Skip to content
All systems operational0 AI providers monitored, polled every 2 minutes
Live status

AI Infrastructure

The physical buildout behind the AI ecosystem: gigawatt-class data centers, nuclear power deals, hyperscaler campuses, and the long-dated compute commitments backing them.

Machine-readable JSON/api/ai-infrastructure/projects.json

We cover the model layer at /pricing, the funding layer at /funding/portfolio, and the live operator state at /status. This page covers the layer underneath all of them. It is the answer to the question where does the AI actually run, and how does it get the power.

Editorial curation. Every entry sourced to a public announcement, regulatory filing, or established trade report. We track the physical buildout and the announced capacity, not the politics. The politics changes faster than the steel.

Projects tracked
24
Operational
5
Construction
12
Announced capacity
30.4 GW

Tracked projects

Stargate Abilene

OpenAI with Oracle, Crusoe Energy, SoftBank, MGX
Construction
Location
Abilene, Texas
Capacity
1.2 GW
Power
natural gas plus grid
Online
2026

First flagship campus of the $500 billion Stargate program, announced from the White House in January 2025. Crusoe Energy is the campus developer; Oracle Cloud is the anchor tenant; OpenAI is the primary compute consumer. Phase one of a multi-site rollout. On its September 10, 2026 earnings call Oracle said 6 of the 8 campus buildings, representing 618 MW and 75% of total capacity, have been delivered to the customer.

$100B capexannounced 2025-01-21

Meta Hyperion

Meta with Blue Owl Capital, Entergy
Construction
Location
Richland Parish, Louisiana
Capacity
5.0 GW
Power
natural gas plus grid
Online
2030

Meta's largest single data center campus. Originally announced December 2024 as a $10 billion independent project; restructured in October 2025 into a $27 billion joint venture with Blue Owl Capital (80% Blue Owl, 20% Meta). On July 13, 2026 Meta said it is expanding the Richland Parish site to 5 GW of compute capacity, the largest in its fleet, representing an investment of more than $50 billion in the region; the site houses Hyperion, Meta's largest multi-gigawatt AI training cluster. Phase 1 targets 1.5 GW by late 2027, using roughly 1.3 million NVIDIA Blackwell GPUs. Entergy Louisiana is the serving utility.

SpaceXAI Colossus

SpaceXAI with Supermicro, Dell, Anthropic (tenant)
Operational
Location
Memphis, Tennessee
Capacity
300 MW
Power
natural gas turbines plus grid
Online
2024

Operational since late 2024, initially with 100,000 NVIDIA H100s, expanding to 555,000+ GPUs across Colossus 1 and 2 by 2026. SpaceX acquired xAI in February 2026 and merged the AI division into SpaceXAI in May 2026. In May 2026, Anthropic leased the entire Colossus 1 cluster to supplement its Google Cloud TPU capacity; Anthropic's own May 6, 2026 announcement confirms over 300 MW of new capacity and more than 220,000 NVIDIA GPUs but states no price or term. The SpaceX IPO filing puts the Anthropic rent at $1.25 billion a month through May 2029 and directly references 130 MW of compute at Colossus and 210 MW at Colossus 2, with a next Colossus 2 phase of at least 220,000 GB300 processors and over 400 MW. Local controversy persists over temporary gas turbines and air quality in nearby neighborhoods.

Three Mile Island Unit 1 Restart

Microsoft with Constellation Energy
Permitted
Location
Londonderry Township, Pennsylvania
Capacity
835 MW
Power
nuclear
Online
2027

Constellation will restart Three Mile Island Unit 1 (renamed Crane Clean Energy Center) under a 20-year power purchase agreement with Microsoft. Unit 1 is the undamaged reactor on the site; Unit 2 was the one that suffered the 1979 partial meltdown and remains permanently shut. In its Q2 2026 results (August 6, 2026) Constellation said FERC granted a waiver to transfer capacity interconnection rights to Crane and the NRC approved a fuel license amendment, moving it closer to restarting operations in 2027.

