Commonwealth Fusion Systems, Nvidia, and Siemens are developing a digital twin that utilizes artificial intelligence to accelerate the commercialization of fusion. They are pairing Siemens’ advanced engineering tools with Nvidia’s AI and simulation technology to create a digital twin of CFS’ SPARC, the fusion demonstration facility that the company is developing in collaboration with MIT.
CFS said last year that the SPARC plant in Massachusetts could begin producing energy sometime after this year.
Bob Mumgaard, CFS’ CEO, announced the collaboration today on the main stage of CES, the tech industry’s annual trade show in Las Vegas. The digital twin is expected to allow the company to “compress years of manual experimentation into weeks of virtual optimization,” Mumgaard said in the press release accompanying the announcement. This is because it provides engineers with a close digital representation of the physical machine, allowing them to both run simulations and test hypotheses.
“For the vast majority of history… you build a piece of hardware and have very little data coming from it….You also didn’t have anything in software that represented the plant,” Mumgaard explained in a press briefing with reporters before the announcement, adding that the lack of insight is fine for most industrial machines. “But in processes that have lots of complexity…you really want to have insights into what’s happening inside your machine.”
While research on fusion has been ongoing for decades, actually creating a reaction that creates more energy than is put in so far remains out of reach. The process involves controlling unstable plasma at extremely high temperatures, which makes physical testing a slow and expensive affair. But the rise of both AI and high-performance computing has led to hope that tools like digital twins could accelerate things.
A 2024 report summarizing existing research found that AI has the potential to significantly shorten the “years-to-decades-long” manual process of “building machines and testing ideas, generating datasets, and then choosing the next design based on empirical scaling.” In addition to digital twins, the report also highlighted other applications, such as predicting how materials behave in harsh conditions to create more resistant components, as well as optimizing the manufacturing of high-temperature superconductors.
With the dramatic increase in energy demand, fueled in part by the AI boom, the promise of fusion has become more and more attractive, and fusion startups are increasingly embracing AI in their race towards commercialization.
Commonwealth Fusion Systems is just the latest of a series of fusion startups that have announced digital twins over the past months. In October, for instance, General Atomics unveiled its own digital twin using the same Nvidia platform. This followed the lead of U.K.-based Tokamak Energy, which said it would begin integrating a digital twin software into testing workflows in 2024.
But CFS is certainly the richest and most prominent startup to embrace AI for fusion. The company has raised nearly $3 billion since it was founded in 2018; in August, NVentures, Nvidia’s venture capital arm, joined its $863 million Series B2 round. And it has secured two of the world’s largest fusion power purchase agreements: an agreement for 200 megawatts with Google and a $1 billion one with Italian energy giant Eni.
The company aims to start running the demonstration facility SPARC by 2027, and to bring its first commercial-scale plant — known as ARC, in Virginia — online in the early 2030s.
