$META Scales Prometheus AI Supercluster in Ohio Toward 1 Million GPUs and Roughly 1 GW of Capacity

Meta ($META) is scaling its Prometheus AI supercluster in Ohio toward 1 million GPUs and roughly 1 GW of capacity, making it one of the largest AI training facilities in the world.
Powered by NVIDIA ($NVDA) GB200 and GB300 systems, the cluster has achieved over 85% training efficiency on Meta's Muse Wonder models, with plans to adopt next-generation Vera Rubin hardware.
Prometheus relies on advanced liquid cooling, high-speed networking, and infrastructure from Arista ($ANET) and Broadcom ($AVGO) to connect hundreds of thousands of GPUs across multiple buildings.
Meta is also securing long-term nuclear power through partnerships with Vistra ($VST), Oklo ($OKLO) and TerraPower, supporting its growing AI infrastructure needs.