Amazon Susquehanna Campus

Amazon Web Services with Talen Energy
Construction
Location
Salem Township, Pennsylvania
Capacity
1.9 GW
Power
nuclear
Online
2032

AWS acquired the Cumulus data center campus adjacent to Talen Energy's Susquehanna nuclear plant for $650 million. After FERC blocked an amended co-located interconnection in late 2024, the parties restructured: in June 2025 Talen signed a PPA to supply Amazon with 1,920 MW of carbon-free nuclear power through 2042, ramping to full volume no later than 2032, with the co-located arrangement moving to a front-of-the-meter structure after transmission reconfigurations expected in spring 2026. Amazon tied the deal to a $20 billion Pennsylvania investment.

$650M capexannounced 2024-03-04

Google Council Bluffs Expansion

Google with MidAmerican Energy
Construction
Location
Council Bluffs, Iowa
Capacity
capacity not disclosed
Power
mixed including wind PPA
Online
ongoing

Multi-phase expansion of Google's flagship Iowa cloud region, the heart of its TPU deployment footprint. Most recent announced phase brought cumulative Iowa investment past $7 billion. Google has paired the buildout with multiple wind and solar PPAs through MidAmerican.

$5.4B capexannounced 2023-04-26

Anthropic TPU Capacity Commitment

Anthropic with Google Cloud, Broadcom
Construction
Location
multi-region, Google Cloud regions
Capacity
capacity not disclosed
Power
Google Cloud regional mix
Online
2027

Five-year, $200 billion compute commitment from Anthropic for Google Cloud capacity and Broadcom-built TPU chips, reported on May 5, 2026. Anthropic's own April 6, 2026 announcement confirms an agreement with Google and Broadcom for multiple gigawatts of next-generation TPU capacity coming online starting in 2027, but does not state the dollar figure, which comes from reporting. The buildout is a derived AI-infrastructure project rather than a single named campus. See the TF original analysis for the math.

$200B capexannounced 2026-05-05

Microsoft Mount Pleasant Campus

Microsoft with We Energies
Operational
Location
Mount Pleasant, Wisconsin
Capacity
capacity not disclosed
Power
mixed including new gas plants
Online
2026 (first facility), 2028 (second facility)

Sits on the former Foxconn site and houses Microsoft's Fairwater AI datacenter. Microsoft announced an additional $3.3 billion commitment in 2024 on top of earlier phases. On June 23, 2026 Microsoft said the first facility is fully operational after equipment came online in April; a second facility immediately adjacent is under construction and scheduled for completion in 2028. We Energies has separately filed for approval of two new natural gas peaker plants partly attributed to the data center load.

$3.3B capexannounced 2024-03-08

Meta Prometheus

Meta with AEP Ohio
Construction
Location
New Albany, Ohio
Capacity
1.5 GW
Power
grid
Online
2026

Meta's New Albany campus expansion, codenamed Prometheus. Meta has been on the site since breaking ground in 2017 and states a $1.5 billion plus investment in Ohio, 100 percent clean and renewable energy matching, and 300 plus operational jobs. Mark Zuckerberg announced Prometheus in July 2025 as Meta's first multi gigawatt supercluster coming online in 2026, and press coverage describes it as the first 1 GW class data center; the 1.5 GW figure here is the campus target at completion and is not a Meta published number. Sits inside the broader AEP Ohio service area which has filed multiple data center load adjustments. Smaller than Hyperion but on track for earlier operational status.

