US launches STELLAR-AI to slash nuclear fusion simulation time from months to real-time

US system to cut nuclear fusion simulation time from months to real-time

This configuration meets the technical needs of private companies developing commercial fusion power. The Princeton Plasma Physics Laboratory (PPPL) has launched a computing platform titled Simulation, Technology, and Experiment Leveraging Learning-Accelerated Research enabled by AI, or STELLAR-AI. This initiative aims to address the computational delays that currently hinder fusion energy research by integrating artificial intelligence with high-performance computing. Fusion research involves complex simulations of plasma behavior that can require several months to complete using existing infrastructure. "Fusion is a complex system of systems. We need AI and high performance computing to really optimize the design for economic construction and operation," said Jonathan Menard, deputy director for research at PPPL in a press release. STELLAR-AI is designed to shorten these timelines by connecting computing resources directly to experimental devices. This configuration allows for the analysis of data in real-time as experiments occur. A primary experimental connection for the platform is the National Spherical Torus Experiment-Upgrade (NSTX-U) at PPPL, which is scheduled to begin operations later this year. The hardware architecture of STELLAR-AI consists of a combination of central processing units (CPUs), graphics processing units (GPUs), and quantum processing units (QPUs). CPUs are utilized for standard computing tasks, while GPUs provide the parallel processing capabilities necessary for training AI models. QPUs are included to solve specific complex calculations that traditional computers cannot process efficiently. This hardware configuration is intended to support the technical requirements of private fusion companies working to develop commercial power solutions. Under the framework of the Genesis Mission, STELLAR-AI contributes fusion-specific computer codes and scientific models to a broader national system. The project supports the DOE's Fusion Science and Technology Roadmap, which outlines the development of an AI-Fusion Digital Convergence platform. The stated objectives of this roadmap include the commercialization of fusion power plants and the provision of energy for future computing and AI infrastructure. "One effort will create a digital twin of NSTX-U: a computer model that mirrors the physical machine so closely that scientists can test ideas virtually before running actual experiments," added the press release. "Another project, called StellFoundry, uses AI to speed the design of stellarators, a type of fusion device with a twisted, pretzel-like shape that some scientists believe could offer advantages over other designs." The design process for these machines involves analyzing large datasets to determine optimal configurations, a task the platform is intended to accelerate. The project is managed by PPPL and involves a network of national laboratories, academic institutions, and private companies. Partner laboratories include the UK Atomic Energy Authority (UKAEA), while academic participation includes Princeton University, the Massachusetts Institute of Technology, and the University of Wisconsin-Madison. Princeton University provides operational support, software engineering, and training for the infrastructure. Technical contributions are provided by NVIDIA, which is working to improve the performance of fusion-related computer codes, and Microsoft, which provides cloud integration through its Azure service. The platform also involves collaboration with private industry partners, including Commonwealth Fusion Systems, General Atomics, Type One Energy, and Realta Fusion. This coalition is intended to provide tools and AI models for the U.S. fusion industry to use in the development of power plant designs.

PPPL Launches STELLAR-AI Platform to Accelerate Fusion Energy Research

Newswise -- A new computing platform that pairs artificial intelligence (AI) with high performance computing aims to end the bottleneck holding back fusion energy research by speeding the simulations needed to advance the field. The project -- known as the Simulation, Technology, and Experiment Leveraging Learning-Accelerated Research enabled by AI (STELLAR-AI ) -- will be led by the U.S. Department of Energy's (DOE) Princeton Plasma Physics Laboratory (PPPL). STELLAR-AI will expand far beyond the Lab's walls, however, bringing together national laboratories, universities, technology companies and industry partners to build the computational foundation the fusion community needs. It can take months to run a single high-fidelity computer simulation or to train an artificially intelligent (AI) system capable of designing an ideal fusion system using existing infrastructure. STELLAR-AI is designed to reduce that timeline by using artificial intelligence. The platform connects computing resources directly to experimental devices, including PPPL's National Spherical Torus Experiment-Upgrade (NSTX-U), which is scheduled to go live this year, allowing researchers to analyze data as experiments occur. Building the Computational Foundation for Fusion Jonathan Menard, deputy director for research at PPPL, sees STELLAR-AI as a cornerstone of the U.S. fusion ecosystem: a dedicated, AI-driven research environment built specifically for the fusion energy mission. STELLAR-AI will pair speed with precision, accelerating the path to commercially viable fusion power. "Fusion is a complex system of systems. We need AI and high performance computing to really optimize the design for economic construction and operation," said Menard. "We want to link simulation technology and experiments -- in particular, NSTX-U -- with AI and partnerships to get to accelerated fusion." STELLAR-AI will achieve this goal by integrating CPUs, GPUs and QPUs in an ideal configuration of hardware for tackling the challenges facing private fusion companies as they race to bring a solution to market. CPUs, or central processing units, are standard computer chips that handle everyday computing tasks. GPUs, or graphics processing units, are specialized chips that excel at the parallel calculations needed for artificial intelligence. QPUs, or quantum processing units, use the principles of quantum physics to solve certain complex problems that would take traditional computers far longer to complete. A critical part of the Genesis Mission STELLAR-AI is part of the Genesis Mission, a national effort launched by executive order in November 2025 to use AI to speed up scientific discovery across DOE laboratories. "The Genesis platform is an integrated, ambitious system that will bring together the various unique DOE assets: experimental and user facilities, the supercomputers, data archives and, importantly, the AI models," said Shantenu Jha, head of PPPL's Computational Sciences Department. While Genesis provides that broad infrastructure, STELLAR-AI contributes fusion-specific computer codes, data and scientific models back into the national system. The project also aligns with the DOE's Fusion Science and Technology Roadmap, which calls for building an AI-Fusion Digital Convergence platform to accelerate commercialization of a fusion power plant, achieve U.S. energy dominance, and provide the abundant power needed to drive the next generation of AI and computing. Researchers plan to use STELLAR-AI for projects that span simulation, design and real-time experiment support. One effort will create a digital twin of NSTX-U: a computer model that mirrors the physical machine so closely that scientists can test ideas virtually before running actual experiments. Another project, called StellFoundry, uses AI to speed the design of stellarators, a type of fusion device with a twisted, pretzel-like shape that some scientists believe could offer advantages over other designs. Stellarator design requires sifting through enormous amounts of data to find the best configurations, a process that traditionally takes months or years and will greatly benefit from the STELLAR-AI platform. A Network of Public and Private Partners The strength of STELLAR-AI lies in PPPL's partnerships with DOE National Laboratories, AI and HPC companies, academic institutions, as well as fusion and engineering companies. The team includes world-leading capabilities from national laboratories, including PPPL and UKAEA as well as top universities such as Massachusetts Institute of Technology and University of Wisconsin-Madison. Princeton University, which manages the laboratory for the U.S. DOE's Office of Science, is also a key partner. Princeton will support operations, research software engineering, and user training for the STELLAR-AI infrastructure. Crucial technical support comes from tech giants like NVIDIA which is providing expertise to improve the performance of several critical fusion codes, and Microsoft, which will federate Azure's leading cloud capabilities. We also have direct collaboration with the fusion industry, including Commonwealth Fusion Systems, General Atomics, Type One Energy and Realta Fusion. This unique combination of partners will deliver proven AI models and key tools for the U.S. fusion industry. STELLAR-AI is just one of several initiatives that position PPPL as a hub for public-private collaboration in fusion energy. The laboratory's seven decades of plasma research, combined with experimental facilities like NSTX-U and computational expertise, have made it a destination for companies and research institutions seeking to accelerate fusion development.