Supporting case
Meta says the first phase of Prometheus is operating; the gigawatt buildout remains a target
What happened
Meta's October 19 engineering account says it has converted its New Albany, Ohio, site from conventional web-service infrastructure into the first operating phase of the Prometheus AI supercluster. It says that phase trained its Muse models and describes hundreds of thousands of GPUs across interconnected buildings. Approximately one million GPUs and gigawatt scale describe the intended buildout, not verified fully operating capacity. October 23 reporting brought the account back into discovery. Meta's primary account ↗.
Meta describes grid/utility coordination, batteries and several cooling arrangements, including closed-loop liquid cooling in newer buildings. Its assertion that it pays the full energy cost is a company statement, not independent verification of utility contracts, public infrastructure costs or total water consumption.
Operating training capacity and the engineering knowledge needed to coordinate it reinforce incumbent scale. Publishing hardware designs does not give outsiders access to the cluster. This account does not establish control of dedicated generation.
Dates and implementation stage
Development 2026-10-19
