How NVIDIA AI and simulation libraries and Siemens tools can accelerate fusion energy
SPARC, the cutting-edge fusion machine we’re building at Commonwealth Fusion Systems (CFS) headquarters in Devens, Massachusetts, is a massive metal machine. When we switch it on, electrical currents will surge through its superstrong magnets.
But alongside that physical machine will be digital representations of SPARC — compilations of data and simulations that can supercharge first how we design the machine, then how we plan experiments once it’s operational.
Two companies help supply us with that digital capability: AI and simulation libraries from NVIDIA and design and engineering expertise from Siemens. Siemens’ tools let us design SPARC and its components and manage all its data so we can rapidly improve our design and manufacturing processes. We can then import these designs into a physically accurate 3D digital twin of SPARC built with NVIDIA Omniverse libraries and the OpenUSD framework. This digital twin won’t just be a 3D design: it’ll also incorporate conventional and AI-accelerated models that let us simulate fusion physics.
These digital tools are a powerful enabler for complex physics and engineering projects like SPARC. Ultimately, that helps us fulfill our mission to bring fusion’s clean, abundant power to the grid as soon as possible.
Boosting fusion with AI and digital twins
SPARC is a type of fusion machine called a tokamak. Its superstrong magnets bottle up and control a superhot cloud of charged particles called a plasma — particles that fuse together and release energy. Digital tools are crucial to solving many of its challenges, like managing temperature extremes and the unruly plasma itself.
Fusion is a complicated process that demands complicated machinery and software. Sophisticated control systems govern how we ramp electrical current up and down in some of SPARC’s magnets during a pulse of fusion-generating activity. We use plasma simulation tools to understand how to control this plasma and enhance its performance. Many of these simulations are computationally demanding and too slow to run in day-to-day operations. This is where emerging AI tools can help, by speeding up simulations to run in milliseconds instead of months.
That’s why we’re working with NVIDIA to explore what’s possible.
NVIDIA AI infrastructure is powering the artificial intelligence race. The same infrastructure that enables modern large language models can also power the fusion AI models that CFS and our partners are developing. Additionally, Omniverse libraries can power digital twin technology that combine designs with conventional and AI simulations. That can help physicists in the SPARC control room compare real-world experimental data with simulation predictions, making it more intuitive to spot trends and use them to guide operations.
“Our digital twin supports a wide array of machine learning and artificial intelligence workflows,” said Tom Looby, Manager of Boundary Operations and Digital Twins at CFS. “This enables physicists to explore operational scenarios quickly while planning pulses or between pulses in the control room.”
The work begins with traditional computer simulations that, while powerful, demand massive computing resources and weeks to run. These simulations can be modernized to run on NVIDIA AI infrastructure, then further modernized by training an AI model to capture how plasma works. The result from these AI “surrogate” models: simulations that run millions of times faster.
CFS also is working with Google DeepMind on applying AI to SPARC, using its TORAX simulation software and exploring how AI could identify good ways to configure SPARC options like fueling, radio-frequency heating, and electrical current.
CFS received investments from NVIDIA and Google in our $863 million Series B2 funding round in 2025.
Siemens software for computer-aided design
CFS uses multiple Siemens software packages. For example, Siemens’ Teamcenter software offers a central data hub essential to rapidly improve new versions of our designs and our manufacturing processes. And Siemens’ Designcenter NX software for computer-aided design (CAD) helps CFS carefully ensure that all of SPARC’s components fit together during both assembly and operation, SPARC Chief Engineer Valeria Riccardo.
It’s a lot to manage. SPARC has more than 2 million parts — about as many as a Boeing 737 passenger jet.
SPARC components will shift as the tokamak runs. When the magnets get cold they contract, when the vacuum vessel gets hot it expands, electromechanical loads move some of SPARC’s magnets, and components attached to the magnets move with them. We can simulate those changes, import the results into Designcenter NX, and check there is sufficient clearance between components.
“The magnets move and take all that’s attached to them on a ride,” Riccardo said. “Designcenter NX is very powerful in this sort of analysis.”
Today’s digital tools, for engineering to simulation to operations, are starkly different. To our scientists and engineers, the data coursing through these computing systems is as real as the electrons flowing through our magnets.
To hear a little about what NVIDIA and Siemens think of this collaboration, check our press release on our digital twin news.
SPARC® and ARC™ are trademarks of Commonwealth Fusion Systems®.
