Particle.news
Download on the App Store

DeepMind and CFS Launch AI Partnership to Speed SPARC Toward Net‑Energy Fusion

The collaboration applies TORAX simulation with reinforcement learning to plan pulses, study heat‑load control on the under‑construction device.

Overview

  • DeepMind formally announced a research tie‑up with Commonwealth Fusion Systems to apply AI across SPARC’s modeling, optimization and control.
  • TORAX, DeepMind’s open‑source, JAX‑based plasma simulator, is being integrated into CFS workflows to run millions of virtual experiments before operations begin.
  • AI agents using reinforcement learning and evolutionary search will probe operating scenarios to find robust paths to maximizing fusion energy within machine limits.
  • Early work targets real‑time control research to manage intense divertor heat, including strategies that sweep exhaust energy across plasma‑facing surfaces.
  • CFS says SPARC in Devens, Massachusetts is about two‑thirds complete with commissioning expected in 2026, as Google backs CFS financially and has agreed to buy 200 MW from the first ARC plant planned near Richmond, Virginia.