A pair of former Google DeepMind researchers are turning years of work controlling experimental fusion machines into a startup that wants to solve one of the industry's less visible problems: how to reliably operate the reactors themselves.

Swiss startup Fusionality has raised CHF 3 million, approximately $3.7 million, in pre-seed funding from Founderful and Playfair to develop measurement, simulation and control technologies for fusion energy companies.

Based in Lausanne, Switzerland, Fusionality was founded in 2026 by Federico Felici, who serves as CEO, and Jonas Buchli, the company's CTO.

Instead of attempting to build another fusion reactor, Fusionality is positioning itself deeper in the industry's supply chain.

Its bet is relatively straightforward: as billions of dollars flow into companies attempting to commercialise fusion energy, those companies will increasingly need specialised suppliers capable of providing the technology required to actually monitor and control their machines.

Today, much of that infrastructure is still being developed internally.

Fusionality wants to change that.

The founders spotted the same problem repeatedly

The idea emerged from conversations Felici and Buchli had with companies developing fusion technology.

Fusion developers were willing to buy components of their control systems rather than develop everything internally, Felici told TechCrunch, but there were few suppliers that combined control-system engineering with a deep understanding of fusion physics.

That presented an opportunity.

Felici and Buchli had already spent years working on many of the same technical challenges their potential customers were attempting to solve.

Rather than continuing to solve those problems inside individual research projects, they decided to package that expertise into technology that could potentially be deployed across multiple fusion companies.

The opportunity is particularly interesting because, according to Felici, as much as 80% of the control infrastructure being developed by different fusion companies can be fundamentally similar.

Yet many companies continue building their own systems from scratch.

If Fusionality can standardise a significant portion of that technology, it could allow reactor developers to concentrate more resources on the parts of their machines that actually differentiate them.

Controlling fusion is a massive engineering challenge

Nuclear fusion attempts to recreate the physical process that powers stars.

Instead of splitting heavy atoms, as conventional nuclear fission reactors do, fusion combines lighter atomic nuclei and releases enormous amounts of energy in the process.

The promise is enormous.

If commercial fusion can be achieved economically and reliably, it could eventually provide large quantities of low-carbon electricity.

But producing a fusion reaction is only one part of the problem.

Keeping the reaction under control is extraordinarily difficult.

In magnetic-confinement fusion systems, fuel is heated until it becomes an extremely hot plasma. Powerful magnetic fields are then used to keep that plasma away from the reactor walls while maintaining the conditions required for fusion.

Temperature, plasma shape, fuel delivery, magnetic fields and numerous other variables have to be constantly monitored.

Changes can happen extremely quickly.

That means a fusion reactor requires sophisticated sensors, software, simulations and real-time control systems capable of processing information and responding almost instantly.

This is the part of the fusion stack Fusionality wants to address.

Think of it as infrastructure for the companies building reactors

Fusionality plans to develop technologies spanning measurement, control, simulation and data systems.

Instead of creating a completely bespoke system for every reactor company, the startup wants to develop reusable building blocks that customers can adapt to their particular reactor designs.

Felici has compared the approach to Lego.

Fusionality can build individual technological blocks and gradually expand the collection until customers have access to much of the infrastructure required to operate their fusion devices.

The company says its expertise covers fusion and plasma physics, control engineering, data engineering, artificial intelligence and machine learning, whole-device simulation, real-time diagnostics and software-hardware integration.

That combination could become increasingly valuable as fusion companies move from laboratory experiments toward larger demonstration machines and eventually commercial power plants.

The founders have already used AI to control fusion plasma

Fusionality's founders are not approaching the problem for the first time.

Felici and Buchli met through work involving the TCV tokamak at the Swiss Plasma Center at EPFL.

A tokamak is a doughnut-shaped magnetic-confinement device used to contain and study extremely hot plasma.

Their work eventually became part of a collaboration between EPFL and Google DeepMind in which researchers demonstrated that deep reinforcement learning could be used to control plasma inside the TCV tokamak.

The research was published in Nature in 2022.

Felici later spent approximately two and a half years at Google DeepMind after previously working at EPFL, while Buchli also worked at DeepMind, where his experience included robotics, optimal control and reinforcement learning.

That history gives Fusionality an obvious connection to artificial intelligence.

But the founders are taking a more cautious position on how AI should be deployed inside actual fusion reactors.

AI could help run reactors — but Fusionality isn't handing over everything

Artificial intelligence is expected to play an increasingly important role in fusion control.

Machine-learning systems could potentially analyse enormous amounts of sensor data, optimise reactor conditions and help operators respond to changes faster than conventional systems.

But Felici does not believe today's AI should simply be handed complete control of an entire fusion reactor.

Instead, the company sees AI as one component of a broader control architecture.

