Meta has published engineering details on Prometheus, its gigawatt-scale AI supercluster located in New Albany, Ohio, which the company describes as its most ambitious AI infrastructure project to date. The first phase is complete and was used to train Meta's Muse foundational models; the cluster currently comprises hundreds of thousands of GPUs across multiple interconnected buildings and will scale to roughly 1 million distributed GPUs. Prometheus deploys two NVIDIA GPU platforms, the GB200 NVL72-based Catalina and the GB300 NVL72-based Clemente, with a third platform based on NVIDIA's Vera Rubin NVL72 arriving soon. Meta reported achieving more than 85 percent goodput for pre-training runs and a wall-clock mean time between interruptions of approximately 54,000 GPU hours after firmware fixes.
Prometheus represents one of the largest single-site AI compute deployments disclosed by any hyperscaler, with a target of 1 million GPUs at a single Ohio campus. The engineering choices, including 800G scale-out per GPU on the GB200 platform and rack-level liquid cooling, set a benchmark for how the industry is designing and operating gigawatt-scale facilities.
Meta's Engineering blog detailing Prometheus, a gigawatt-scale supercluster, signals a major milestone in hyperscaler AI compute infrastructure and provides rare technical depth from an authoritative first-party source.