Meta is quietly building five separate AI data center clusters, each drawing over a gigawatt of power — and according to semiconductor analysis firm SemiAnalysis, that could be enough to surpass OpenAI and Anthropic in total compute by the end of this year.
Among the projects: a roughly 1GW facility called Hyperion in Louisiana, which could eventually expand to 5GW. In Ohio, the Prometheus cluster is already partially operational, with a planned capacity exceeding 3GW. Prometheus comprises six campuses and 27 data centers, all linked by high-speed networking. Meta plans to eventually connect campuses more than 2,000 kilometers apart.
Because of the high latency over such long distances, Meta will handle pre-training in a single region while distributing reinforcement learning tasks across different campuses.
Meta has also assembled a team of about 3,000 people dedicated to creating reinforcement learning tasks and training environments. The company had begun recording some employees' screens, keyboards, and mouse movements to train AI agents on real workflows, but that effort was paused over privacy and data management concerns.
SemiAnalysis argues that Meta now has the talent, data, and compute all lined up, making it one of the best-positioned companies to catch up to OpenAI and Anthropic.