UK and US Link AI Supercomputers to Speed Up Fusion Research

The UK Atomic Energy Authority and the US Department of Energy's Princeton Plasma Physics Laboratory have signed a Joint Declaration of Intent to link their fusion-focused supercomputing platforms, UKAEA's SUNRISE and PPPL's STELLAR-AI, through a partnership called the SUNRISE–STELLAR-AI Federation. The agreement, signed at the Global Fusion Policy Summit in London, aims to use combined AI training to accelerate fusion power plant design.

 

Why Linking Two AI Platforms Specifically Addresses Fusion's Data Scarcity Problem

 

Machine learning models generally require large volumes of high-quality training data to produce accurate predictions, but fusion research faces a genuine data scarcity constraint that many other AI application domains don't share, since experimental fusion facilities are extraordinarily expensive and few in number globally, meaning any single facility can only generate a limited volume of experimental data over time. By federating UKAEA's SUNRISE and PPPL's STELLAR-AI platforms, researchers gain the ability to train the same AI models on combined experimental data drawn from both UKAEA's MAST Upgrade facility in Oxfordshire and PPPL's NSTX-U facility in New Jersey simultaneously, effectively doubling the pool of real-world experimental data available for any given model training effort compared with either facility working independently.

That combined dataset directly addresses a specific technical limitation UKAEA's Rob Akers described: models built using data from a single machine "can struggle to make accurate predictions that apply across different machines," meaning genuine cross-facility model generalisation, a model's ability to make accurate predictions beyond just the specific facility it was originally trained on, requires training data spanning multiple distinct physical facilities rather than relying on a single machine's experimental results alone.

 

Read more: Planted Raises $32 Million to Build Solar Farms Using Robots on Uneven Terrain

 

Why the Near-Identical Reactor Designs at Both Facilities Matter for Cross-Facility Training

 

The release specifically notes "both MAST Upgrade and NSTX-U are compact spherical tokamaks with similar designs, making them well suited for joint AI training." That design similarity matters considerably for the federation's technical viability, since training a single AI model on data from two facilities with fundamentally different reactor geometries or operating principles would likely produce a model struggling to reconcile genuinely different underlying physical relationships between the two data sources, potentially undermining rather than improving the resulting model's predictive accuracy.

Because both facilities share the same fundamental spherical tokamak design category, the underlying physics governing plasma behaviour across both machines should be considerably more consistent, allowing a jointly trained model to identify genuine underlying physical relationships that hold true across both facilities rather than learning facility-specific idiosyncrasies that wouldn't transfer meaningfully to the other machine. That shared design similarity directly connects to both countries' next-generation fusion ambitions, with PPPL's Jonathan Menard specifically noting the collaboration aims to "accelerate the path to a compact fusion power plant," referencing concepts including the UK's Spherical Tokamak for Energy Production and PPPL's Spherical Tokamak Advanced Reactor, both building on the same compact spherical tokamak design category the federation's shared training data specifically targets.

 

Explore OneStop ESG Marketplace: Renewable Energy

 

Why the Digital Twin Ambition Represents a More Advanced Goal Than Simply Pooling Computing Power

 

Beyond combining raw computing capacity, UKAEA's Rob Akers described the federation's longer-term ambition as developing "digital twins of both machines to support the design of future fusion power plants." A digital twin, as the release explains, is "a detailed virtual copy of a real fusion machine, supported by regular data from fusion experiments, so researchers can test changes and predict how the machine will behave in software before applying experiments to hardware."

That distinction matters considerably: simply combining computing power or training data produces improved individual predictive models, but a functioning digital twin represents a more comprehensive, continuously updated virtual replica capable of simulating how proposed design changes would actually perform before any physical experiment or hardware modification is attempted. Akers specifically noted the federation would "use simulation to fill the gaps and extend those datasets into regimes we have yet to explore," suggesting the digital twin ambition extends beyond simply modelling already-observed experimental conditions into predicting how these fusion machines might behave under conditions neither facility has yet physically tested, a capability with direct value for designing future, larger-scale fusion power plants that will necessarily need to operate under conditions extending beyond what current smaller experimental facilities have already demonstrated.

 

Why the Government Backing and Infrastructure Context Signal Sustained National Investment

 

The release notes SUNRISE is "backed by £45 million from the UK government" and "forms the first phase of computing infrastructure underpinning the UK's first AI Growth Zone, established at UKAEA's Culham Campus." That specific government funding and infrastructure designation situates this transatlantic federation within a broader, sustained UK government investment strategy in fusion-specific AI computing capability, rather than representing an isolated research collaboration disconnected from wider national technology infrastructure planning, a pattern that connects to the broader theme of national energy security and technological sovereignty examined throughout recent climate and energy technology coverage, where governments increasingly treat next-generation energy technology development as a strategic national capability warranting dedicated infrastructure investment rather than purely relying on private sector or international collaboration alone.

 

Source: UK Atomic Energy Authority