cited in amended S-1 coverage

Counterparty

SpaceX (xAI Colossus campuses)

Where

Memphis, Tennessee, US (Colossus / Colossus II)

Amount

—

MW

—

GPUs

325,000

SpaceX S-1 (May 2026) disclosed Cloud Services Agreements with Anthropic for Colossus/Colossus II at $1.25B/month through May 2029; either party may terminate on 90 days’ notice after ramp. GPU count ~325k from amended S-1 coverage; MW not reliably stated as a single leased figure. amountUsd left null (term depends on duration/termination).

Source

Under construction · AI-optimized data center campus (self-build)

AI-optimized campus; >$1B regional investment; MW not stated in release

Counterparty

Self-build

Where

Tulsa, Oklahoma, US

Amount

$1B

MW

—

GPUs

—

Meta Apr 21, 2026 (page also shows Apr 20): first Oklahoma AI-optimized DC; >$1B regional investment. Campus MW not disclosed.

Source

Announced · Cloud compute offtake / capacity envelope

Up to 5 GW new capacity; >$100B AWS tech spend over 10 years; ~1 GW Trainium2+3 targeted by end-2026

Counterparty

Amazon Web Services (Trainium / Graviton)

Where

AWS regions (incl. Project Rainier / Indiana cited in secondary coverage; sites not fully enumerated in Anthropic release)

Amount

$100B

MW

—

GPUs

—

Anthropic Apr 20, 2026 primary: up to 5 GW and >$100B over ten years on AWS technologies (Trainium2–4, Graviton). Not a named single-site lease; Amazon equity ($5B + up to $20B more) is separate from the offtake spend figure used in amountUsd.

Source

Announced · AI cloud capacity offtake (multi-year)

~$21B expanded agreement through Dec 2032; Vera Rubin deployments; prior relationship brings combined ~$35B cited in coverage

Counterparty

CoreWeave

Where

Multiple CoreWeave locations (undisclosed)

Amount

$21B

MW

—

GPUs

—

CoreWeave Apr 9, 2026: expanded long-term Meta agreement ~$21B through Dec 2032 for dedicated AI cloud capacity across multiple sites, including early Vera Rubin. MW/GPU counts not in the company release.

Source

Announced · GPU capacity expansion / reservation

>1 million Nvidia GPUs starting 2026 (Blackwell/Rubin architectures cited)

Counterparty

Nvidia

Where

AWS regions (global)

Amount

—

MW

—

GPUs

1,000,000

AWS blog at Nvidia GTC 2026: deploy more than 1M Nvidia GPUs across AWS regions starting 2026. announced date set to mid-March GTC window (exact keynote day not pinned in the blog slug). Aug 2026 +2M expansion tracked separately as aws-nvidia-gpus-2027-2028.

Source

Under construction · AI / hyperscale data center campus (self-build)

Designed for 1 GW capacity; >$10B investment cited

Counterparty

Self-build

Where

Lebanon, Indiana, US (Boone County)

Amount

$10B

MW

—

GPUs

—

Meta Feb 11, 2026: groundbreaking on 1GW Lebanon, Indiana campus; investment over $10B. IT vs gross capacity not separated.

Source