A new US-led computing initiative aims to remove one of the biggest obstacles to progress in fusion energy research: the sheer time and computing power required to model, design, and optimise fusion systems.
By tightly integrating artificial intelligence with high-performance computing and live experimental data, the project promises to dramatically accelerate the pace of discovery across the fusion sector.
The platform, called STELLAR-AI, is being developed under the leadership of the Department of Energy’s Princeton Plasma Physics Laboratory (PPPL).
Rather than serving as a standalone facility, STELLAR-AI is designed as a shared computational backbone for the wider fusion community, connecting national laboratories, universities, technology firms and private fusion companies.
Ending the simulation bottleneck
Modern fusion energy research relies heavily on advanced computer simulations to predict plasma behaviour, test reactor designs and refine operating conditions.
These simulations are extraordinarily complex, often taking weeks or even months to complete on conventional computing infrastructure. Training AI systems capable of assisting with fusion design can be just as time-consuming.
STELLAR-AI aims to compress these timelines by orders of magnitude. The platform directly links powerful computing resources to experimental fusion devices, including PPPL’s National Spherical Torus Experiment-Upgrade (NSTX-U), which is expected to resume operations this year.
This connection allows researchers to analyse experimental data in near real time, rather than waiting until after experiments conclude.
By closing the loop between simulation, AI analysis and live experiments, STELLAR-AI transforms fusion research from a stop-start process into a continuous, adaptive workflow.
A purpose-built digital foundation for fusion
At its core, STELLAR-AI is designed specifically for the unique challenges of fusion energy research.
Fusion systems involve tightly coupled physics processes, engineering constraints and economic considerations. Optimising them requires both speed and precision.
To meet this need, the platform integrates multiple types of computing hardware. Traditional central processing units handle general workloads, graphics processing units accelerate AI training and inference, and emerging quantum processing units are incorporated to explore problems that may benefit from quantum algorithms.
This hybrid approach creates a flexible computing environment capable of tackling everything from plasma turbulence modelling to reactor design optimisation.
The ultimate goal is to shorten the path from scientific insight to commercially viable fusion power by enabling faster, smarter design decisions.
Digital twins and faster reactor design
Researchers plan to deploy STELLAR-AI across multiple high-impact projects. One flagship effort is the creation of a detailed digital twin of NSTX-U.
This virtual replica will allow scientists to test control strategies and experiment designs in software before applying them to the physical machine, reducing risk and improving efficiency.
Another major initiative, known as StellFoundry, focuses on stellarators – fusion devices with complex, twisted magnetic geometries.
Designing stellarators traditionally requires scanning vast parameter spaces, a process that can take years. By applying AI-driven optimisation on STELLAR-AI, researchers expect to identify promising designs far more quickly.
Part of a national AI science push
STELLAR-AI operates within the broader Genesis Mission, a nationwide Department of Energy initiative launched to accelerate scientific discovery across federal laboratories through the use of artificial intelligence.
Genesis provides shared access to experimental facilities, leadership-class supercomputers, data repositories and advanced AI models.
Within this ecosystem, STELLAR-AI contributes fusion-specific simulation codes, curated datasets and scientific models, strengthening the national infrastructure for fusion energy research.
The project also supports federal goals outlined in the Fusion Science and Technology Roadmap, which calls for AI-driven digital platforms to speed fusion commercialisation and strengthen US energy leadership.
A powerful public-private collaboration
A defining feature of STELLAR-AI is its extensive partnership network. The project brings together expertise from national laboratories, leading universities, and global technology companies, as well as direct collaboration with the private fusion industry.
Academic partners contribute cutting-edge research and training, while technology firms provide advanced hardware, cloud infrastructure and performance optimisation.
Fusion companies gain access to validated AI models and simulation tools that can directly support reactor development and commercial timelines.
This collaborative structure ensures that advances made through STELLAR-AI flow rapidly from the laboratory to industry, strengthening the entire fusion ecosystem.
Accelerating the future of fusion
By uniting AI, high-performance computing and live experiments into a single, purpose-built platform, STELLAR-AI represents a major step forward for fusion energy research.
If successful, it could significantly shorten development cycles, lower costs and bring practical fusion power closer to reality – helping unlock a new era of clean, abundant energy.
