Sam Altman is stepping down as board chair of Helion Energy, the fusion startup he’s backed for over a decade, as the company enters early-stage talks to sell power directly to OpenAI. TechCrunch AI reports the potential deal could guarantee OpenAI 12.5% of Helion’s future production: five gigawatts by 2030 and 50 gigawatts by 2035.
Those numbers are staggering. And they tell us something important about where AI infrastructure is heading.
🔋 What’s on the Table
The deal, first reported by Axios and confirmed in part by TechCrunch AI, would make OpenAI Helion’s second major customer. Microsoft signed a similar agreement back in 2023 to buy fusion power starting in 2028.
Helion hasn’t confirmed the OpenAI talks directly. A spokesperson told TechCrunch the company hasn’t announced any new customer agreements beyond Microsoft and steelmaker Nucor. But the company did confirm Altman’s departure from the board, and the language was telling.
Sam is stepping down from Helion’s Board of Directors after more than a decade. This decision enables Helion and OpenAI to partner on future opportunities to bring zero-carbon, safe electricity to the world, Helion co-founder and CEO David Kirtley said in a statement to TechCrunch.
That’s about as close to a confirmation as you’ll get without one.
⚡ The Scale Problem
Here’s what stands out. If the reported figures are accurate, Helion expects to build roughly 800 reactors by 2030 and 8,000 by 2035. Each reactor generates about 50 megawatts of electricity. That’s an extraordinary manufacturing ramp for a technology that hasn’t reached commercial scale yet.
Helion is currently running its Polaris prototype. In February, the company hit 150 million degrees Celsius inside the reactor, closing in on the 200 million degrees it believes commercial operations require. Most fusion competitors are targeting early 2030s for their first commercial plants. Helion wants to get there first.
The startup’s approach is also different from most fusion companies. Instead of harvesting heat and running steam turbines, Helion uses magnets to convert fusion energy directly into electricity. Inside its hourglass-shaped reactor, plasma is fired from both ends, collides in the middle, gets compressed by magnets until fusion occurs, and the reaction pushes back on those magnets to generate power.
🧠 Why This Matters for AI
AI’s electricity appetite is becoming one of the industry’s defining constraints. Training and running large models requires massive amounts of power, and that demand is growing faster than the grid can keep up. OpenAI isn’t just thinking about chips and data centers anymore; it’s thinking about where the electrons come from.
This mirrors a broader pattern:
- Microsoft locked in Helion power in 2023
- Google has signed nuclear energy agreements
- Amazon has invested in small modular reactors
The race for AI-dedicated energy supply is real, and fusion represents the most ambitious bet on the board.
🔍 The Altman Factor
Altman’s involvement creates obvious conflict-of-interest questions: he’s a major Helion investor and was board chair, while also running the company that would become a major customer. His departure from the board follows the same playbook he used with Oklo, a small modular nuclear reactor startup. He stepped down from Oklo’s board last year to let it “explore strategic partnerships with leading AI companies, including potentially with OpenAI,” according to Oklo’s co-founder Caroline Cochran.
Pattern recognized.
Helion raised $425 million last year from investors including Altman, Mithril, Lightspeed, and SoftBank. The company has serious backing. The question isn’t whether it has funding; it’s whether fusion technology can move from prototype to commercial scale on the timeline these deals demand.
Five gigawatts by 2030 is four years away. That’s either visionary planning or a very ambitious promise. Either way, it signals that the biggest AI companies are planning for an energy future that doesn’t exist yet, and betting billions that it will.
For the full breakdown, check the original reporting at TechCrunch AI.