A small research team is revealing the rapid growth of datacenter infrastructure in the United States through innovative mapping techniques. According to a report from 404 Media, the non-profit research institute Epoch AI is utilizing publicly available information and satellite imagery to document facilities that are often overlooked in public discussions. This initiative aims to enhance understanding of the scale and pace of artificial intelligence development, which is becoming an increasingly critical component of the economy.
Mapping the Infrastructure Growth
Datacenter construction has emerged as a significant issue across the nation. These facilities require enormous quantities of electricity and water, and many local communities only become aware of their existence once construction is underway. Epoch AI’s interactive map visualizes known sites, providing detailed information linked to satellite views and project specifics. For instance, one marker indicates the location of Meta’s “Prometheus” datacenter complex in New Albany, Ohio. Epoch AI estimates that this project has already cost approximately $18 billion and draws 691 megawatts of power. The organization describes this facility as a combination of weatherproof tents, colocation facilities, and traditional datacenter buildings, noting it reflects Meta’s strategic pivot towards AI.
Visitors to the map can explore a timeline showing the complex’s growth over time, with satellite images illustrating the addition of new buildings and cooling systems.
Understanding Power Consumption Estimates
A significant portion of Epoch AI’s analysis focuses on the cooling infrastructure necessary for modern AI systems, which generate substantial heat. Often, cooling units are installed outside the buildings or on rooftops, with Epoch AI explaining, “Modern AI data centers generate so much heat that the cooling equipment extends outside the buildings.”
The researchers meticulously count fans, measure their sizes, and analyze their placements to feed this data into a custom model that estimates energy consumption. This power estimate plays a crucial role in inferring compute capacity and construction costs. Senior researcher Jean-Stanislas Denain emphasized the importance of cooling systems, stating, “We focus on cooling because it’s a very useful clue for figuring out the power consumption.”
Although this model provides valuable insights, it does carry some uncertainty. Variations in fan speed and configuration mean that actual cooling capacity could be either significantly higher or lower than estimates suggest.
The map is still a work in progress. Variability in state and local disclosure laws can leave gaps in the data. Some projects intentionally avoid public attention, and smaller facilities may go unnoticed. Epoch AI estimates that the current dataset represents only about 15 percent of global AI compute capacity provided by chipmakers as of November 2025.
Markers on the map extend across the United States, including one near Memphis, Tennessee, which highlights xAI’s Colossus 2 project. Epoch AI notes that the company has installed natural gas turbines across the Mississippi border to expedite the approval process. Based on this information and previous comments from Elon Musk, the organization estimates that 110,000 NVIDIA GB200 GPUs are currently operational at this site.
Despite the detailed mapping efforts, significant unknowns persist. Epoch AI acknowledges that even with comprehensive analysis of a datacenter, crucial information regarding its users and their consumption remains elusive. The organization intends to broaden its search globally, aiming to illuminate the infrastructure that is increasingly shaping the future economy, often without sufficient public visibility.