$1.5B capexannounced 2024-08-14

Apple Maiden Campus Expansion

Apple with Duke Energy
Operational
Location
Maiden, North Carolina
Capacity
capacity not disclosed
Power
renewable (Apple created regional projects)
Online
ongoing

Apple's flagship east coast data center, serving iCloud, the App Store, Apple Music, iMessage and Apple TV+ for North America. In its August 2025 US investment announcement Apple said construction is underway in Maiden to expand capacity, including for the growth of Apple Intelligence, building on more than $5 billion already invested in Catawba County. The site runs on 100 percent renewable energy from Apple created projects in the region. Apple discloses neither campus capacity nor a capex figure for the expansion, so both fields are null; the prior source link on this entry pointed at an unrelated European renewables release and was replaced.

announced 2025-08-06

CoreWeave AI Cloud Fleet

CoreWeave with Microsoft, Nvidia
Operational
Location
multi-region, 32+ US data centers
Capacity
capacity not disclosed
Power
colocation grid mix
Online
ongoing

AI-specialized GPU cloud, IPO on Nasdaq in March 2025 (CRWV). Operates 49 data centers as of mid-2026. In Q2 2026 results (August 11, 2026) CoreWeave said it expanded active power by nearly 500 MW to 1.5 GW, grew total contracted power to about 3.7 GW, and completed the first bring-up of NVIDIA Vera Rubin NVL72. Microsoft is the anchor customer, accounting for roughly half of revenue. NVIDIA holds a meaningful equity stake ($2 billion investment, January 2026). Heavy spend on multi-year capacity commitments tied to Microsoft and OpenAI deals.

$20B capexannounced 2017-01-01

IREN AI Compute Fleet

IREN with Nvidia, Dell
Construction
Location
multi-region, Childress TX, West Texas, British Columbia
Capacity
2.9 GW
Power
renewables (wind, solar, hydro)
Online
2027

Originally Iris Energy, Australian-founded Bitcoin miner that pivoted to AI compute. In May 2026 NVIDIA and IREN announced a partnership to deploy up to 5 GW of AI infrastructure across IREN's pipeline over time; IREN issued NVIDIA a five-year right to purchase up to 30 million shares at $70, a right to invest up to $2.1 billion (an option, not a completed equity stake or project capex). Future deployments are expected to focus on IREN's 2 GW Sweetwater campus in Texas: Sweetwater 1 (1,400 MW) and Sweetwater 2 (600 MW) were conditionally included as Base Load in the ERCOT Batch Zero process on September 8, 2026, with the Sweetwater 1 substation already energized and 300 MW of data center capacity in construction toward Q4 2027 delivery. IREN describes an announced global development portfolio of more than 5 GW. Renewable-power-focused, with sites in Childress and West Texas and British Columbia. Nasdaq ticker IREN.

Hut 8 AI Hosting Portfolio

Hut 8 with Coatue, Fluidstack, Google (lease backstop), Anthropic (reported)
Construction
Location
multi-region, River Bend LA, Salt Creek TX, Vega TX, North Dakota
Capacity
1.5 GW
Power
natural gas plus grid
Online
2027

Originally a Bitcoin mining operator (Nasdaq HUT), now pivoting infrastructure toward AI hosting. Joint venture with Coatue (Highrise AI) targeting gigawatt-class HPC capacity. Mixed-use sites still run some Bitcoin workloads while new AI capacity comes online. The flagship is now River Bend, Louisiana: in December 2025 Hut 8 signed a 15-year, $7.0 billion lease with Fluidstack for 245 MW of IT capacity, with Google providing a financial backstop for the base term, a right of first offer on up to 1,000 MW more, and the initial data hall scheduled for Q2 2027.

$1.2B capexannounced 2024-06-04

Lambda GPU Cloud Fleet

Lambda Labs with Microsoft, Nvidia
Operational
Location
multi-region, primarily US colocation
Capacity
capacity not disclosed
Power
colocation grid mix
Online
ongoing

Established AI-specialized GPU cloud with H100 and H200 SuperPOD deployments across US colocation partners. Smaller than CoreWeave on raw capacity but well-known among AI research teams and frontier labs running shorter-cycle workloads. Multi-year Microsoft capacity agreement disclosed in 2024.

announced 2012-01-01

Stargate Michigan (The Barn)

OpenAI with Oracle, Walbridge, Related Digital, Blackstone, SoftBank
Construction
Location
Saline, Michigan
Capacity
1.0 GW
Power
grid
Online
2027

Second major Stargate campus after Abilene, Texas. OpenAI broke ground on June 1, 2026 alongside Oracle, Related Digital, and Walbridge in Saline Township, Washtenaw County, describing a 1 GW campus. Related Digital and Blackstone announced financing for the $16 billion campus, purpose-built for Oracle, in April 2026; it comprises three single-story buildings with more than a gigawatt of capacity on 250 acres. Part of the broader $500 billion Stargate program announced from the White House in January 2025.

