AI
Meta opens up the design of its gigawatt-scale Prometheus cluster
Meta published an engineering account of Prometheus, the gigawatt-scale supercluster in New Albany, Ohio, that trained its Muse models. The first phase is complete and the company is working toward about one million GPUs, relying on closed-loop liquid cooling to keep density manageable. Meta notes the site carried ordinary web traffic two years ago before being rebuilt for frontier training, a reminder that AI capacity is being retrofitted onto existing data-centre footprints.