Early in 2014, a young Sam Altman entered the small Redmond, Washington office of Helion Energy while carrying textbooks on nuclear fusion. At the time, the business was a research operation, cautious and incremental like most deep-science startups are when they’re still attempting to establish the fundamentals of physics. Altman stayed there for a few days. Helion was moving more quickly after he departed. He made a $9.5 million investment a year later.
He contributed an additional $375 million by 2021. Around the same time, Altman was co-founding OpenAI, and the relationship between those two bets—on fusion and AI—was not accidental. He has stated unequivocally that an energy breakthrough is essential to the future of AI. For years, the question of where that energy originates has been quietly at the heart of the tech industry’s aspirations, and it is becoming more and more prominent.
| Category | Details |
|---|---|
| Total Industry Funding Growth | $1.7 billion in 2020 → $15 billion as of September 2025 (Fusion for Energy, EU body report) |
| Sam Altman / Helion | Invested $9.5 million in Helion Energy in 2015; added $375 million in 2021; serves as chairman — Helion based in Redmond, Washington |
| Microsoft–Helion Deal | Helion signed contract to deliver 50 megawatts of fusion power to Microsoft — faces financial penalties if it misses its 2028 timeline from Malaga, WA site |
| Key Corporate Investors | Microsoft (Helion), Google (Commonwealth Fusion Systems + TAE Technologies), Nvidia (Commonwealth Fusion Systems), SoftBank (Helion), Chevron, Bill Gates, Jeff Bezos |
| 2025 Annual Fusion Fundraising | $2.6 billion raised globally — a 180% jump year-over-year; number of fusion companies doubled from 23 to 53 between 2021–2025 |
| China’s Investment | ~$1.5 billion annual government investment in fusion — nearly double the US government’s 2024 research budget; EAST “artificial sun” reactor recently broke plasma stability threshold at extreme densities |
| AI Role in Fusion | DeepMind (Google) collaborating with Commonwealth Fusion Systems on AI plasma control; Nvidia and General Atomics building a digital twin model to simulate plasma at DIII-D facility in San Diego |
| Helion’s Method | Hurls two plasma rings together at ~1 million mph; currently recaptures ~96% of energy input; 7th-generation prototype “Polaris” first fired in late 2024 |
| Energy Potential | Per kg of fuel, fusion produces ~4 million times more energy than coal or oil; a gallon of seawater could yield as much energy as 300 gallons of petrol |
| Revised Timeline | Oak Ridge National Laboratory’s 2020 report projected a fusion pilot plant by early 2040s — director Troy Carter now believes mid-2030s is achievable |
It is hard to overlook the figures that have accumulated regarding fusion investment over the last five years. As of September 2025, total industry funding had increased from $1.7 billion in 2020 to $15 billion. Fusion developers raised $2.6 billion worldwide in 2025 alone, a 180 percent increase from the previous year. Between 2021 and 2025, the number of businesses engaged in fusion more than doubled, from 23 to 53. Google has supported TAE Technologies as well as Commonwealth Fusion Systems.
Commonwealth Fusion Systems also includes Nvidia. Microsoft has surpassed all others by entering into an actual supply contract with Helion that calls for the production of 50 megawatts of fusion-generated electricity by 2028, with financial penalties if the deadline is missed. It’s worth sitting with that final detail. Penalty clauses in a contract indicate that at least one large corporation has determined that fusion is sufficiently real to be included in a formal agreement rather than merely a portfolio.
It’s practical and unglamorous, which is why the tech sector is so concerned. Data centers use a lot of energy, and the AI systems that operate within them are growing bigger and more power-hungry every generation. Wind and solar power are helpful, but they are sporadic. Although current nuclear fission plants are dependable, they have lengthy construction schedules and problems with waste storage.
If it succeeds, Fusion will provide the grid with something it has never had: fuel that is partially derived from seawater and power that is always on, carbon-free, and free of long-lived radioactive waste. Compared to coal or oil, fusion fuel generates about four million times more energy per kilogram. Once commercially available, that ratio would do more than just reduce electricity costs. One of the strict physical restrictions on the amount of computing power available would be eliminated.
However, the engineering problem is still very difficult. In order to extract useful power from fusion, hydrogen plasma must be heated to temperatures of about 50 million degrees Celsius, which is hotter than the sun’s core, and kept stable for a sufficient amount of time. Helion’s method is remarkably inventive: instead of using fusion heat to boil water to power a turbine, it throws two plasma rings together at a speed of about a million miles per hour, directly harvesting electricity from the disruption of the magnetic field caused by the collision.
Approximately 96% of the energy input is recaptured by the company’s current machine, placing it near the break-even point, where the reaction produces more than it consumes. In late 2024, Polaris, the company’s seventh-generation prototype, was first put to use. Depending on one’s level of optimism, CEO David Kirtley’s decision to withhold results from the public can be interpreted as either cautious or telling.
The underlying science is becoming easier to understand thanks to AI. Google-owned DeepMind has been collaborating with Commonwealth Fusion Systems to use deep reinforcement learning to manage plasma stability, which is a recurring flaw in all fusion designs. A digital twin of the DIII-D fusion facility in San Diego has been constructed by Nvidia and General Atomics, enabling researchers to virtually test the reactor and simulate plasma behavior before anything breaks.
These contributions are not incidental. Timelines have begun to compress because machine learning is now being applied to one of the field’s most challenging problems—plasma control—with significant resources. Oak Ridge National Laboratory’s Troy Carter, who oversaw a 2020 study that predicted a fusion pilot plant by the early 2040s, now believes the mid-2030s is feasible.
China isn’t holding out. Its EAST reactor recently broke a plasma stability threshold at densities previously thought to be unachievable, and its government spends an estimated $1.5 billion a year on fusion research—nearly twice the US federal allocation. People keeping a close eye on this believe that Beijing’s competitive pressure is contributing to the acceleration of private investment in the US and Europe. It’s difficult not to feel that the industry has crossed some invisible threshold, going from a long-running scientific project that might eventually matter to something that a significant number of serious people now believe will happen within a working lifetime, as you watch the capital flow in and the timelines tighten. It’s still genuinely unclear if 2028 is feasible for Helion. However, there is no longer any significant uncertainty about the bet’s direction.
