From years to weeks: How a fusion startup aims to use AI to speed breakthroughs
Add Axios as your preferred source to
see more of our stories on Google.
Inside Commonwealth Fusion Systems' demonstration facility in early 2026. Photo: Courtesy of Commonwealth Fusion Systems
A trio of companies including Nvidia announced Tuesday that they're using AI to accelerate the development of fusion that's eluded innovators for decades.
Why it matters: Fusion, the energy that powers stars, has long had unrealized potential to provide vast swaths of clean, stable electricity for the AI boom and to cut emissions.
Driving the news: Commonwealth Fusion Systems, leveraging data from Siemens and an AI platform from Nvidia, is creating what it calls a "digital twin" of its demonstration machine to speed progress toward commercial fusion.
- "CFS will be able to compress years of manual experimentation into weeks of virtual optimization using the digital infrastructure developed by Nvidia and Siemens," said Bob Mumgaard, co-founder and CEO of CFS.
- By creating a digital twin of its Boston-area demonstration facility, CFS will be able to more quickly test, in the digital version, things that traditionally required testing in the hardware version, Mumgaard said.
The big picture: Tuesday's news offers a concrete example of AI's potential to accelerate energy innovation — one of the forces undergirding the AI energy boom, alongside surging power demand.
Inside the room: The announcement was made on the main stage of the Consumer Electronics Show in Las Vegas, with Mumgaard joined by Nvidia CEO Jensen Huang and Siemens CEO Roland Busch.
- "Fusion is debuting at CES," Mumgaard said on a call with reporters Monday.
- It's a shift, he said, that shows "fusion is no longer a science project — it's actually the next big thing in tech."
What they're saying: "The large tech companies are pulling forward innovation across several areas," said Charles Boakye, an equity analyst for sustainability and transition at Jefferies, in an email to Axios.
- He cited demand-response technology and large electricity load flexibility solutions.
Where it stands: CFS aims to hit a key technical milestone in 2027 — generating net energy at its demonstration machine — before producing commercial-scale fusion energy in the early 2030s.
Reality check: Mumgaard declined to specify how much time this type of AI application could accelerate those timelines.
- "There's this general accelerant that happens," Mumgaard said, while acknowledging it's "not as confident an answer of, 'Oh, it'll save three months.' "
- But he said hindsight will show a "meaningful" acceleration of the process.
Flashback: Tuesday's news follows another in October where Google DeepMind announced it is partnering with CFS to accelerate one particular part of the demonstration plant regarding the physics of plasma.
- If Tuesday's digital twin is the virtual airplane, the DeepMind partnership is more like a co-pilot, a CFS spokesperson said.
Follow the money: Both Nvidia and Google are investors in CFS.
What we're watching: CFS also announced it has installed the first of 18 huge magnets required of the demonstration facility.
- These aren't your everyday fridge magnets. They each weigh 24 tons — about as much as four adult African elephants.
What's next: The rest are on a schedule to be installed throughout the first half of this year.
