Google DeepMind alumni are building tools to accelerate fusion power for the grid

Federico Felici and Jonas Buchli started Fusionality after repeatedly hearing from fusion teams that there was a shortage of off-the-shelf components and expertise for reactor control systems. The Lausanne-based startup raised a $3.7 million (CHF 3 million) pre-seed round from Founderful and Playfair and is building a small team to produce modular control software and simulation environments tailored to fusion experiments.

Many fusion developers still build control systems from the ground up even though much of the underlying functionality is shared across projects. Fusionality plans to exploit that commonality by creating reusable “Lego-like” blocks — both hardware and software — that customers can adapt to their specific reactor designs. The company emphasizes simulation tools as well, enabling startups to test and tune control strategies before deploying them on live devices.

The founders bring hands-on experience in fusion control and AI. They met while teaching AI to operate a tokamak; Felici worked at EPFL and later at Google DeepMind developing simulations and machine learning interfaces for fusion devices, while Buchli worked at DeepMind. The founders see artificial intelligence as a valuable enhancer of control systems, useful for optimization and specific tasks, but not yet as a wholesale replacement for human-engineered control logic.

Initially Fusionality is concentrating on magnetic confinement approaches, which use strong magnetic fields to confine hot plasma — a method pursued by companies such as Commonwealth Fusion Systems, Realta Fusion, Proxima Fusion, and Type One Energy. The pre-seed funding will support development of a limited set of core technologies the team considers broadly applicable; the founders say they intend to expand the product set over time but did not disclose exact technical details.