The growing private fusion industry has attracted billions into magnets, lasers, reactors, and power-plant engineering, while a newly funded Lausanne company is betting that another bottleneck will emerge in the less visible software needed to control and understand increasingly complicated machines.

Fusionality has raised CHF3 million in pre-seed financing from Founderful and Playfair to develop measurement, simulation, diagnostics, data, and control technology for fusion systems. Founded by Federico Felici and Jonas Buchli, the company is concentrating initially on magnetic-confinement machines.

The founders arrive from technical work rather than a generic application of AI to energy. Felici worked at EPFL’s Swiss Plasma Center before joining Google DeepMind, while Buchli also worked at DeepMind, and the pair were involved in research using reinforcement learning to control plasma inside EPFL’s experimental TCV tokamak.

Their commercial premise is that individual fusion programmes repeatedly construct substantial parts of the same operating infrastructure. Although reactor designs differ, every machine requires ways to measure conditions, simulate behaviour, analyse data, and translate control decisions into reliable physical actions.

Fusion develops a specialist software layer

Magnetic-confinement projects have to maintain extremely hot plasma inside carefully controlled magnetic fields, continually adjusting variables such as position, shape, and stability. As experimental devices become larger and more powerful, those control systems move from an academic research tool towards an industrial engineering requirement.

Fusionality intends to provide reusable components that customers can adapt to individual reactor designs rather than rebuilding a complete environment each time. Founderful describes the product area as operations infrastructure spanning measurement, simulation, and control, while the funding will support hiring and early commercial development.

As Techopia examined in August, expertise developed inside fusion programmes is beginning to escape the laboratory and become commercial engineering technology in its own right. Fusionality approaches the same transition from inside the reactor stack, where control and simulation capabilities may eventually become products sold across competing projects.

The company is not arguing that artificial intelligence should autonomously operate a commercial fusion plant. Felici has taken a more restrained position, treating AI as one component capable of optimising parts of the control system alongside conventional engineering rather than as a replacement for the entire architecture.

Simulation may prove just as important. Reactor teams need environments in which proposed control strategies can be tested before they reach physical machinery, while engineers need to reproduce failures and understand how adjustments to one subsystem affect another. A reusable simulation layer could reduce repeated engineering even where final control algorithms remain specific to each machine.

Standardisation is the difficult commercial step

Fusion remains a technologically fragmented market, with tokamaks, stellarators, inertial systems, and other approaches making fundamentally different engineering choices. Even companies pursuing magnetic confinement vary in geometry, magnets, heating systems, diagnostics, materials, and intended operating regimes.

Narrowing Fusionality’s initial market to magnetic confinement reduces that problem without removing it. The company still has to demonstrate that enough of the control and data layer can be shared across customers to create repeatable economics rather than a series of bespoke engineering engagements.

Reactor developers also have reasons to retain control capability internally because it sits close to core intellectual property and operating knowledge. A system able to interpret plasma behaviour and determine how a machine responds contains information that could influence performance, making some developers cautious about dependence on outside suppliers.

Keeping every layer proprietary carries costs of its own, particularly as companies duplicate infrastructure that provides little differentiation once the underlying engineering is mature. Aerospace manufacturers, for example, buy simulation software, flight computers, sensors, and specialist engineering systems without outsourcing the design of the aircraft itself.

Fusion could develop a comparable division of labour if suppliers prove that common components remain reliable across different reactor programmes. Such a market would allow developers to concentrate more resources on the physics and hardware distinguishing their machines while purchasing mature infrastructure elsewhere.

A broader industry is forming around the reactors

Private fusion investment is beginning to create opportunities beyond companies promising to generate electricity themselves. Magnet manufacturers, materials businesses, robotics specialists, diagnostic suppliers, software developers, and engineering contractors can all serve multiple programmes, reducing their dependence on correctly predicting which reactor concept reaches commercial operation first.

Fusionality remains at an early stage. Its financing is modest by energy-infrastructure standards, named commercial customers have not yet been disclosed, and commercial fusion generation itself remains some distance from routine operation. Software used inside future power plants will also have to demonstrate levels of reliability far beyond experimental AI demonstrations.

Even so, larger experimental machines will need control, diagnostics, simulation, and data infrastructure whether or not every project eventually produces grid electricity. The market for those tools can therefore begin forming before the sector resolves its largest scientific and commercial uncertainties.

The next useful evidence will come from deployments. If reactor developers are willing to purchase substantial portions of an operating stack from an external supplier, fusion will be taking another step towards the industrial structure found in aerospace and semiconductors, where specialist businesses beneath the headline manufacturer become indispensable to the finished system.