Particle.news

Download on the App Store

Meta Unveils Hyperion and Prometheus to Build Multi-Gigawatt AI Clusters

Meta plans to fuel its Superintelligence Labs with unprecedented compute power that rivals its top AI competitors.

Image
Image
Image

Overview

  • Meta’s Hyperion data center will scale up to five gigawatts of compute power over multiple years and occupy an area equivalent to Manhattan
  • Prometheus is slated to come online in 2026 as a one-gigawatt supercluster to support advanced AI model training
  • Meta is on track to invest between $64 billion and $72 billion in AI data center capex this year as part of a multiyear plan involving hundreds of billions in spending
  • Combined, Hyperion and Prometheus will consume power equal to millions of homes and have already strained local resources, including causing dry taps in Newton County, Georgia
  • The build-out heightens competition with OpenAI’s Stargate, Google DeepMind and xAI’s Colossus while drawing public backing from the White House and the Energy Department for expanded power generation