Particle.news

Download on the App Store

Meta Builds Multi-Gigawatt AI Superclusters in Push for Superintelligence

The infrastructure and talent strategy is designed to overtake AI rivals, drawing regulatory scrutiny over power and water consumption in host regions.

Overview

  • Meta is constructing multiple multi-gigawatt AI superclusters with the first, Prometheus, slated to go live in 2026 and Hyperion set to scale up to 5 GW over the coming years.
  • The company has raised its 2025 capital expenditure forecast to $64–72 billion and plans to invest hundreds of billions more to fuel its AI compute ambitions.
  • Meta Superintelligence Labs has recruited elite researchers from OpenAI, Google DeepMind and other leaders, led by Scale AI co-founder Alexandr Wang.
  • To accelerate deployments, some of the new AI data centers are being assembled in temporary tent structures using prefabricated power and cooling modules.
  • Permit applications show the data centers could draw up to six million gallons of water per day, prompting local authorities to weigh constraints on supplies.