Meta details Prometheus gigawatt-scale supercluster, targeting ~1M GPUs
BenBajarin · x · 2026-10-20
Meta's engineering blog details Prometheus, its gigawatt-scale supercluster in New Albany, Ohio. Phase one is complete and was used to train the Muse foundation models, including the frontier Muse Wonder models for personal and professional agentic tasks; the cluster will scale to roughly 1 million GPUs.
Key points:
- Two years ago the site ran ordinary web traffic; it was rebuilt for frontier training via changes in hardware, networking and building design.
- It uses closed-loop liquid cooling for efficiency.
- After reliability work, pretraining runs achieve over 85% goodput, and Muse Wonder's RL runs exceed 90% goodput.
- The post covers cluster topology across compute, networking and software. Meta says only Google has comparable scale for serving AI to billions.
Related event: Meta Unveils Gigawatt Prometheus Supercluster, Targets One Million GPUs(4 posts)→
More from Infra
- Tenstorrent Raises $1.5B at $6.2B Valuation After Takeover Interest — Sethwinterroth · 2026-10-20
- Perplexity's Open-Source Sandbox Platform SPACE Joins Crusoe Cloud's Managed AI Portfolio — AravSrinivas · 2026-10-20
- vLLM Team Details Agent-Assisted GPU Kernel Work: Kimi K3 Hits 1.73x Stock vLLM — lmoroney · 2026-10-20
- Gartner: $3.5T of AI infra spend may face write-downs by 2030, 80% of neoclouds to fold — rohanpaul_ai · 2026-10-20
- Stripe paid $7.5B for a router, but routing is easy — verifying outputs is the real moat — AIGODSEND · 2026-10-20
- Google Cloud upgrades VPC Service Controls with folder-level perimeters for agent traffic, free — rseroter · 2026-10-20