Meta says first phase of its Prometheus supercluster is complete

On October 19, 2026, Meta Engineering published a post detailing progress on Prometheus, its gigawatt-scale supercluster. According to the post, the first phase of Prometheus is already complete, and the company is working toward scaling the cluster to roughly 1 million GPUs. The post ties the infrastructure to the recently released Muse family of foundational models, describing it as the infrastructure that made those models possible. It also notes that Prometheus incorporates technologies such as closed-loop liquid cooling.

The following day, October 20, Techmeme relayed and added detail from Meta's engineering team, reporting that Prometheus is located in Ohio, that it will have about 1 million GPUs at full scale, and that it has been used to train the Muse foundational models. Both reports agree on the roughly 1 million GPU figure; they describe the current state somewhat differently, with one saying the first phase is complete and the other framing full scale as still ahead. Public information currently stops at Meta's disclosure that phase one is built, with further expansion ongoing.