A new initiative by Epoch AI aims to map out America's rapidly expanding AI datacenters using open-source intelligence. By analyzing satellite images, construction permits, and local regulatory data, the researchers have created an interactive map that estimates the costs and power consumption of these facilities. The project highlights how many datacenters are built without community awareness, as local residents often learn about them only after construction starts. For instance, the map features Meta's “Prometheus” datacenter in New Albany, Ohio, which has already cost about $18 billion and consumes 691 megawatts of power. The researchers focus heavily on cooling systems, which are critical for managing the heat generated by AI operations. They assess the size and placement of cooling fans to estimate energy use, which in turn helps gauge compute capacity and overall construction costs. However, their model carries uncertainty, with actual cooling capacity potentially being significantly higher or lower than estimates. Despite the detailed analysis, there are still gaps in visibility as some projects remain undisclosed, and smaller facilities can easily go unnoticed. As of November 2025, Epoch AI estimates their dataset captures only about 15% of the global AI computing capacity delivered by chipmakers. The initiative also exposes the challenges of tracking the specific usage of these datacenters. Even with accurate geographical data, the actual users and their consumption patterns often remain unknown. Epoch AI plans to expand its mapping efforts globally, aiming to provide a clearer picture of the infrastructure that underpins the growing AI economy, which continues to operate largely off the public radar.