Researchers Reveal Hidden AI Datacenters Across America

A recent investigation by 404 Media has unveiled a significant initiative aimed at mapping the rapid growth of datacenters throughout the United States. Led by Epoch AI, a non-profit research institute focused on artificial intelligence, this project utilizes open-source intelligence and satellite imagery to track facilities that often remain overlooked in public discussions. The research team has developed interactive maps detailing costs, ownership, and power consumption of these expanding infrastructures.

Mapping the Expansion of Datacenters

The construction of datacenters has emerged as a contentious issue nationwide. These facilities require substantial amounts of electricity and water, often leaving local communities unaware of their existence until construction is underway. Epoch AI’s interactive map features visual markers identifying known sites, linking users to satellite images and specific project details. For instance, one marker highlights Meta’s “Prometheus” datacenter complex in New Albany, Ohio, which Epoch AI estimates has incurred costs of approximately $18 billion and draws 691 megawatts of power.

The map provides a timeline that allows users to observe the complex’s growth, with satellite images illustrating the addition of new buildings and cooling systems over time. This level of detail offers insights into an industry that has been expanding rapidly, often without sufficient public scrutiny.

Understanding Power Consumption Estimates

Epoch AI’s research places particular emphasis on the cooling systems utilized within these datacenters. Modern AI technologies generate significant heat, necessitating advanced cooling solutions. According to Epoch AI, “Modern AI datacenters generate so much heat that the cooling equipment extends outside the buildings.” This approach involves counting fans, measuring their size, and analyzing their placement, which contributes to a custom model estimating energy use and, consequently, overall compute capacity and construction costs.

Jean-Stanislas Denain, a senior researcher at Epoch AI, explained, “We focus on cooling because it’s a very useful clue for figuring out the power consumption.” Despite the thoroughness of the analysis, Epoch AI acknowledges the model’s limitations, indicating that actual cooling capacity could significantly differ from their estimates.

The current dataset compiled by Epoch AI is incomplete, as state and local disclosure laws vary, allowing some projects to remain undisclosed. Smaller facilities often evade detection altogether. Epoch AI estimates that the data set currently represents around 15 percent of global AI compute capacity delivered by chipmakers as of November 2025.

Further investigations reveal markers across various 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 to expedite approvals. Epoch AI noted that, based on earlier statements from Elon Musk, approximately 110,000 NVIDIA GB200 GPUs are operational as part of this project.

Even with detailed mapping efforts, there remain significant gaps in understanding the infrastructure’s usage. Epoch AI remarked, “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.” Looking ahead, Epoch AI plans to broaden its research globally, aiming to illuminate the infrastructure that shapes the future economy, often without public visibility.