OpenAI Bets on Fusion Power That Doesn't Exist Yet

OpenAI is negotiating to buy fusion power from Helion Energy — 5 gigawatts by 2030 and 50 gigawatts by 2035. The catch: no private fusion company has ever produced net electricity. Helion would need to build 800 reactors in four years using technology that hasn't reached scientific breakeven. The deal reveals less about fusion's readiness than about how desperate AI companies are for power they can't get from existing grids.

The Deal That Tells on Itself

Sam Altman stepped down as board chair of Helion Energy on March 23. The same day, Axios reported that OpenAI and Helion are negotiating a deal that would guarantee OpenAI 12.5% of Helion's total power output.

The numbers are staggering. Five gigawatts by 2030. Fifty gigawatts by 2035. Each Helion reactor generates 50 megawatts. That means 800 reactors in four years, then 7,200 more by 2035.

Zero commercial fusion reactors exist anywhere on Earth today.

Altman, who invested in Helion's $500 million Series E round in 2021 and led additional fundraising since, has recused himself from negotiations. Helion told TechCrunch it hasn't announced any new customer agreements beyond existing deals with Microsoft and steelmaker Nucor. But confirming that Altman is leaving the board "to enable Helion and OpenAI to partner on future opportunities" isn't exactly subtle.

Why OpenAI Is Shopping for Imaginary Electrons

Here's what makes this story interesting. It's not a fusion story — it's an AI energy crisis story wearing a fusion costume.

OpenAI has announced plans for data center facilities requiring more than 30 gigawatts of total power — more than the peak electricity demand of all New England. AI data centers are projected to consume 9% of US electricity by 2030. Server rack power density has nearly doubled from 8 kilowatts in 2024 to 15 kilowatts in 2026.

The grid can't keep up. Interconnection queues in the US have ballooned past 2,600 gigawatts of proposed projects waiting for approval. In Virginia's Data Center Alley, 78% of residents blame the facilities for rising electricity costs. AI companies are scrambling for any power source they can lock down — natural gas, nuclear fission, off-grid generation, and now fusion.

OpenAI isn't betting on fusion because fusion is ready. It's betting because everything else is already spoken for.

Microsoft signed a similar power purchase agreement with Helion in 2023, targeting delivery by 2028. That deadline is two years away, and Helion hasn't demonstrated net electricity generation. The company's Polaris prototype reached 150 million degrees Celsius in February 2026 — a genuine milestone, and a first for private fusion — but still short of the 200 million degrees Helion says it needs for commercial operations.

The 800-Reactor Problem

Here's where the math gets uncomfortable.

Building 800 fusion reactors by 2030 requires manufacturing roughly 200 per year starting now. For context, the entire global nuclear fission industry — a technology we've understood for 70 years — added about 5 reactors per year over the past decade.

Helion's design does have advantages. Its field-reversed configuration reactors are smaller than traditional tokamaks. The company built a production facility in Everett, Washington, designed for high-volume manufacturing. Sofia Gizzi, Helion's director of production, told GeekWire that the factory's capacity extends well beyond the current Orion prototype, targeting 2030 and beyond.

But "designed for high-volume" and "producing 200 reactors annually of a technology that hasn't reached breakeven" aren't the same sentence. Fusion licensing pathways in the US are still being sketched. Site selection for the OpenAI deal hasn't begun. And as one industry analysis noted: the industry "has yet to publish bankable, project-level costs."

This is Musk's $25 billion Terafab all over again — the announcement reveals the crisis, not the solution.

Three Fusion Races, Three Different Stories

The OpenAI-Helion deal is the loudest fusion story this week, but it's not the only one. And depending on where you read about it, the entire fusion landscape looks different.

The American version: AI money rescues fusion. Private capital, venture-backed startups, and billionaire founders will crack the code faster than government programs. Helion, Commonwealth Fusion Systems, and TAE Technologies lead the charge. The market will solve it. The Chinese version: China is building BEST — the Burning plasma Experimental Superconducting Tokamak — on track for completion by 2027. The state-funded China Fusion Engineering Demo Reactor targets commercial power by 2030. The Economist reported this week that China is "a serious contender" in the race, combining state planning with industrial investment at a scale private companies can't match. The European version: The EU just committed €330 million to fusion energy through its 2026-2027 Euratom Work Programme. It's a fraction of what private US startups have raised, but it's patient money aimed at ITER and next-generation reactor designs — a bet on the long game while the Americans sprint.

The perception gap isn't about physics. It's about what fusion is for. In the US, fusion is an AI infrastructure play. In China, it's a strategic energy independence program. In Europe, it's a climate and decarbonisation tool. Same technology, three completely different civilisational priorities.

Meanwhile, fusion developers are going public for the first time, tapping pension funds and sovereign wealth funds for capital that venture firms alone couldn't provide. Reuters reported this week that the AI boom has widened funding sources for fusion startups beyond the usual Silicon Valley circuit.

What the Deal Actually Means

Strip away the gigawatt projections and the reactor math, and the OpenAI-Helion deal tells you three things:

One. AI companies have already committed to more electricity than the existing grid can deliver. They're not expanding supply — they're competing for it. Every gigawatt that goes to a data center is a gigawatt that doesn't go to homes, hospitals, or factories. As the US wind and solar sector showed, 59 gigawatts of clean power sits stalled in interconnection queues while AI companies jump the line with private generation. Two. The fusion industry's business model has shifted. It's no longer "solve clean energy for humanity." It's "solve clean energy for AI companies willing to pay above-market rates for dedicated supply." Microsoft, OpenAI, and Google are the new anchor tenants. The rest of us get whatever's left. Three. Nobody is asking the hardest question: what happens if fusion doesn't deliver on time? If Helion can't hit 200 million degrees and start generating net electricity by 2028, OpenAI's 5-gigawatt-by-2030 target evaporates. And AI's power demand won't wait. It'll get met with natural gas, revived coal plants, or the kind of grid strain that raises everyone's electricity bills.

The Energy Race Behind the AI Race

Fusion's promise has always been "too cheap to meter" electricity with zero carbon emissions. That promise is still real. Helion's 150-million-degree achievement is genuinely impressive. China's state-backed program and Europe's €330 million commitment show this isn't vaporware — serious institutions are spending serious money.

But the timeline pressure created by AI demand is warping the entire field. When OpenAI needs 30 gigawatts and the grid can't provide it, every energy source becomes a race against an artificial deadline — not a physics deadline, not a climate deadline, but a compute deadline.

The question isn't whether fusion will work. Most physicists now say it will. The question is whether it'll work fast enough to power the AI industry — or whether AI's hunger will devour every other energy priority in the queue first.

Eight thousand reactors by 2035. It's either the most ambitious clean energy target in history, or the most expensive proof that AI needs power faster than physics can provide it.