$16B capexannounced 2026-06-01

Anthropic AI Infrastructure (Fluidstack)

Anthropic with Fluidstack
Construction
Location
Texas and New York (multi-site), TX, NY
Capacity
capacity not disclosed
Power
grid
Online
2026

Anthropic committed $50 billion to American AI infrastructure, with Fluidstack as the build partner across initial sites in Texas and New York. Announced November 12, 2025 alongside an expected 800 permanent jobs and 2,400 construction jobs. Sites are designed as custom data centers optimized for Anthropic workloads and are coming online in phases through 2026. The gigawatt-scale capacity complements Anthropic's Google Cloud TPU commitments and its lease of SpaceXAI Colossus 1.

$50B capexannounced 2025-11-12

PORTS-Pike Technology Campus

OpenAI with SB Energy, NVIDIA, SoftBank, AEP Ohio, U.S. Department of Energy
Announced
Location
Pike County, Ohio
Capacity
8.0 GW
Power
new dedicated generation plus grid
Online
2028-2032

OpenAI agreed to secure approximately 8 GW of IT capacity at the PORTS-Pike Technology Campus on the former Portsmouth Gaseous Diffusion Plant site. SB Energy will build, own, and operate the data center under a 20-year lease to OpenAI. NVIDIA is the exclusive AI compute provider and is providing credit support on land, power, and shell to secure the initial 4.25 IT-GW, with an option on the remaining 3.75 IT-GW, and is investing $1.5 billion in SB Energy. SB Energy and SoftBank plan at least 10 GW of new generation and at least $4.2 billion of grid investment with AEP Ohio. Capacity comes online in phases beginning in 2028 over a six-year buildout through 2032. Capacity figure is IT load; total project capex not disclosed.

Paducah Site AI Data Center Campus

Brookfield with NextEra Energy, Big Rivers Electric Power Corporation, Jackson Purchase Energy Cooperative, Paducah Power System, U.S. Department of Energy
Announced
Location
Paducah, Kentucky
Capacity
1.2 GW
Power
dedicated natural gas plus battery storage
Online
2032

The Department of Energy selected Brookfield to lease land and develop and operate a data center campus at DOE's former enrichment site in Paducah, and selected NextEra Energy to build and own dedicated generation of up to 2 GW of natural gas and up to 2.6 GW of battery storage. The $100 billion privately funded project is planned to support up to 1.8 GW of utility capacity and over 1.2 GW of compute capacity once fully constructed in 2032 (capacity figure is the compute number). The project is subject to negotiation and execution of definitive documentation; no tenant has been named.

$100B capexannounced 2026-07-29

Anthropic Justified Data Campus Lease

Anthropic with TeraWulf
Announced
Location
Hawesville, Kentucky
Capacity
401 MW
Power
grid
Online
2027-2028

TeraWulf signed a 20-year lease with Anthropic for a purpose-built AI infrastructure campus at the Justified Data site in Hawesville, Kentucky, accommodating approximately 401 MW of critical IT load developed in multiple phases. Initial capacity is expected in the second half of 2027, ramping to the full 401 MW by early 2028. TeraWulf expects about $19 billion of contracted lease revenue over the initial term (lease revenue, not capex). Announced alongside TeraWulf's sale of its Abernathy, Texas joint venture stake to Fluidstack.

