A recent investigation by 404 Media reveals how a small, dedicated team at Epoch AI is meticulously mapping the swift growth of datacenter infrastructure across the United States. Utilizing publicly available data and satellite imagery, the researchers are illuminating facilities that often lack visibility in public discussions. Their efforts aim to assess the scale and pace of artificial intelligence (AI) development, shedding light on an industry that expands rapidly, often outpacing public scrutiny.
Tracking the Invisible Infrastructure
The construction of datacenters has emerged as a contentious issue across various communities. These facilities, which require enormous amounts of electricity and water, frequently surprise local residents who only discover their existence after construction has already commenced. Epoch AI’s interactive map provides users with visual markers indicating known sites, complete with links to satellite images and project specifics. For instance, a green marker highlights Meta’s “Prometheus” datacenter complex located in New Albany, Ohio. According to Epoch AI, this project has incurred costs exceeding $18 billion and consumes approximately 691 megawatts of power.
Epoch AI describes the Prometheus facility as a blend of weatherproof tents, colocation spaces, and traditional datacenter buildings, emphasizing Meta’s strategic shift towards AI. The map allows users to scroll through a timeline, observing the gradual expansion of the complex as new buildings and cooling systems are added over time.
Understanding Energy Consumption
The core of Epoch AI’s analysis focuses on the cooling systems essential for modern AI operations. These systems are critical as AI technologies generate significant heat. Epoch AI explains that many datacenters now feature cooling units placed outside their buildings or on rooftops. “Modern AI datacenters generate so much heat that the cooling equipment extends outside the buildings,” the organization noted on its website.
To assess energy usage, the team counts the number and size of fans and examines their configurations. This data is fed into a custom model that estimates energy consumption, which in turn helps infer compute capacity and construction costs. Jean-Stanislas Denain, a senior researcher at Epoch AI, stated, “We focus on cooling because it’s a very useful clue for figuring out the power consumption.” However, the model includes uncertainties, as fan configurations can vary widely, leading to potential discrepancies in actual cooling capacity.
Epoch AI acknowledges that their current dataset captures only around 15 percent of global AI compute provided by chipmakers as of November 2025. This limitation stems from varying state and local disclosure laws and the tendency for some projects to fly under the radar. For example, a marker near Memphis, Tennessee, points to xAI’s Colossus 2 project, which reportedly installed natural gas turbines across the Mississippi border to facilitate faster project approvals.
“Based on this information and earlier tweets from Elon Musk, 110,000 NVIDIA GB200 GPUs are operational,” Epoch AI noted, illustrating the scale of infrastructure growth in the AI sector.
Despite their detailed mapping efforts, Epoch AI recognizes that gaps 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,” they stated. The organization plans to extend its research globally, aiming to illuminate infrastructure that significantly influences the future economy, often without the public’s awareness.
