The physical buildout behind the AI ecosystem: gigawatt-class data centers, nuclear power deals, hyperscaler campuses, and the long-dated compute commitments backing them.
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.
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.
Meta's largest single data center campus, roughly 5 million square feet on completion. Entergy Louisiana is the serving utility and has filed multi-gigawatt expansion plans tied to the load. Construction underway across phased buildings.
Operational since late 2024 with an initial 100,000 Nvidia H100 build, expanded to roughly 200,000 GPUs through 2025. Local controversy over the temporary methane gas turbines and downstream air-quality concerns in nearby neighborhoods. Memphis Light, Gas and Water is the serving utility.
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. NRC license renewal in progress.
AWS acquired the Cumulus data center campus adjacent to Talen Energy's Susquehanna nuclear plant for $650 million. Initial direct-feed agreement of 480 MW with provisions to scale to 960 MW. FERC blocked an amended interconnection in late 2024, sending the parties back to renegotiate the grid-bypass structure.
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.
Five-year, $200 billion compute commitment from Anthropic for Google Cloud capacity and Broadcom-built TPU chips, reported on May 5, 2026. The buildout is a derived AI-infrastructure project rather than a single named campus; the gigawatt-scale capacity backing it begins arriving in 2027 across Google Cloud regions. See the TF original analysis for the math.
Sits on the former Foxconn site. Microsoft announced an additional $3.3 billion commitment in 2024 on top of earlier phases. We Energies has separately filed for approval of two new natural gas peaker plants partly attributed to the data center load.
Meta's New Albany campus expansion, codenamed Prometheus in some internal references. Sits inside the broader AEP Ohio service area which has filed multiple data center load adjustments. Smaller than Hyperion but earlier to operational status.
Apple's flagship east-coast data center, expanded multiple times to support Apple Intelligence and the M-series silicon trained at AWS plus Apple's own private cloud compute. Paired with Apple-owned solar farms in Catawba and Conover.
AI-specialized GPU cloud, IPO on Nasdaq in March 2025 (CRWV). Operates 32+ data centers as of 2026 with hundreds of thousands of Nvidia GPUs (H100, H200, GB200) deployed. Microsoft is the anchor customer, accounting for roughly half of revenue. Nvidia holds a meaningful equity stake. Heavy spend on multi-year capacity commitments tied to Microsoft and OpenAI deals.
Originally Iris Energy, Australian-founded Bitcoin miner that pivoted to AI compute. Nvidia announced a $2.1 billion equity stake in 2026 covering five gigawatts of DSX-branded rack deployments over the next several years. Renewable-power-focused, with sites in Childress and West Texas and British Columbia. Nasdaq ticker IREN.
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. Vega TX site is the flagship new build.
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.
Multiple separately-reported feasibility programs into placing AI training or inference capacity in low Earth orbit. Anthropic and SpaceX have publicly discussed orbital extensions of Colossus-class compute; 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.
/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.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