Open-Source Mapping Reveals US Datacenter Expansion Trends

A recent investigation by the non-profit research organization Epoch AI has unveiled the rapid growth of datacenter infrastructure across the United States. Utilizing open-source information and satellite imagery, the team is mapping facilities that frequently escape public attention. This research highlights a sector that is expanding quickly, often outpacing community awareness and regulatory scrutiny.

Understanding the extent of this infrastructure is crucial, as datacenters require significant resources, including electricity and water. Communities often learn about these developments only after construction has commenced. Epoch AI’s interactive map allows users to visualize these sites, linking to satellite images and detailed project information. For instance, a marker over New Albany, Ohio, identifies Meta’s “Prometheus” datacenter complex, which has incurred costs of approximately $18 billion and demands 691 megawatts of power.

Mapping the Future of AI Infrastructure

The construction of datacenters has become a contentious issue nationally, particularly due to their substantial energy needs. Epoch AI’s mapping project provides unprecedented visibility into this expanding industry. Each marker on the map correlates with specific facilities, enabling users to explore their development over time. The timeline feature displays the growth of these complexes, showing how new buildings and cooling systems are added to accommodate rising demand.

Epoch AI’s research places a strong emphasis on the cooling mechanisms employed by these facilities. Modern AI systems generate considerable heat, necessitating effective cooling solutions. The team assesses the cooling infrastructure by counting fans, measuring their dimensions, and analyzing their configurations. This information feeds into a custom model that estimates energy consumption and subsequent construction costs. Senior researcher Jean-Stanislas Denain noted, “We focus on cooling because it’s a very useful clue for figuring out the power consumption.” However, the estimates carry inherent uncertainty, as variations in fan speed and arrangement can lead to significant discrepancies.

Challenges and Future Prospects

Despite the detailed mapping efforts, the dataset remains incomplete. Variations in state and local disclosure laws, combined with the tendency for some projects to operate under the radar, mean that many smaller facilities go untracked. Epoch AI estimates that the current dataset represents about 15 percent of the global AI compute capabilities delivered by chipmakers as of November 2025.

Markers on the map extend across the country, with one near Memphis, Tennessee, pointing to xAI’s Colossus 2 project. According to Epoch AI, the company has installed natural gas turbines across the Mississippi border, presumably to expedite the approval process for the project. They estimate that approximately 110,000 NVIDIA GB200 GPUs are currently operational at this site.

Despite the advancements in mapping, significant gaps remain. Epoch AI acknowledges that even with a comprehensive analysis of a datacenter, it may still lack insight into who utilizes the facility and the extent of their usage. Looking ahead, Epoch AI intends to broaden its focus globally, aiming to illuminate the critical infrastructure that will shape future economies, all too often without public oversight.