DeepMind Alumni Raise $3.7M for Fusionality Fusion Controls
Google DeepMind alumni launch Fusionality with $3.7 million to build reusable reactor controls and simulations for fusion startups targeting the power grid.
Summary
Federico Felici and Jonas Buchli founded Lausanne-based Fusionality in 2026 to supply fusion reactor controls that many developers currently build themselves. Felici, the CEO, estimates 80% of each company’s system is identical. The startup raised a $3.7 million, CHF 3 million, pre-seed round from Founderful and Playfair; Buchli is CTO, and the team has seven people. Kyoto Fusioneering and precision manufacturers serve other parts of fusion’s emerging supply chain, but few suppliers specialize in reactor control hardware and software.
Fusionality plans reusable control systems and simulation environments that customers can adapt to their designs. Such systems must make split-second adjustments to plasma temperature, shape and fuel levels. Felici and Buchli met while teaching AI to control EPFL’s experimental tokamak, when Felici worked at EPFL and Buchli at Google DeepMind. Felici later joined Google DeepMind to develop fusion simulations and machine learning interfaces. He expects AI to enhance or optimize selected functions, not control an entire reactor.
The company initially targets magnetic confinement, which uses powerful electromagnets to keep fusion fuel sufficiently hot and dense. Commonwealth Fusion Systems, Realta Fusion, Proxima Fusion and Type One Energy use this approach. Fusionality plans to expand into other fusion methods, while its funding supports a small, modular set of initial technologies that Felici has not identified, with more components intended to follow.
Positives
- $3.7 million, or CHF 3 million, gives Fusionality funding to begin developing modular reactor-control technologies.
- An estimated 80% of fusion companies’ control systems is identical, creating scope for reusable products that reduce duplicated engineering.
- Felici and Buchli bring direct experience using AI to control EPFL’s experimental tokamak.
- Commonwealth Fusion Systems, Realta Fusion, Proxima Fusion and Type One Energy provide an initial market for magnetic-confinement controls.
- Fusionality plans to expand its control systems beyond magnetic confinement as its modular portfolio grows.
Risks & concerns
- Fusionality has not disclosed which control technologies its seven-person team will develop first.
- AI is not ready to control an entire fusion reactor and can currently enhance or optimize only selected functions.
- Plasma temperature, shape and fuel require split-second management, making reactor controls complex to develop and deploy.
- Fusionality initially supports only magnetic confinement, with systems for other fusion approaches deferred.
- Many fusion companies still build control systems from scratch because specialized suppliers remain scarce.