Bringing AI to the next generation of fusion energy

We are partnering with Commonwealth Fusion Systems (CFS) to bring clean, safe, limitless fusion energy closer to reality.

Nuclear fusion is the process that powers the sun and promises clean, abundant energy without long-lived radioactive waste. Making it work on Earth means keeping the ionized gas, known as plasma, stable at temperatures above 100 million degrees Celsius, all within the limits of fusion energy machines. This is a very complex physics problem that we are working to solve using artificial intelligence (AI).

Today, we are announcing a research partnership with Commonwealth Fusion Systems (CFS), a world leader in fusion energy. CFS is pioneering a faster path to clean, safe, and virtually limitless fusion energy using a compact and powerful tokamak device called SPARC.

SPARC aims to leverage powerful high-temperature superconducting magnets to become the first magnetic fusion machine ever to generate net fusion energy, or more fusion energy than is needed to sustain energy. This breakthrough is known as exceeding the “break-even point” and is an important milestone on the path to viable fusion energy.

This partnership builds on our groundbreaking work using AI to successfully control plasma. In collaboration with our academic partners at the EPFL Swiss Plasma Center (Polytechnic Institute of Lausanne), we have shown that deep reinforcement learning can control tokamak magnets to stabilize complex plasma shapes. To cover a wider range of physics, we developed TORAX, a fast differentiable plasma simulator written in JAX.

Now we are bringing that work to CFS to accelerate the timeline for delivering fusion energy to the grid. We have collaborated in three main areas so far:

Produce fast, accurate, differentiable simulations of fusion plasmas. Find the most efficient and robust path to maximize fusion energy. Discover new real-time control strategies using reinforcement learning.

The combination of our AI expertise and CFS’ cutting-edge hardware makes this an ideal partnership to advance fundamental discoveries in fusion energy for the benefit of the global research community and, ultimately, the entire world.

Fusion plasma simulation

Optimizing tokamak performance requires simulating how heat, current, and matter flow through the plasma core and interact with surrounding systems. Last year, we released TORAX, an open-source plasma simulator built for optimization and control, expanding the range of physics problems we can address beyond magnetic simulation. Because TORAX is built into JAX, it can easily run on both CPUs and GPUs, and you can seamlessly integrate AI-powered models, including your own, for even better performance.

TORAX helps CFS teams test and refine operational plans by running millions of virtual experiments before SPARC is turned on. It also gives you the flexibility to adapt your plan as soon as the first data arrives.

This software is a cornerstone of CFS’s daily workflow, helping you understand how your plasma behaves under different conditions, saving you valuable time and resources.

Find the shortest path to maximum energy

Tokamak operation involves countless choices on how to adjust the various “knobs” available, such as magnetic coil current, fuel injection, and heating output. Manually finding the optimal settings for a tokamak that produces maximum energy while staying within operating limits can be highly inefficient.

Using TORAX in conjunction with reinforcement learning and evolutionary exploration approaches such as AlphaEvolve, our AI agents can explore a vast number of potential operational scenarios in simulation and quickly identify the most efficient and robust paths to producing net energy. This allows CFS to focus on the most promising strategies, increasing the probability of success from day one, even before SPARC is fully up and running.

We have been building the infrastructure to explore various SPARC scenarios. As you learn more about your machine, you can consider maximizing the fusion power produced under different constraints and optimizing its robustness.

Here is an example of a standard SPARC pulse simulated in TORAX. Our AI system can evaluate many possible pulses to find the settings that are expected to give the best performance.

Through a growing collaborative network within the fusion research community, TORAX will be able to be validated and calibrated against historical tokamak data and high-fidelity simulations. This information gives you confidence in the accuracy of your simulations and helps you adapt quickly once SPARC goes live.

Developing an AI pilot for real-time control

In our previous work, we showed that reinforcement learning can control the magnetic configuration of a tokamak. We are now increasing complexity by adding simultaneous optimization of more aspects of tokamak performance, such as maximizing fusion power and managing SPARC thermal loads, allowing it to operate at high performance with greater margins to the machine’s limits.

When SPARC operates at full power, it releases an enormous amount of heat concentrated in a small area, which must be carefully managed to protect the solid-state materials closest to the plasma. One strategy that SPARC can use is to magnetically sweep this exhaust energy along the wall, as shown below.

In the early stages of our collaboration, we are investigating how reinforcement learning agents can learn to dynamically control the plasma to effectively dissipate this heat. In the future, AI may learn adaptive strategies that are more complex than those created by engineers, especially when balancing multiple constraints and goals. Reinforcement learning can also be used to quickly tune traditional control algorithms to specific pulses. Combining pulse optimization with optimal control could move SPARC even faster toward achieving its historic goals.

Combining AI and fusion to build a cleaner future

Alongside our research, Google has invested in CFS to support promising scientific and engineering breakthrough efforts and move the technology toward commercialization.

Looking to the future, our vision extends beyond optimizing SPARC operations. We are building the foundation for AI to become the intelligent, adaptive system at the heart of future fusion power plants. This is just the beginning of our collaborative journey. We hope to share more details about our collaboration as we reach new milestones.

By integrating the transformative potential of AI and nuclear fusion, we are building a cleaner, more sustainable energy future.