It could enhance or optimise individual parts of reactor operations while more traditional engineering and control systems continue handling other critical functions.

That distinction matters.

Fusion reactors are highly complex physical machines where errors could damage expensive equipment and interrupt experiments. Reliability and predictability therefore matter just as much as raw AI capability.

For Fusionality, the opportunity may consequently be less about creating an "AI fusion reactor" and more about combining machine learning with proven control engineering.

Fusionality is starting with magnetic confinement

The startup is initially concentrating on magnetic-confinement fusion, one of the industry's leading approaches.

Magnetic confinement uses powerful magnetic fields to contain extremely hot plasma.

Several heavily funded fusion companies are pursuing variations of this approach, including Commonwealth Fusion Systems, Proxima Fusion, Type One Energy and Realta Fusion.

But those companies are not necessarily building identical machines.

Some use tokamaks, while others are pursuing stellarators, magnetic mirrors or other configurations.

That makes Fusionality's strategy particularly challenging — and potentially valuable.

Its technology has to provide enough common infrastructure to be reusable while remaining flexible enough to work with substantially different reactor architectures.

The company eventually plans to expand beyond magnetic-confinement technologies as well.

Fusion's supply chain is becoming a business of its own

The emergence of companies such as Fusionality points to an important shift in the fusion industry.

For years, most attention has gone to the companies trying to achieve the actual fusion reaction.

But commercialising fusion will require much more than a reactor breakthrough.

The industry will need specialised magnets, materials, cooling equipment, diagnostics, power electronics, control systems, simulation software, maintenance technology and eventually equipment capable of converting fusion energy into electricity.

That is creating a second layer of businesses around the reactor developers themselves.

Some companies are building components that could eventually help fusion plants produce grid electricity. Others are manufacturing specialised hardware.

Fusionality is betting that control and operations infrastructure will become another major category.

The model resembles what has happened in other technology industries.

As an industry matures, companies gradually stop building every component internally. Specialised suppliers emerge, standardised tools develop and businesses concentrate resources on their core intellectual property.

Fusion could eventually follow the same path.

Billions are flowing into fusion

The timing of Fusionality's launch is significant.

Private investment in fusion has accelerated as investors bet that technological advances in magnets, computing, materials science and artificial intelligence could move commercial fusion closer to reality.

Industry data cited in recent reports shows private fusion investment reached approximately $4.48 billion during the 12 months ending July 2026.

Since 2021, more than $14 billion has reportedly flowed into dozens of fusion companies.

That money is creating demand beyond the headline reactor developers.

Every new experimental device needs measurement systems.

Every reactor needs software.

Every plasma needs to be monitored.

And every company eventually attempting to produce electricity will need increasingly reliable control infrastructure.

Fusionality's opportunity therefore depends less on picking which individual reactor design wins and more on whether the overall fusion industry continues expanding.

The $3.7 million is only the beginning

Fusionality currently has a team of around seven people.

The new pre-seed funding will allow it to develop an initial set of carefully selected technologies before gradually expanding its product portfolio.

Felici has not publicly disclosed exactly which components will come first.

But the broader ambition is clear.

The startup wants to eventually provide a wide range of the technologies required to operate fusion devices.

That could include everything from simulation environments and data infrastructure to measurement and real-time control technology.

Building that platform will take time.

Commercial fusion itself also remains uncertain. Despite enormous investment and significant scientific progress, the industry has yet to demonstrate economically competitive fusion electricity operating continuously on the commercial grid.

That means Fusionality is building infrastructure for an industry whose ultimate market is still developing.

But that may also explain why the founders believe this is the right moment to start.

If fusion companies are going to move from experiments toward commercial machines, they will increasingly need suppliers capable of turning years of specialised research into repeatable industrial technology.

And Fusionality wants to become one of them.

The most interesting part of the fusion race may eventually be the companies that never build a reactor.

Billions of dollars are flowing into startups competing to create commercially viable fusion machines, but an entire technology ecosystem will have to grow around those reactors if fusion is ever going to become a real energy industry.

Fusionality is making a classic infrastructure bet.

Instead of predicting which reactor company wins, it wants to sell critical technology to many of them.

The bigger opportunity is standardisation.

If fusion companies are genuinely rebuilding large portions of similar control infrastructure internally, specialised suppliers could reduce development costs and allow reactor companies to focus their engineers on the technologies that actually distinguish their machines.

AI will be part of that transition, but Fusionality's approach also offers a useful counterpoint to today's AI hype: some of the most consequential applications of artificial intelligence may not involve replacing entire systems.

They may involve quietly helping extremely complicated machines work better.

For Fusionality, the $3.7 million pre-seed round is small compared with the billions being invested in fusion reactors.

But if fusion becomes a commercial industry, the companies building its picks, shovels — and control systems — could become just as important as the companies chasing the breakthrough itself.