Software company Google DeepMind has announced a research partnership with US-based start-up Commonwealth Fusion Systems (CFS). Since it was spun-out from the Massachusetts Institute of Technology (MIT) in 2018, CFS has raised almost $3bn in capital, which it plans to use to complete its SPARC demonstration tokamak under construction in Devens, Massachusetts. It will also progress development work on its planned 400 MWe ARC fusion power plant in Chesterfield County, Virginia, with grid connection expected in the early 2030s.
Earlier in October, CFS announced that the results of rigorous performance tests of its toroidal field (TF) magnet, a key component of SPARC, had been validated by an independent panel of reviewers under the US Department of Energy’s (DOE’s) Milestone-Based Fusion Development Program.
DeepMind said the partnership with CFS builds on work using AI to successfully control a plasma. “With academic partners at the Swiss Plasma Center at EPFL (École Polytechnique Fédérale de Lausanne), we showed that deep reinforcement learning can control the magnets of a tokamak to stabilise complex plasma shapes. To cover a wider range of physics, we developed TORAX, a fast and differentiable plasma simulator written in JAX. Now, we’re bringing that work to CFS to accelerate the timeline to deliver fusion energy to the grid.”
TORAX will help CFS teams test and refine their operating plans by running millions of virtual experiments before SPARC is even turned on. It also gives them flexibility to quickly adapt their plans once the first data arrives.
Collaboration between CFS and DeepMind is already underway in three key areas:
- Producing a fast, accurate, differentiable simulation of a fusion plasma;
- Finding the most efficient and robust path to maximising fusion energy; and
- Using reinforcement learning to discover novel real-time control strategies.
“The combination of our AI expertise with CFS’s cutting-edge hardware makes this the ideal partnership to advance foundational discoveries in fusion energy for the benefit of the worldwide research community, and ultimately, the whole world,” DeepMind noted.
TORAX software has become a linchpin in CFS’s daily workflows, helping them understand how the plasma will behave under different conditions, saving precious time and resources. “TORAX is a professional, open-source plasma simulator that saved us countless hours in setting up and running our simulation environments for SPARC,” said Devon Battaglia, Senior Manager of Physics Operations at CFS.
Operating a tokamak involves countless choices in how to manipulate magnetic coil currents, fuel injection and heating power. Manually finding optimal settings to produce the most energy, while staying within operating limits, could be very inefficient. Using TORAX in combination with reinforcement learning or evolutionary search approaches allows AI to explore vast numbers of potential operating scenarios in simulation, rapidly identifying the most efficient and robust paths to generating net energy. This can help CFS focus on the most promising strategies, increasing the probability of success, even before SPARC is fully commissioned and operating at full power, which is planned for 2026 or 2027.
“It’s a big deal,” said CFS. “AI tools carry serious potential to accelerate learning on SPARC and then to push toward making our ARC fusion power plants more efficient and economical.” Alongside research, Google deepened its investment in CFS earlier in 2025 and agreed to purchase 200 MWe of fusion power from the first ARC power plant in the early 2030s.
According to CFS, TORAX, which emerged in 2024, is a key enabling technology. Its framework can link together multiple computing engines, including AI tools, into a single, fast, modern computing engine that greatly accelerates the pace of discovery. CFS said it is on the fastest path to commercial fusion energy in part because we’re able to build on a physics understanding of tokamaks developed over decades of research. That research included computer simulations validated by real-world tokamak operation.
However, before TORAX, those simulations required a patchwork of software projects and programming languages. “Now we have something better thanks to TORAX and its use of Google numerical computation software called JAX that can integrate results from multiple AI models,” CFS noted. “Using JAX, TORAX lets us distil plasma calculations into a blazingly fast computation that makes it easier to combine AI models built from that patchwork of projects. And TORAX can run that software on processors with the latest hardware acceleration features.”
CFS is aiming for commercial availability of fusion power in the early part of the next decade, a target which many believe is overly ambitious. However, CFS CEO Bob Mumgaard told Axios that AI will help to make it a reality. He acknowledged that fusion energy has long been viewed as something for the distant future, but noted that AI was once seen the same way. “We are close to breaking that meme,” he said.
