Why the AI Industry Is Betting on Fusion Energy

When Sam Altman arrived at Helion Energy’s small Redmond, Wash., office in early 2014, nuclear-fusion textbooks tucked under his arm, the company was focusing its efforts on research and development. By the time he left, several days later, he had persuaded the fusion-energy startup to chart a more aggressive path toward deployment, CEO David Kirtley recalls. A year later, Altman, who was co-founding OpenAI around the same time, invested $9.5 million in Helion, taking the role of chairman. He plowed a further $375 million into Helion in 2021, making it one of the largest personal bets in his multibillion-dollar portfolio.

Once a government-led pursuit, nuclear fusion is now a private-capital race, much of it financed by the same people building energy-hungry AI and pursuing the goal of creating systems with human-like intelligence, known as artificial general intelligence (AGI). The fusion-energy industry’s total funding has jumped from $1.7 billion in 2020 to $15 billion as of September 2025, according to a report by E.U. body Fusion for Energy. Alongside Altman, who has said AI’s future depends on an energy breakthrough, investors in Helion include OpenAI funder SoftBank as well as Facebook co-founder and early Anthropic backer Dustin Moskovitz. Nvidia has backed Helion rival Commonwealth Fusion Systems (CFS). So too has Google, which has also invested in another player, TAE Technologies. “AI is a big driver [due to] the energy needs … to power their data centers,” says Troy Carter, director of Oak Ridge National Laboratory’s fusion-energy division.

Recent engineering progress and the flood of cash from investors willing to chase moon shots have some firms promising grid power within years rather than decades. They must still prove the technology works, but if fusion delivers, it would provide carbon-free power without solar and wind’s seasonal fluctuations or nuclear fission’s long-lived radioactive waste—a breakthrough that wouldn’t just lower power bills, but reshape what’s possible.

Fusion, the same reaction that powers the sun, makes energy via the opposite process of today’s nuclear power plants, joining light atoms rather than splitting heavier ones. Deep in a star’s core, this occurs in plasma, a super-hot, electrically charged gas. Re-creating that on earth has proved the mother of all engineering problems.

For decades, even as scientists were able to spark a fusion reaction, it generated less energy than required to heat the plasma, known as scientific break-even. But in 2022, researchers at Lawrence Livermore National Laboratory made history. Using giant lasers to briefly crush a tiny fuel pellet, they demonstrated for the first time a fusion reaction that generated more energy than was used to heat the plasma. No private company has hit this milestone.

If or when they do, getting power onto the grid will require going a step further: generating not just more energy than was used to heat the plasma, but enough to power the entire generator, known as engineering break-even. Of the firms racing to hit that milestone, Helion is running to the most bullish schedule. The company expects a commercial version of its machine to provide electricity by 2028 from a site in Malaga, Wash., where construction began in July. Helion has already signed a deal to sell 50 megawatts of fusion power to Microsoft and faces financial penalties if it veers from its timeline.

Unlike most fusion efforts, which boil water to spin a turbine, Helion plans to harvest electricity by hurling two plasma rings together at about a million miles an hour. The collision would trigger fusion, perturbing a magnetic field, which in turn produces power. Kirtley says the setup currently recaptures about 96% of its energy input—sort of like how an EV uses regenerative braking to recharge a bit during a trip. That puts it within a hair of break-even already. Polaris, Helion’s seventh-generation prototype, was slated to demonstrate engineering break-even in 2024. The prototype was fired up for the first time late that year. Kirtley declined to share results.

Kirtley, who credits Altman with encouraging him to “go faster and at bigger scale,” envisions not only being first to build a fusion plant. “Our goal is to … build a generator per day and deploy fusion systems all over the world. And do that quickly.”

That kind of ambitious thinking has credibly brought fusion’s arrival closer, Oak Ridge’s Carter says. In 2020 he led a Department of Energy report that said a pilot nuclear-fusion plant could be built by the early 2040s, but he now thinks it’s possible that goal could be achieved by the mid-2030s. And beyond capital, AI is a useful tool for scientific progress. “The advent of AI has made some very challenging problems in the plasma space more accessible,” says Nuno Loureiro, director of MIT’s plasma science and fusion center.

Read More: A Nuclear Fusion Breakthrough May Be Closer Than You Think

Even if Helion’s plan were to run a few years behind schedule, it could still hit many world firsts. And if its approach doesn’t pan out, others are lining up right behind it.

California-based startup Pacific Fusion says it has designed a machine that would hit engineering break-even using the same approach as the Lawrence Livermore National Laboratory’s device. It publicly launched in 2024, revealing $900 million from investors including former Google CEO Eric Schmidt and Microsoft AI CEO Mustafa Suleyman.

CFS, which spun out of MIT in 2018 to become the best-funded fusion startup, is pursuing a different approach: creating a magnetic bottle that holds an ultra-hot plasma in place. CFS is building a pilot that it believes will pass scientific break-even in 2027. The company is so bullish that it has begun work in parallel on a commercial plant that it expects to deliver that energy to the grid in the early 2030s. Google has already agreed to buy 200 megawatts. “[Having] these big hyperscalers behind us is really helpful,” says Brandon Sorbom, CFS’s co-founder and chief science officer, adding that it signals to suppliers who provide superconducting magnets and other difficult-to-manufacture materials that “this isn’t a one-off science experiment.” (Investors in Commonwealth Fusion Systems include TIME co-chair and owner Marc Benioff.)

Meanwhile, New Zealand–based upstart OpenStar generated plasma in late 2024 with a relatively modest $10 million in funding and has since raised a further $14 million. Its prototype flips the “magnetic bottle” concept inside out, with an ultra-strong magnet at the core of the reactor, around which the plasma is confined.

While optimistic about the number of startups vying to become industry leaders, Carter warns that a high-profile failure could spook investors and dent the field’s credibility. “You hope the hype doesn’t get too much; [that] a failure of one of the more visible companies does not pull the plug on progress that we have elsewhere,” he says.

Still, fusion can’t come fast enough for the likes of Google and Microsoft. Both are building new data centers to power AI, even as Microsoft targets being carbon-negative by 2030 and Google targets net-zero. The data centers powering AIs run 24/7; without an energy-storage breakthrough, variable wind and solar won’t reliably cover that load. And supply is tight: U.S. power generation has barely budged since 2010, the year it was overtaken by China as the world’s largest electricity producer. Even fossil fuel might struggle to scale as compute soars. Altman and Nvidia’s Jensen Huang now call energy the key bottleneck.