announced 2026-07-06

Anthropic AMD Helios Deployment

Anthropic with AMD
Announced
Location
multi-site, not disclosed
Capacity
2.0 GW
Power
not disclosed
Online
2027

AMD and Anthropic announced a partnership for Anthropic to deploy up to 2 GW of AMD Instinct MI450 Series GPUs in AMD Helios rack-scale systems, with deployment of the first gigawatt beginning in the first half of 2027. AMD committed to a strategic equity investment of up to $5 billion in Anthropic, and the companies will collaborate on ROCm and workload optimization. Like the TPU entry, this is a derived compute commitment rather than a single named campus; sites were not disclosed.

announced 2026-07-22

Duane Arnold Energy Center Restart

Google with NextEra Energy, Central Iowa Power Cooperative, U.S. Department of Energy
Announced
Location
Palo, Iowa
Capacity
615 MW
Power
nuclear
Online
2029

NextEra Energy plans to restart Iowa's only nuclear plant, shut in 2020, backed by a 25-year agreement for Google to buy power from the 615 MW plant to support its cloud and AI infrastructure in Iowa; Central Iowa Power Cooperative buys the remaining output on the same terms. On September 8, 2026 NextEra and the DOE reached conditional commitment and financial close on a loan of up to $1.9 billion for the restart. The plant is expected back online no later than the first quarter of 2029, pending regulatory approvals.

Meta Sturgeon County Data Centre

Meta with Greenlight Limited Partnership, AltaLink, Capital Power, Alberta Electric System Operator
Construction
Location
Sturgeon County, Alberta
Capacity
1.0 GW
Power
new dedicated generation plus grid
Online
not disclosed

Meta's first data center in Canada and the 33rd in its global fleet: a 1 GW AI-optimized campus representing more than CA$13 billion (roughly US$9 billion), plus about CA$60 million for local infrastructure. Meta broke ground in July 2026, says it is fully funding new Alberta generation and grid infrastructure, and will match electricity use with 100 percent clean and renewable energy; cooling is closed loop and liquid with dry cooling, so there is no operational cooling water use. Roughly 3,000 construction workers at peak and more than 300 operational jobs. CBC reported that the campus will be served by a new natural gas plant built by a consortium that includes Pembina Pipeline. Meta has not published a target operational date.

$9B capexannounced 2026-07-08

HyperVault Hyderabad AI Campus

HyperVault (Tata Consultancy Services) with Government of Telangana
Announced
Location
Hyderabad, Telangana
Capacity
1.0 GW
Power
grid plus green energy contracts
Online
not disclosed

HyperVault, the data center arm of Tata Consultancy Services, secured 264 acres at Hyderabad for an AI campus of up to 1 GW. HyperVault and its partners expect to invest up to 70,000 crore rupees (about $7.4 billion) to build and operate it. The campus targets frontier AI companies and hyperscalers running high density GPU fleets, using direct to chip liquid cooling, larger power blocks and a green energy plus water neutral design, and will be built in phases tied to customer demand. Among the largest AI campuses announced in India. No anchor tenant, construction start or completion date has been announced by the company.

$7.4B capexannounced 2026-09-05

Orbital AI Compute (Concept)

Anthropic, SpaceX, Google (separately exploring) with Starcloud, various
Announced
Location
low Earth orbit, feasibility studies
Capacity
capacity not disclosed
Power
solar (free, continuous, vacuum-radiated cooling)
Online
2030 or later

Multiple separately-reported feasibility programs into placing AI training or inference capacity in low Earth orbit. In its May 6, 2026 compute announcement Anthropic said it has expressed interest in partnering with SpaceX to develop multiple gigawatts of orbital AI compute capacity, which is an expression of interest and not a signed build; Google has separately explored similar concepts via Starcloud-style partners. Power thesis: continuous solar plus vacuum heat-rejection sidesteps the grid and water constraints driving terrestrial buildouts. Capacity, capex, and timeline figures are all aspirational at this stage. Concept-stage entry; we will reclassify when a launch contract or specific construction commitment lands.

