New open-source map charts the scale of US AI datacenters buildout

A non-profit research institute, Epoch AI, is using open-source intelligence—including satellite imagery, construction permits, and regulatory filings—to map the rapid expansion of AI datacenters across the United States. Their interactive map provides detailed estimates on cost, ownership, and power consumption of these facilities, offering rare transparency into an industry growing faster than public oversight. For example, Epoch AI highlights Meta’s “Prometheus” datacenter in New Albany, Ohio, estimating it has cost $18 billion and consumes 691 megawatts of power, reflecting Meta’s strategic pivot toward AI infrastructure. Epoch AI’s methodology centers on analyzing cooling infrastructure visible in satellite images, as modern AI datacenters generate extreme heat requiring extensive external cooling units. By counting and measuring fans and cooling systems, they estimate energy use, which in turn informs compute capacity and construction cost estimates. However, these estimates carry uncertainty due to variable fan configurations and speeds. The map currently covers about 15% of global AI compute capacity as of November