Sam Altman’s AI company is in talks to buy electrity from Sam Altman’s fusion startup. - The Verge

Sam Altman is on both sides of a deal that could shape how OpenAI powers its data centers for the next decade — and the governance mechanism designed to resolve that tension is a board resignation that leaves his financial stake intact.

Axios reported this morning, citing a single unnamed source familiar with the talks, that OpenAI, the AI lab Altman leads, is in advanced discussions to purchase electricity from Helion Energy, the fusion startup he has backed since 2015. Altman led Helion's $500 million Series E in 2021 and participated in its $425 million Series F in January 2025, which valued the company at $5.4 billion. His exact stake has never been publicly disclosed by Helion or Altman. A CNBC estimate — not confirmed by either company — put the value at approximately $375 million, but that figure predates the January 2025 Series F and has not been updated or verified since. Treat it as an unconfirmed data point, not a current holding.

To clear a path for the deal, Altman is stepping down as Helion's board chair — a position he's held for more than a decade. "Sam is stepping down from Helion's Board of Directors after more than a decade," Helion CEO David Kirtley confirmed to TechCrunch. "This decision enables Helion and OpenAI to partner on future opportunities to bring zero-carbon, safe electricity to the world." OpenAI declined to comment.

The step-down is real governance. But Altman retains his equity — which appreciates directly if OpenAI, the company he runs, commits to buying Helion's power at scale. That's not a resolved conflict; it's a managed one.

What a single anonymous source says about scale

Per Axios — and picked up by The Verge and The Next Web from the same original report — the deal framework calls for OpenAI to receive an initial 12.5% of Helion's power production, targeting 5 gigawatts by 2030 and 50 gigawatts by 2035. These figures come entirely from one anonymous source. Helion has not confirmed them. The talks are described as early — site selection hasn't happened, conditions remain unmet.

Working from those Axios-sourced production targets — again, a single anonymous source, unconfirmed by Helion — TechCrunch's Tim De Chant calculated that each Helion reactor produces roughly 50 megawatts. On that projection, 5 gigawatts by 2030 would require approximately 100 reactors built in four years; 50 gigawatts by 2035 would require 1,000. To be clear: this is arithmetic derived from an unconfirmed anonymous leak, not a disclosed Helion roadmap. Helion has published no such build schedule. The math is useful for contextualizing the ambition, not for holding the company to it.

What Helion has actually committed to: one signed power purchase agreement, a 50 megawatt deal with Microsoft announced in 2023, delivery expected 2028. That's the only confirmed production target on the books.

Going from a 50 MW commitment to a 50,000 MW commitment — a 1,000x jump — while the underlying technology hasn't yet demonstrated net energy output takes some audacity to sign. Whether it reflects genuine confidence in Helion's roadmap or optionality hedging is something only Altman and OpenAI's board know.

Where Helion actually stands

Helion's approach to fusion is distinct from the tokamak designs being pursued by companies like Commonwealth Fusion Systems, the MIT spinout backed by Google and others. Helion uses a field-reversed configuration reactor and a deuterium-helium-3 fuel cycle intended to convert fusion energy directly to electricity via magnets — bypassing the steam turbine step that reduces efficiency in most nuclear designs.

In February 2026, Helion announced its Polaris prototype reached 150 million degrees Celsius during deuterium-tritium testing — a milestone its CEO called "record-setting." The Department of Energy's Jean Paul Allain called the results evidence of "strong progress." But D-T testing is a stepping stone: commercial D-He3 operations require temperatures closer to 200 million degrees Celsius. And reaching the right temperature is not the same as demonstrating net electricity output, which Helion has not yet done.

As TechCrunch reported at the time of the Polaris announcement, Helion is racing toward its 2028 deadline for the Microsoft PPA. The OpenAI deal, if it closes, would layer an even steeper ramp on top of that commitment.

A playbook Altman has run before

This is not the first time Altman has used a board resignation to create transactional distance between himself and a clean energy startup he's backed. In April 2025, he stepped down as board chair of Oklo, a small modular reactor startup he had backed, with Oklo co-founder Caroline Cochran confirming to CNBC at the time that the move was designed to "avoid conflict of interest" and "open up opportunities for future deals between OpenAI and Oklo."

The Helion move follows the identical structure: Altman exits the board, retains equity, and the company is now positioned to do business with OpenAI. Eleven months between the two instances. That's not coincidence — it's a playbook.

The question the governance arrangement doesn't fully answer is this: if OpenAI commits to purchasing 5 GW — or eventually 50 GW — from Helion, Helion's valuation rises materially. Altman's stake appreciates. He has been recused from the deal discussions, but he benefits from their outcome. Whether OpenAI's board has fully priced that into their deliberations isn't public.

What it means for the AI energy race

Altman has reportedly said he envisions 250 gigawatts of new electricity to power AI infrastructure. Fifty gigawatts from Helion alone would represent 20% of that stated vision — roughly equivalent to the entire generating capacity of France's nuclear fleet, delivered by a startup that has yet to prove commercial viability.

If Helion delivers at that scale, OpenAI would have locked in a structural energy advantage that no competitor could easily replicate. The most promising clean energy source in the world — controlled, in meaningful part, by the CEO of the company consuming it. That's the scenario Altman is betting on. The counterargument is equally straightforward: fusion has been "ten years away" for seventy years, and the 2030 deadline assumes a technology leap that hasn't happened yet.

The confirmed facts here are narrow but significant: Altman stepped off the Helion board (Kirtley on record), Helion has a 50 MW / 2028 PPA with Microsoft (Helion's own announcement), and Altman ran the same board-exit structure eleven months ago at Oklo. Everything else — the deal terms, the production targets, the 100- and 1,000-reactor projections, and the $375 million stake — is either unconfirmed, estimated, or derived from a single anonymous source. Read accordingly.

The deal remains unconfirmed by OpenAI, contingent on conditions not yet met, and founded on a single anonymous source. But the Kirtley quote and the board resignation make it real enough to watch. The next data point is whether OpenAI says anything — or whether the silence holds until a term sheet is signed.