announced 2026-05-09

Free agent endpoint

  • /api/ai-infrastructure/projects.jsonFull registry as static JSON. Each entry returns id, name, operator, partners, location, status, capacity_mw_announced, power_source, announced_date, target_operational, capex_usd_billion, context, and primary_sources. License: TF curation CC-BY-4.0; underlying sources retain their original license.
Capacity is announced, not measured. The numbers on this page reflect what operators have publicly committed to. Real delivered capacity tracks announced capacity loosely and tends to slip. Use these figures for relative magnitude and pipeline timing, not as guaranteed power draw at the wall.
Long-arc thesis

When terrestrial runs out of room, the steel goes up.

The four constraints terrestrial AI infrastructure hits (grid, water, permits, NIMBY) all go away in orbit. Continuous solar plus vacuum cooling sidesteps the bottlenecks that are hardening on the ground. The catch is launch cost, and that is the one curve actively bending in the right direction. Concept-stage today; 2030-plus timeline. We track it on this page as it moves from feasibility into engineering.

Read the orbital thesis

Frequently asked questions

What does this page cover?
Major AI buildout projects across four operator categories: hyperscalers selling capacity (Microsoft, Google, Amazon, Meta, Apple), frontier-lab compute farms consuming capacity (Stargate, Colossus, the Anthropic TPU commit), AI-specialized GPU clouds (CoreWeave, Lambda), and Bitcoin-pivot AI hosting (IREN, Hut 8). Plus the nuclear PPAs and restarts powering them and a concept-stage orbital entry. Editorial curation, every entry sourced to public records. Free JSON at /api/ai-infrastructure/projects.json.
Why a separate page for this?
AI infrastructure is the physical substrate of every model, every API call, every agent on this site. We already cover the model layer, the pricing layer, and the funding layer. This is the layer underneath them all. The next two years will be defined by which projects come online on time, which slip, and which utility relationships hold up.
Is xAI a hyperscaler?
Strictly speaking, no. The traditional definition of hyperscaler is a cloud provider operating at massive scale that sells capacity to third parties: AWS, Microsoft Azure, Google Cloud, Oracle Cloud, plus arguably Meta and Apple operating hyperscale infrastructure for internal use. xAI Colossus today is a single-tenant frontier-lab compute farm: xAI builds it to train Grok, not to rent capacity. But the buzz term has expanded. Reporters and analysts increasingly use "hyperscaler" to mean "any operator running compute at hyperscale", which includes xAI. Elon has also signaled that future Colossus generations may open multi-tenant capacity, which would make the strict definition apply too. We track xAI Colossus as a frontier-lab compute farm in this registry, but flagging the terminology drift here so the search-term traffic finds the right entry.
How are projects added?
Editorial cadence. An entry is added when an authoritative source (company announcement, regulatory filing, utility commission docket, or established trade reporting) confirms a project at the gigawatt class or with an unusual structural feature (nuclear PPA, dedicated build, grid-bypass). We would rather ship 10 well-sourced entries than 100 stale ones.
Why no opinions on the politics?
Because the politics changes faster than the steel and concrete. We track the physical buildout, the power deals, the timelines, and the announced capacity. Reasonable people disagree on whether a 2 GW data center is good or bad for a community; reasonable people do not disagree on whether it is being built. We stay on the latter.
How does this connect to the rest of TensorFeed?
Cross-linked with /funding/portfolio (the capital flowing into AI infrastructure), /pricing (the model prices these data centers serve), and /status (the live state of the operators that run them). Together they form a closed-loop view of the AI ecosystem from capital to silicon to inference.
Will you cover environmental impact?
Yes, factually. Water draws, grid strain, peaker plant filings, emissions, and community pushback all appear in the context paragraph of relevant entries with their primary sources. Pros and cons both get reported. No endorsements, no advocacy, just sourced facts.
Last reviewed: reviewed weeklyNext review:

New projects added when announced; updates when status (announced/permitted/construction/operational) changes.