Bringing AI to the next generation of Fusion Energy

We are working with Commonwealth Fusion Systems (CFS) to bring clean, safe, unlimited marine energy closer to reality.

Fusion, the process that powers the sun, promises clean, energy-rich energy without long-lasting radioactive waste. Making it work here on earth means keeping ionized gas, known as plasma, in domestic stables over 100 million degrees Celsius – all within the limits of the Fusion machine. This is a very complex physics problem that we are working to solve with artificial intelligence (AI).

Today, we are announcing our research partnership with Commonwealth Fusion Systems (CFS), the global leader in Fusion power. CFS pioneered a fast clean, safe and unlimited power source with their powerful tokamak machine called SPARC.

SPARC achieves high-temperature energy and aims to be the first nuclear reactor in history to produce net fusion energy – More energy from fusion than is needed to support us. The critical breakthrough is known as crossing the “leap,” and is an important milestone on the path to effective toxicology.

This collaboration builds on our worldwide work using AI to effectively control plasma. With study partners at the Swiss Plasma Center at EPFL (École Polytechnique Fédérale de Lausanne), we have shown that deep learning can control the magnets of a tokamak to strengthen the structure of a complex plasma. To cover a wide range of physics, we created torax, a fast and unique plasma simulator written in Jax.

Now, we're bringing that work to CFS to accelerate the timeline to bring renewable energy to the grid. We have partnered in three key areas so far:

• To produce a fast, accurate, unique fusion plasma coating.
• Finding an efficient and robust way to maximize the power of Fusion.
• Reinforcement learning is used to derive real-time control strategies.

The combination of our AI technology with CFS hardware at the CFS edge makes this a great collaboration before getting the circuit available to the entire community, and eventually, the world.

It simulates fusion plasma

In order to use the Tokamak operation, we need to simulate how the heat, produced by electricity, flows through the plasma core and interacts with the surrounding systems. Last year, we released Torax, an open source simulator designed for efficiency and control, to expand the range of physics questions we can answer with this magnetic simulation method. Torax is built on jax, so it can easily run on both CPU and GPU and can integrate well with AI-enabled models, including ours, to achieve better performance.

Torax will help CFS teams test and refine their operational strategies by running millions of virtual tests before SPARC opens. It also gives them the flexibility to adapt their plans as soon as the first data arrives.

This software has been a linchpin in CFS's day-to-day operations, helping them understand how the plasma will behave under different conditions, saving valuable time and resources.

Finding the fastest way to higher power

Operating a Tokamak involves countless decisions on how to use the various “knots” available, such as magnetic coil currents, fuel injection and thermal power. Manually finding the right Tokamak settings to produce the most power, while staying within operating limits, can be inefficient.

Using Torax in combination with reinforcement learning or natural search methods such as Alphaevolve, our AI agents can evaluate large numbers of possible active scenarios in simulations, quickly identifying efficient and robust methods to generate healthy energy. This can help focus CFS on the most promising strategies, increasing the chances of success from day one, even before SPARC is fully deployed and operating at full capacity.

We are building an infrastructure to investigate various SPARC scenarios. We can look to increase the fusion power produced under different conditions, or to optimize the durability as we learn more about the machine.

Here we show examples of a typical sparc pulse made in the Thorax. Our AI system can test many possible pulses to find the settings we expect to perform best.

Through our growing network of collaborations within the Fusion research community, we will be able to validate and measure Torax with past tokamak data and high throughput. This information will provide confidence in the accuracy of the simulation and help us quickly adapt when SPARC goes live.

Developing an AI driver for real-time control

In our previous work, we have shown that reinforcement can control the magnetic configuration of a Tokamak. We are now adding to the specification of the optimal use of additional Tokamak functions, such as increasing the fusion power or managing the thermal load of SPARC, so it can work with high efficiency with the machine's margits.

When operating at full power, SPARC will emit a large amount of heat concentrated in a small area that must be carefully managed to protect solids near the plasma. One strategy SPARC can use is to sneakily trip over the wall's draining power, as shown below.

In the first phase of our collaboration, we are investigating how energy-enhancing learning agents can learn how to manipulate the plasma through regeneration to distribute this heat effectively. In the future, AI could learn more complex dynamic strategies than anything an engineer could devise, especially when balancing multiple constraints and objectives. We can also use reinforcement to make the traditional algorithms faster. A combination of pulse generation and good control can push sparc further and faster to achieve its historic goals.

Combining AI and Fusion to create a cleaner future

In addition to our research, Google has invested in CFS, supporting its work to improve lives in science and engineering, and commercializing its technology.

Looking ahead, our vision extends beyond SPARC operations. We are building AI foundations to be the intelligent, adaptive system at the heart of future Fusion Power Plags. This is just the beginning of our journey together, and we hope to share more details about our collaboration as we reach new heights.

By combining the transformative power of AI and farming, we are building a clean and sustainable energy future.