In mid-July 2025 Mark Zuckerberg said Meta was building several multi-gigawatt computing clusters to train its frontier models, naming the first two. Prometheus, in New Albany, Ohio, was described as a roughly gigawatt-scale supercluster slated to come online in 2026 - making it one of the first single AI data center campuses planned to draw on the order of a gigawatt of power. The second, Hyperion, in Richland Parish, Louisiana, was described as scaling toward 5 gigawatts over several years.
The numbers reframed what a data center is. A gigawatt is the output of a large nuclear plant; a 5-gigawatt campus would draw power comparable to several such plants, or to a mid-sized city. Building computing facilities at this scale meant power procurement and grid capacity became inseparable from the project itself, which is why Meta paired the announcements with large generation deals - behind-the-meter gas plants near the Ohio site and, later, a multi-gigawatt nuclear portfolio confirmed in January 2026 to feed Prometheus.
Prometheus and Hyperion sat alongside xAI’s Colossus and the Stargate Project as evidence that the frontier-model race had become a race to assemble unprecedented concentrations of compute and the power to run it. The named, branded superclusters signaled a shift: cutting-edge AI was no longer constrained mainly by algorithms or even chips alone, but by how quickly a company could stand up gigawatts of energized, networked infrastructure.