AI Datacenters Need Nuclear Fusion

The environmental impact of artificial intelligence (AI) computing is huge. Asking ChatGPT or any AI model a question is like driving a Mack truck to the corner store for a pint of milk.

AI’s power needs are so voracious that the U.S. Department of Energy (DoE) stated in late 2024 that “datacenter load growth has tripled over the past decade and is projected to double or triple by 2028.” Its 2024 Report on U.S. Data Center Energy Use also said that “Datacenters consumed about 4.4% of total U.S. electricity in 2023 and are expected to consume approximately 6.7% to 12% of total U.S. electricity by 2028.”

Such requirements raise questions about where all that energy will come from. A striking development is emerging as a possible answer: nuclear fusion. Two tech stalwarts—Microsoft and Google—have signed agreements to purchase power from yet-to-be-built fusion reactors, by 2028 for Microsoft, and by the early 2030s for Google. Microsoft is turning to an Everett, WA, company, Helion, while Google is teaming up with Devens, MA-based Commonwealth Fusion Systems.

In addition to the power agreements, the list of technology companies and individuals who have invested in fusion development companies is growing, and includes Google, Nvidia, Japan’s SoftBank, Japan’s NTT, the Gates Frontier, and others. And then there’s OpenAI CEO Sam Altman, who is not only an investor in Microsoft fusion provider-to-be Helion, but who doubles as Helion’s chairman. Amazon chairman Jeff Bezos has financially backed Vancouver-area fusion firm General Fusion since 2011.

Spurred by their full-throttle flight into AI, IT outfits are pairing with the energy world in a big act of industry convergence. These combinations includes not just the brass-ring reach for fusion, but also nuclear fission. Microsoft is helping to restart the Three Mile Island nuclear site in Pennsylvania, for example, while Amazon, Google, and Oracle have committed to the construction of small modular reactors near AI datacenters.

However, it is the commitment to fusion, long heralded as a viable energy source with little to no carbon emissions, that is most notable.

Nuclear fusion is regarded as a Holy Grail of cheap, clean, “limitless,” affordable energy. Fusion combines atoms, as opposed to the fission of civilian nuclear power, which splits atoms. Compared to fission, fusion’s radioactive waste is extremely short-lived. If and when perfected, it could greatly diminish the role of carbon-emitting fossil fuels. (Both nuclear fusion and fission are carbon-free in their generation process.)

The knock on fusion energy is that it has been 20 years in the future for the last 70 years. It has remained in the distance ever since the idea of using it for civilian electricity entered the mainstream with the United Nation’s Atoms for Peace conference in Geneva in 1958.

While several government-backed fusion projects around the world have made varying degrees of progress, they are still a long way from producing a machine that could start delivering electricity to grids. One of the best known, the multinational ITER (International Thermonuclear Experimental Reactor) in Cadarache, France, has suffered many delays and is not expected to be ready to produce energy until at least 2040.

Another fusion project, the DoE’s National Ignition Facility (NIF) at Lawrence Livermore National Laboratory, Livermore, CA, has made impressive advances but is designed as a research facility. NIF and ITER use different approaches, with NIF based on lasers and ITER on a torus (donut-shaped) design called a tokamak, using magnets. Both draw on principles of the Sun, aiming to hit temperatures of around 100 million degrees C to fuse isotopes of hydrogen.

An underlying issue is that until recently, most fusion endeavors take more energy to trigger a fusion reaction than the amount of energy produced. In the parlance of thermodynamics, the “gain” has been below zero.

Even at NIF, which over the last three years has shown a positive gain on nine different occasions, the gain has not been large enough for the purposes of commercial electricity production.

However, things could be rapidly changing on the fusion landscape. In addition to government projects, there are now at least 53 privately held fusion development companies, according to the Washington, D.C.-based Fusion Industry Association.

Some of these companies say they could be in a position to deliver fusion-based power within the next several years, in no small measure to Big Tech money aiming to help them over fusion’s Himalayan-sized hump.

OpenAI’s Altman has been investing in Helion since 2015, led a $500 million round of investment in 2021, and participated in a $425 million round in January 2025.

Helion claims to be so close to cracking fusion that Microsoft (which has invested billions in OpenAI) has signed a power purchase agreement to receive fusion electricity from Helion by 2028. Helion and Microsoft signed the deal in May 2023; Helion in July 2025 started construction in Malaga, WA, on the commercial fusion reactor that will supply Microsoft.

“Fusion represents an inspiring frontier in the world’s pursuit of clean and abundant power,” Microsoft chief sustainability officer and corporate vice president Melanie Nakagawa said at the time of the Malaga groundbreaking.

The Microsoft agreement with Helion involves an intermediary power distributor, Constellation Energy, a Baltimore-based transmission and nuclear fission company that will be restarting Three Mile Island for Microsoft. Constellation could find other sources of electricity for Microsoft if the Helion plant is not ready.

Helion is sticking to its 2028 target. “We have good confidence in the technology that we’re going to make it,” Helion co-founder and head of research and development Anthony Pancotti told Communications.

Look, No Turbine

Helion’s approach to fusion differs from those taken by NIF and ITER in several ways. Whereas NIF and ITER will use heat created by neutrons emitted during the fusion process to drive a steam turbine that would generate electricity, Helion’s machine, currently called Polaris, will emit electricity directly to the grid, no steam turbine necessary.

The Helion approach is called “aneutronic” (without neutrons) because it skips the thermal step of bombarding a blanket (typically containing lithium) with neutrons to capture heat for steam turbine generation, and to serve other purposes.

“Lithium blanket seems like a very challenging technology,” said Helion’s Pancotti. “It’s one of the aspects that our approach gets to step around.”

The direct production of electricity means Helion will not require the same size energy “gains” as with the thermal approach to fusion. “The amount of gain we have to produce is on the order of two or three; single-digit kind of gains in order to make a system that’s viable,” Pancotti said.

Taking the Pulse

Helion says its process is more efficiently than other fusion projects is because it does not have to achieve “ignition,” the state in which a fusion reaction self-sustains. Helion is taking a “pulsed” (magneto-inertial) approach of compressing and expanding fusion plasma, which Pancotti likens to the way a diesel engine operates.

By comparison, NIF is shooting 192 lasers at a tiny pellet of hydrogen isotopes deuterium and tritium to coax them into fusing and maintaining the reaction. The approach is called inertial confinement. ITER is magnetically confining deuterium and tritium in a giant tokamak torus to elicit the same reaction, a process called magnetic confinement.

Helion claims it has already proven the fusion part of the equation via its previous prototype, Trenta. Unlike the deuterium/tritium formula of NIF and ITER and many others, Helion is fusing deuterium with a rare isotope of helium called helium-3 (“helion” is another name for the nucleus of helium-3). Helium-3 is so uncommon that some have suggested mining it from the Moon. Helion’s method is more earthly: it obtains the helium-3 by first fusing two deuterium atoms, thus performing a fusion double act.

All that’s left is to magnetically convert its pulsed fusion into electricity, which Pancotti said should happen “later this year” with Polaris, a machine he called “our seventh and final prototype.”

A follow-up News article will look at Google’s deal with Commonwealth Fusion, at some gargantuan sized AI datacenters under construction, and speak with a leading fusion scientist at Lawrence Livermore National Laboratory.

Mark Halper is a freelance journalist based near Bristol, England. He covers everything from media moguls to subatomic particles.