On Thursday, energy startup Commonwealth Fusion Systems (CFS) announced a collaboration with Google’s DeepMind to optimize and potentially enhance the performance of its upcoming Sparc reactor through artificial intelligence.
Their joint initiative involves simulating the plasma that will ignite within CFS’s reactor using DeepMind’s specialized software, Torax. Additionally, they intend to integrate Torax with AI models to assist CFS in determining the most effective strategies for achieving fusion energy.
Fusion energy holds the promise of generating vast amounts of power with no emissions, using water as an almost inexhaustible fuel source. Companies focused on AI have shown strong interest in fusion startups as a means to supply electricity for power-intensive data centers, and Google is also considering them as future clients.
This is not Google’s initial venture into nuclear fusion. The company has previously partnered with fusion startup TAE Technologies, leveraging AI to analyze plasma dynamics within TAE’s fusion apparatus.
Google’s repeated involvement in fusion research is no coincidence: AI may be particularly well-suited to overcoming the challenges of making fusion energy viable.
A major hurdle for fusion startups is maintaining plasma at sufficiently high temperatures for extended periods. Unlike nuclear fission, which sustains itself, fusion reactions are difficult to keep going outside of stars like our sun. Without the immense mass and gravity of a star, plasma tends to dissipate and extinguish itself.
CFS’s reactors use strong magnets to replace gravity and contain the plasma, but this method isn’t flawless. Operators must create control software that allows the reactor to adapt in real time to fluctuating plasma conditions.
The challenge lies in the sheer number of variables involved—far more than a human could manage. This is precisely the kind of complex problem where AI excels. Industry experts have identified AI as a critical factor behind the sector’s significant progress in recent years.
CFS is in the process of constructing Sparc, its demonstration reactor, in a suburb near Boston. The project is about two-thirds finished, and once completed in late 2026, the company expects it to be the first fusion device to generate more energy than it consumes.
According to Google, Torax can work alongside reinforcement learning or evolutionary search models to identify the “most efficient and robust paths to generating net energy.” The partners are also investigating whether AI can be used to directly manage the reactor’s operations.
In August, Google joined Nvidia in CFS’s $863 million Series B2 funding round. Earlier this year, Google also committed to purchasing 200 megawatts of electricity from CFS’s inaugural commercial facility, Arc, which is set to be constructed near Richmond, Virginia. The tech giant is also an investor in TAE Technologies, a competitor of CFS.
