Commonwealth Fusion Systems (CFS) said on CES 2026 which has installed the first magnet in its Sparc fusion reactor, a demo device that hopes to turn on next year.
The magnet is the first of 18 that, when the reactor is complete, will form a donut-like shape that will generate a strong magnetic field to compress and compress the superheated plasma. If all goes well, the plasma will release more energy than it needs to heat and compress.
After decades of promises and delays, fusion power seems just around the corner – CFS and its competitors locked in a race to deliver the first electrons to the grid in the early 2030s. If the pot pans out, fusion power could unlock nearly unlimited clean energy in a package that resembles traditional power plants.
The key components of Sparc’s magnets are complete, and the company expects to install all 18 by the end of the summer, said Bob Mumgaard, CFS founder and CEO. “It’s going to be bang, bang, bang for the first half of this year as we integrate this revolutionary technology.”
When installed, the D-shaped magnet will sit upright in a 24-foot wide, 75-ton stainless steel circle known as a cryostat, which set last March. The machines themselves weigh about 24 tons each and can generate 20 tesla magnetic fields, about 13 times stronger than a typical MRI machine. “It’s the type of magnet that can be used to, like, lift an aircraft carrier,” Mumgaard said.
To achieve this power, the magnet will be cooled to -253˚ C (-423˚ F) in order to safely conduct currents in excess of 30,000 amps. Inside the donut, the plasma will burn at more than 100 million degrees C.
To work out the many kinks before Sparc is turned on, CFS said on Tuesday that it is working with Nvidia and Siemens to develop a digital twin of the reactor. Siemens is providing design and manufacturing software, which will help the company collect data to feed into Nvidia’s Omniverse library.
Techcrunch event
San Francisco
|
13-15 October 2026
This won’t be CFS’s first simulation — the company has run many simulations to predict the performance of various parts of the reactor — but existing efforts provide results that are out of the question, Mumgaard said. With digital twins, he said, “these are no longer isolated simulations that are only used for design. They will be together with the physical in every way, and we will constantly compare them.”
The hope is that CFS can conduct experiments or tweak parameters on the digital twin before applying to Sparc itself. “It will work together so we can learn from the machine even faster,” he said.
Building Sparc has been a costly endeavor. CFS has raised nearly $3 billion to date, incl $863 million Series B2 round in August that included investments from Nvidia, Google, and nearly three dozen other investors. The company’s first commercial-scale power plant, Arc, will be the first. As a result, it will cost several billion dollars more, CFS estimates.
Mumgaard hopes that digital twins and AI technology will help the company deliver fusion power to the network faster. “As the machine learning tool becomes better, as the representation becomes more precise, we can see it becomes faster, which is good because we have an urgency for the fusion to get to the grid,” he said.
Follow all TechCrunch coverage of the annual CES conference here.