A recent report by 404 Media reveals that a small research team has been diligently mapping the rapid expansion of artificial intelligence (AI) datacenters across the United States. Utilizing publicly available information and satellite imagery, the non-profit research institute Epoch AI aims to shed light on the scale and pace of AI infrastructure development that often goes unnoticed in public discussions.
The team employs open-source intelligence to identify, analyze, and document datacenters, creating an interactive map that estimates critical factors like cost, ownership, and power consumption. By reviewing satellite images, construction permits, and local regulatory filings, Epoch AI is providing unprecedented visibility into a burgeoning industry that is expanding faster than public scrutiny.
Datacenter Construction Under Scrutiny
The construction of datacenters has emerged as a contentious issue nationwide due to their substantial demands for electricity and water. Many communities are unaware of these facilities until construction is underway. Epoch AI’s mapping project visually marks known sites, linking each marker to satellite views and detailed project information. For instance, a green circle on the map indicates Meta’s “Prometheus” datacenter complex in New Albany, Ohio, which has an estimated construction cost of $18 billion and draws 691 megawatts of power.
Epoch AI describes the Prometheus facility as a combination of weatherproof tents, colocation facilities, and traditional datacenter buildings, illustrating Meta’s strategic shift toward AI. Users can explore a timeline of the complex’s development, observing the addition of new structures and cooling systems through satellite imagery.
Innovative Analysis of Cooling Infrastructure
Epoch AI’s analysis predominantly focuses on cooling infrastructure, a critical aspect since modern AI systems generate significant heat. The datacenters often place cooling units outside buildings or atop roofs. “Modern AI datacenters generate so much heat that the cooling equipment extends outside the buildings,” the organization notes on its website.
The research team meticulously counts fans, measures their dimensions, and analyzes their configurations. These details feed into a custom model designed to estimate energy usage, which in turn informs calculations of computational capacity and construction costs. According to Jean-Stanislas Denain, a senior researcher at Epoch AI, “We focus on cooling because it’s a very useful clue for figuring out the power consumption.”
Nevertheless, the model carries inherent uncertainties. Variations in fan speed and configuration can lead to discrepancies in estimates, with real cooling capacity potentially being twice as high or half as low as projected.
Despite its thoroughness, the mapping remains incomplete. Variations in state and local disclosure laws, along with some projects that intentionally avoid publicity, create challenges. Epoch AI estimates that the dataset currently reflects only about 15 percent of global AI compute capacity delivered by chipmakers as of November 2025.
Markers on the map extend across the United States, including one near Memphis, Tennessee, which points to xAI’s Colossus 2 project. This facility reportedly installed natural gas turbines across the Mississippi border, likely to expedite approval processes. Epoch AI indicates that approximately 110,000 NVIDIA GB200 GPUs are operational at this site, based on previous statements from Elon Musk.
Despite the detailed mapping efforts, significant blind spots remain. “Even if we have a perfect analysis of a datacenter, we may still be in the dark about who uses it and how much they use,” Epoch AI acknowledges. The organization plans to expand its research on a global scale, aiming to illuminate infrastructure that influences the future economy while often remaining hidden from public view.
