The UK Atomic Energy Authority and the US Department of Energy’s Princeton Plasma Physics Laboratory have agreed to explore linking two specialist AI supercomputers, creating a transatlantic computing platform intended to accelerate the design of future fusion power plants.
The proposed SUNRISE–STELLAR-AI Federation would connect the UK’s SUNRISE system with Princeton’s STELLAR-AI platform, allowing researchers in both countries to train artificial-intelligence models using data from leading fusion facilities.
A declaration of intent was signed at the Global Fusion Policy Summit in London, building on a memorandum of understanding agreed between the laboratories in June.
The collaboration will use experiments from the UK’s MAST Upgrade facility in Oxfordshire and Princeton’s NSTX-U machine in New Jersey. Both are compact spherical tokamaks with related designs, creating useful conditions for testing AI models across more than one experimental system.
Joe Milnes, executive director for engineering and computing at UKAEA, said: “Fusion is one of the great scientific and engineering challenges of our time.”
Combining the computing platforms would increase both the volume and diversity of data available for model training.
An AI system developed using results from one fusion machine may not perform reliably when conditions change. Training models against experiments from both countries could help researchers determine whether they generalise effectively and identify where assumptions fail.
The project forms part of a wider effort to reduce the cost and time involved in developing fusion technology.
A commercial fusion reactor would have to integrate plasma physics, advanced materials, heat management, magnets, control systems, fuel handling, robotics, maintenance, power conversion, and other complex engineering disciplines.
Advanced computing can allow researchers to test larger numbers of design options before expensive physical components or experiments are required.
The UK has made that capability a central part of its fusion strategy. The Government is investing £45m in SUNRISE, which is intended to become a major AI computing resource dedicated to fusion research.
It has also committed £1.3bn to the next phase of the STEP programme at West Burton in Nottinghamshire, where the UK plans to develop a prototype fusion power plant.
STEP is expected to support thousands of construction jobs during the 2030s and a substantial permanent workforce once the plant and surrounding industrial ecosystem are established.
The economic objective extends beyond electricity generation. Government policy is intended to create domestic capabilities in engineering, robotics, computing, specialist materials, project management, and manufacturing that can supply a wider international fusion industry.
Private investment in fusion has risen substantially as companies pursue different reactor designs and development models. Commercial power nevertheless remains unproven, and substantial scientific, engineering, regulatory, and financing hurdles remain.
That uncertainty has not prevented competition between countries seeking to turn research strength into industrial capacity. Britain has decades of expertise in fusion science but has repeatedly faced the wider challenge of converting research leadership into scaled commercial industries.
Collaboration with Princeton can accelerate scientific progress while still allowing each country to develop its own workforce, companies, facilities, and supply chains.
Regulatory coordination is another element of that international effort. Developers need clarity over how future fusion facilities will be approved if they are to make long-term capital commitments.
The UK’s regulatory approach distinguishes fusion from nuclear fission in several respects because the technologies have different fuel, waste, and accident characteristics, although major safety and environmental requirements remain.
The proposed computing federation is a research initiative rather than a commercial-power project, and there is no guarantee that AI will remove the hardest engineering obstacles facing fusion.
Its value lies in making better use of expensive experimental data and computing resources. Models trained across multiple facilities could help researchers eliminate weaker designs earlier, improve plasma control, and identify promising engineering configurations before larger sums are committed.
The agreement also shows how computing capability is becoming part of the industrial infrastructure surrounding fusion alongside laboratories, engineering, regulation, capital, and skills.
Commercial fusion remains a long-term prospect. The competition to develop the technologies, intellectual property, supply chains, and computing systems required to support it is already well under way.
You must be logged in to post a comment.