Google DeepMind partners with Commonwealth Fusion Systems on AI-driven fusion breakthrough

Google DeepMind has announced a partnership with Commonwealth Fusion Systems (CFS) to apply artificial intelligence to the challenge of controlling fusion plasma, bringing clean, abundant energy closer to practical deployment.

The fusion challenge#

Fusion—the process that powers the sun—offers the promise of clean, safe energy without long-lived radioactive waste. The technical hurdle is immense: keeping ionized gas (plasma) stable at temperatures exceeding 100 million degrees Celsius within a fusion machine. CFS is tackling this with SPARC, a compact tokamak powered by high-temperature superconducting magnets, designed to be the first to achieve "breakeven"—generating more power from fusion than it consumes.

Three pillars of the collaboration#

The partnership focuses on three areas:

• Fast, accurate plasma simulation: DeepMind's open-source TORAX simulator, built in JAX, allows CFS teams to run millions of virtual experiments before SPARC is switched on, helping refine operating plans.
• Optimization under constraints: Using reinforcement learning and evolutionary search, AI agents can explore vast numbers of operating scenarios to identify the most efficient paths to net energy generation.
• Real-time control: Building on earlier work where deep reinforcement learning controlled tokamak magnets, the teams are now optimizing multiple performance aspects simultaneously—maximizing fusion power while managing heat loads within machine limits.

Building the future#

DeepMind's work extends a 2023 breakthrough with the Swiss Plasma Center at EPFL, where reinforcement learning successfully controlled plasma shapes in a tokamak. Now, the expanded partnership aims to make AI an adaptive system at the heart of future fusion power plants.

"By uniting the revolutionary potential of AI and fusion, we're building a cleaner and more sustainable energy future," DeepMind stated in announcing the collaboration. Google has also invested in CFS to support their path to commercialization.

The work represents a critical moment in fusion energy: where physics meets machine learning at scale.