Open database of large AI data centers, using satellite and permit data (epoch.ai)

An open, public database and interactive map has been released cataloguing large AI/ML data center sites by combining high-resolution satellite imagery, building and electrical permits, and corporate disclosures. The project links visual footprints (Airbus imagery delivered via Apollo Mapping) with official permit records and public filings to derive structured data about facility size, visible infrastructure (cooling plants, substations, expansion pads) and declared electrical capacity. Map exploration is currently available on desktop only. For the AI/ML community this matters because it creates transparent, reproducible inputs for estimating compute capacity, power draw, and geographic concentration of training infrastructure—useful for researchers modeling carbon footprint, performance scaling, supply-chain dependencies and regional grid impacts. Technically, the team fuses remote sensing with permit metadata and public statements to infer likely IT load and peak electrical capacity, enabling comparative analyses across sites while documenting methods and limitations. The open dataset accelerates empirical study and policy oversight but also raises operational-security and privacy considerations, so users should weigh reuse policies and responsible disclosure when leveraging the resource.