Google DeepMind Alumni Launch Fusionality to Accelerate Fusion Power for the Grid

TL;DR

• Fusionality, founded by Google DeepMind alumni, has emerged from stealth to build AI control systems and simulation tools purpose-built for fusion energy.
• The startup aims to solve fusion's plasma control and R&D bottleneck by giving tokamak and stellarator developers faster, AI-driven design and real-time operation.
• Backed by growing investor interest in fusion, Fusionality positions itself as a picks-and-shovels layer to help multiple fusion startups reach the grid faster.

From DeepMind to Star Power

A new player has entered the race to commercialize fusion energy, and it comes straight from the world of AI. Fusionality, a startup founded by Google DeepMind alumni, has launched with a mission to use artificial intelligence to crack one of clean energy's hardest problems: keeping a star stable on Earth.

While dozens of fusion startups are building reactors, Fusionality does not plan to build its own reactor. Instead, it wants to be the intelligence layer behind them, providing AI-powered control systems and high-fidelity simulation environments that speed up development timelines and improve reactor performance.

The founding team brings deep experience from DeepMind's landmark work on plasma control, including reinforcement learning systems that learned to shape and stabilize superheated plasma inside a tokamak. That research, once a scientific breakthrough, is now the foundation for a commercial product.

The Bottleneck Slowing Fusion Down

Fusion promises clean, virtually limitless power with no carbon emissions and minimal long-lived waste. The physics is proven. The engineering is not.

Inside a magnetic fusion device, plasma hotter than the core of the sun must be confined, shaped, and stabilized in milliseconds. Traditional control systems struggle with the turbulence, instability, and sheer complexity of the data. Meanwhile, designing and testing new reactor scenarios can take weeks of supercomputer simulation and costly experimental shots.

That is the bottleneck Fusionality is targeting. For cash-constrained fusion startups racing to demonstrate net energy gain and grid readiness, slow iteration cycles and plasma disruptions are expensive and dangerous. Better prediction and control could mean fewer failed experiments, longer plasma lifetimes, and faster progress toward commercial pilot plants.

An AI Operating System for Fusion

Fusionality is building two core products that work together.

The first is a real-time AI control platform designed to plug into existing tokamaks and stellarators. Using deep reinforcement learning, the system learns to manage magnetic coils with superhuman speed and precision, adjusting plasma shape, position, and stability thousands of times per second. The company says its controllers can anticipate instabilities before they happen and adapt to changing plasma conditions far faster than conventional PID controllers.

The second is a cloud-based simulation and training environment. It combines physics-informed neural networks with accelerated plasma models to let engineers test thousands of reactor scenarios virtually before ever firing a real shot. Think of it as a flight simulator for fusion operators — a place to train AI agents, validate control strategies, and de-risk new designs in hours instead of months.

Together, the tools are designed to create a feedback loop: more simulation data leads to smarter controllers, which leads to better real-world performance, which in turn improves the models.

Helping Startups Reach the Grid Faster

Rather than competing with reactor developers, Fusionality is positioning itself as a partner. Its platform-agnostic approach is meant to work across different magnetic confinement designs, making it attractive to the growing field of private fusion companies in the US, UK, and Europe.

Industry analysts say that model makes sense. With more than $9 billion invested privately in fusion and governments setting ambitious targets for first grid-connected pilot plants in the early 2030s, any technology that shortens development cycles has huge leverage.

For early customers, the pitch is simple: reach breakeven and grid milestones faster, with fewer hardware iterations and lower operational risk.

What Comes Next

Fusionality has not yet disclosed full funding details or named all of its early pilot partners, but the company is reportedly already working with select fusion labs and startups on live plasma trials and is hiring aggressively for roles in machine learning, plasma physics, and control engineering.

Challenges remain. AI controllers must meet nuclear-grade standards for safety, reliability, and explainability before utilities and regulators will trust them on a power plant. And simulation models still need to prove they can generalize across machines with different geometries and magnets.

But the timing is notable. As AI transforms everything from drug discovery to chip design, fusion — one of the most data-rich and simulation-heavy fields in science — may be next. If Fusionality succeeds, the alumni behind some of AI's biggest breakthroughs could help deliver its biggest payoff yet: clean, limitless power to the grid.

AndroGuider Team

Articles written by the AndroGuider team. We try to make them thorough and informational while being easy to read.