Meta details how it built and stabilized its Prometheus AI supercluster
Original titleInside Prometheus: Meta’s Journey to a Gigawatt-Scale Supercluster
AISummary
Meta says the first phase of its gigawatt-scale Prometheus supercluster in New Albany, Ohio, is complete and was used to train its Muse models, with a goal of roughly 1 million GPUs. The company reports that interruptions fell 99% from their mid-2025 peak through firmware fixes, telemetry, and cross-partner work with NVIDIA, and that MTBI rose to more than 54K GPU hours.
AIWhy it matters
The article explains how Meta built and stabilized its gigawatt-scale Prometheus cluster, giving concrete reliability, network, and power figures useful for judging large AI training infrastructure.
Source: Engineering at Meta · engineering.fb.comPublished · added here