DeepMind alumni build fusion control tools

Federico Felici and Jonas Buchli kept hearing the same frustration from fusion companies: they wanted to buy control system components off the shelf, but no one was selling them. “We would ideally buy many of the components to make our our control system, but there isn’t really anybody providing them — and especially there isn’t anybody who speaks the language of fusion,” Felici told TechCrunch. That gap became the seed of a new startup.

A problem they know firsthand

Felici and Buchli have spent years working on how to control experimental fusion devices. “We spent quite a lot of our time solving exactly the problems that our customers need to solve,” Felici said. “It felt like the right time.”

This year they co-founded Fusionality, based in Lausanne, and quickly raised a $3.7 million (CHF 3 million) pre-seed round from Founderful and Playfair. Felici is CEO, Buchli is CTO.

An emerging supply chain

Fusionality is part of a growing ecosystem serving the fusion power industry. Some startups, like Kyoto Fusioneering, develop components that help turn engineering breakthroughs into electricity for the grid, while others are existing manufacturers with the skills to produce high-precision parts.

But few companies specialize in the hardware and software needed to control fusion reactors themselves. “Across the industry, many companies make their own control systems from scratch,” Felici said. “But actually, 80% of each company’s control system is really exactly the same.”

What fusion control actually involves

Fusion power plants generate electricity by harnessing the energy released when two atoms fuse, which is more likely to happen in a superheated plasma. But plasmas are fickle, and maintaining the right temperature, shape, and fuel level requires split-second decisions.

The control systems that keep those conditions stable are complex to build and deploy, yet much of the fundamental physics is consistent across reactor designs. Fusionality plans to build on that commonality, offering a suite of control systems and simulation environments that startups can fine-tune to their specific design.

From EPFL to DeepMind

Felici and Buchli met while teaching AI to control an experimental tokamak, a donut-shaped device used to study fusion. At the time, Felici was at EPFL in Switzerland, which operates one such tokamak, while Buchli was at Google DeepMind. Felici later moved to Google DeepMind, where he developed simulations and machine learning interfaces for fusion devices.

Their combined experience shapes Fusionality’s approach. They know both the physics and the software challenges firsthand.

A pragmatic view of AI

Felici said AI will “play an important role in future control systems,” but it’s not ready to handle everything. “I wouldn’t be somebody who advocates for AI to take the role of controlling the entire fusion reactor,” he said. Today, AI is more likely to “complement or enhance or optimize” parts of the system.

This measured stance is grounded in real experience. Fusion control is a safety-critical task where a slow or mistaken decision can destabilize the plasma. AI can assist, but human-designed rules will likely remain central for now.

Starting with magnetic confinement

Fusionality is initially focusing on magnetic confinement, an approach that uses powerful electromagnets to keep fusion fuel dense and hot enough to feed a power plant. Commonwealth Fusion Systems, Realta Fusion, Proxima Fusion, and Type One Energy are all developing magnetic confinement devices.

Felici said the company will eventually build systems for other approaches as well.

Building ‘Lego blocks’

The pre-seed funding will help the team, now seven people, develop a “tightly selected number of technologies,” Felici said, though he declined to name them. He compared the strategy to Lego blocks. “We start with a few of these blocks and then we build out more,” he said. “We have the ambition to serve the really wide range of technology that you need to operate a fusion reactor.”

What this means for fusion’s future

Fusion startups have historically built their own control systems, an expensive and time-consuming process that duplicates effort across the industry. If Fusionality can deliver reliable, adaptable tools, it could help shorten the time from a successful experiment to a commercial plant. That’s exactly what the industry needs as it races to bring zero-carbon electricity to the grid.

By standardizing a critical piece of reactor infrastructure, Fusionality could help the entire field move faster — even as it leaves the core physics to the specialists.