Meta completes first phase of Prometheus AI supercluster with hundreds of thousands of Nvidia GPUs

Meta has completed the first phase of its gigawatt-scale Prometheus supercluster in Ohio, using hundreds of thousands of Nvidia GB200 and GB300 GPUs. The cluster trained Meta's Muse Wonder AI models with over 85% productive training time. Meta plans to expand to 1 million GPUs and add Nvidia's next-generation Vera Rubin systems.