In April 2025, something exciting happened at the ITER Private Sector Fusion Workshop in Cadarache. Scientists, engineers, and tech innovators gathered in a vibrant room where the boundaries between major scientific research and commercial innovation began to dissolve.
Three organizations—Microsoft Research, Arena, and Brigantium Engineering—spoke about how artificial intelligence (AI) is stepping up to help unlock the potential of nuclear fusion. Each speaker tackled different pieces of a complex puzzle, but they shared a common message: AI has shifted from a buzzword to a crucial tool in big science and engineering, including fusion projects.
Kenji Takeda from Microsoft Research highlighted the rapid pace of the AI revolution, comparing it to the agricultural and industrial revolutions. Just a month before the workshop, Microsoft signed a Memorandum of Understanding with ITER to explore how AI can speed up research and development. They already implemented Microsoft technologies to empower ITER teams. For instance, they developed an Azure-based chatbot to help staff sift through over a million documents, making technical knowledge more accessible. GitHub Copilot aids in coding, while AI addresses routine IT support issues.
But Microsoft isn’t stopping there. A major challenge in fusion is creating materials that can withstand extreme conditions—like heat and radiation. That’s where AI shines. They introduced MatterGen, a generative AI model that designs new materials based on specific properties. “It’s like ChatGPT,” Takeda explained. “Instead of asking for a poem, we ask it to create a material for a fusion reactor.”
Next comes MatterSim, a simulation tool predicting how these new materials will perform in real-world scenarios. By blending generation and simulation, Microsoft aims to find materials not yet in any catalogue.
While Microsoft focuses on the atomic level, Arena tackles the hardware development speed. As Michael Frei said, “Software innovation happens in seconds. In hardware, that loop can take months—sometimes even years.” Arena’s solution is Atlas, an AI platform that acts like an extra set of hands for engineers. It can read data sheets, interpret lab results, and even interact with lab equipment through software. “Instead of manually adjusting gear,” Frei noted, “you can simply say, ‘Verify the I2C protocol,’ and Atlas handles it.”
Atlas goes further—it can adapt firmware in real-time, enabling quicker feedback, faster prototyping, and fewer late nights in the lab. Arena wants to make hardware development feel as fluid and fast as software creation, with smart tools driving the process.
Now, let’s talk about the construction aspect. Brigantium Engineering is focused on building these massive, unique machines. Founder Lynton Sutton explained their use of “4D planning,” combining 3D CAD models with detailed construction schedules to visualize how projects evolve over time. Instead of struggling with Gantt charts and static models, they create animations that depict the construction process shape by shape, essential for safety reviews and stakeholder meetings.
Brigantium also taps into virtual reality, using Unreal Engine to bring these models to life. One recent initiative reconstructed ITER’s tokamak pit, employing drone footage and photogrammetry for full interactivity, accessible even through a web browser. “We’ve enhanced visualization quality,” Sutton said, noting smoother textures and better graphics. “Soon, anyone in the team can easily navigate the 4D model online.”
Looking ahead, Sutton sees a future where AI could automate syncing construction schedules with 3D models, potentially reaching down to individual bolts and fasteners. This would not only make visualization better but also help eliminate delays.
Despite their different approaches, a common thread ran through all the presentations: AI is more than just a productivity tool; it’s becoming a collaborative partner in creativity and scientific discovery. Takeda mentioned that Microsoft is researching “world models,” inspired by video games, where AI learns about physics by analyzing videos of real phenomena like plasma behavior. “If it watches enough, it might grasp the physics of plasmas,” he said.
This vision may seem futuristic, but it’s rooted in logic. The more AI learns from the world, the better it can assist humans in understanding and mastering complex challenges. The overarching message from the workshop was clear: AI is here not to replace scientists, engineers, or planners, but to support them, enhancing their work and perhaps making it more enjoyable along the way. Takeda emphasized, “These are just a few examples of how AI is starting to be integrated at ITER, and this is only the beginning of that journey.”