AI data center sites

234 data center and power sites in 45 countries, with local water stress, grid carbon intensity and recorded community objections for each. 69 are in basins rated high or extremely high for water stress.

Sites by status

StatusSitesCapacity
Operational1679.0 GW
Partly operational65.8 GW
Halted240 MW
Under construction2817.4 GW
Announced244.7 GW
Cancelled7not counted
Total23436.9 GW

Loading map…

Capacity is the full planned size of each site, from company statements, permits or our estimates; 108 of 234 sites have a figure. Operators rarely publish site capacity: 68 of 167 operational sites have a figure, and 76% of the operational total is our estimate (confidence D or E). Where no source gives a figure, capacity is left blank rather than estimated. Partly operational campuses are counted at their planned size, separately from operational sites. Cancelled sites are listed but not counted. Coverage in China and Russia is less complete: operators there publish few site-level figures. Each site lists its sources and a confidence rating. See the methodology for how sites are selected.

Largest sites

The ten largest sites by planned full capacity. Bar color shows status; most of this capacity is not yet built.

  1. Meta · Under construction · Richland Parish, Louisiana, US

  2. Amazon · Partly operational · Northern Indiana, US

  3. OpenAI · Under construction · Doña Ana County, New Mexico, US

  4. OpenAI · Under construction · Texas, US

  5. OpenAI · Partly operational · Texas, US

  6. OpenAI · Under construction · Texas, US

  7. Crusoe · Announced · Texas, US

  8. Google · Under construction · Andhra Pradesh, India

  9. Meta · Under construction · Texas, US

  10. Meta · Under construction · Central Indiana, US

Sites in water-stressed basins

69 of 234 sites, about 5.6 GW of capacity, are in basins that WRI Aqueduct rates high or extremely high for baseline water stress.

Largest eight by capacity shown. Water stress is assigned per site from WRI Aqueduct 4.0; it describes the basin, not the facility's own draw.

Classification

Explicit AI infrastructure
Purpose-built for AI training or inference workloads.
AI-capable hyperscale
General hyperscale data centers with significant AI workloads.
Associated infrastructure
Substations, transmission, and power projects serving AI sites.

Environmental data

Water stress, grid carbon intensity and drought risk are assigned from published sources (WRI Aqueduct, EPA eGRID, IEA, China MEE, Ember). Each value shows its source and confidence tier.

What we do not claim

We do not attribute environmental harm to specific facilities. Community concern flags record events reported in public sources. They do not imply causation. See the methodology for full definitions.