OpenAI Is Betting on Fusion That Has Not Been Invented Yet

The numbers in the Helion deal are so large they force a question the deal's sponsors have not publicly answered: what's the plan if Helion does not hit net energy gain?

Axios reported Monday that OpenAI is in advanced talks to purchase electricity directly from Helion Energy, the fusion startup that Sam Altman has backed personally to the tune of $375 million. The framework: OpenAI would get 12.5% of Helion's total production, targeting 5 gigawatts by 2030 and scaling to 50 GW by 2035. Altman has stepped down as Helion's board chairman ahead of the deal, a standard governance move meant to blunt conflict-of-interest concerns.

The governance move is fine. The numbers are extraordinary.

To understand why, start with the only comparable commercial fusion deal on record. In May 2023, Microsoft signed a power purchase agreement with Helion for 50 megawatts by 2028. Constellation Energy would serve as power marketer. Helion broke ground on the commercial plant — called Orion — in Malaga, Washington in July 2025. Chelan County issued the conditional use permit in October 2025. The plant is under construction. It has not produced a single electron of grid electricity.

The OpenAI deal would take 100 times that, two years later, from the same company and the same underlying technology.

For that to work, Helion would not be building one commercial plant. They would be manufacturing fusion generators at an industrial pace that no one in the fusion industry — public or private — has ever attempted. The Axios story confirms that many conditions remain unfulfilled, including basic site selection for the OpenAI-adjacent capacity. This is not a contract for electricity. It is a conditional framework for electricity that assumes commercial fusion works at a scale nobody has demonstrated.

What Helion Has Actually Demonstrated

To be fair to Helion: they have real results.

On February 13, 2026, Helion announced that Polaris — their 7th-generation prototype — achieved 150 million degrees Celsius plasma temperature, beating their own record from the 6th-generation Trenta machine. Polaris was also the first privately-funded fusion device to use deuterium-tritium fuel, which is the fuel combination expected in commercial fusion plants. The DOE's Office of Science validated the results.

David Kirtley, Helion's CEO and co-founder, has been building fusion machines since 2013. Seven prototypes over thirteen years. "We've built and operated seven prototypes, setting and exceeding more ambitious technical and engineering goals each time," he said in February. The company's scientific progress is genuine.

But there's a gap between temperature records and commercial power, and it is not a small one. As of December 2025, Polaris had not demonstrated net energy production — meaning energy out exceeds energy in. Helion says they are on the cusp of scientific breakeven. That is a milestone that no private fusion company has yet achieved.

In 2023, when the Microsoft deal was announced, MIT Technology Review spoke with independent experts who were already skeptical of the 50 MW timeline. Jessica Lovering of Good Energy Collective identified two major unproven steps: net energy gain and converting fusion output to grid electricity. Adam Stein of the Breakthrough Institute put it plainly: these are breakthroughs we are talking about. Both were discussing the 50 MW plant. The OpenAI deal is 100 times larger.

The Conflict Is Not Just Procedural

Altman stepping down from the board is the right call. But it does not resolve the structural issue.

His $375 million personal investment led Helion's Series E in 2021. He has publicly called Helion more than an investment. He reportedly spends significant personal time on it alongside OpenAI. The governance attorneys who raised self-dealing concerns after the 2023 board drama were flagging this exact configuration: Altman in a position to direct company resources toward investments he personally benefits from.

Now the man who runs OpenAI has negotiated a major power supply agreement with a company he bankrolled, after stepping off its board. OpenAI's board would have needed to conduct its own independent analysis of whether this deal is arms-length — whether the price, the terms, the conditions, and the counterparty selection serve OpenAI's interests rather than its CEO's. That analysis should be disclosed. It has not been.

This does not mean the deal is bad for OpenAI. Fusion power at scale would be transformative for any data-center-heavy company, and Altman's interest in Helion may simply reflect a bet that paid off on the due-diligence front. But the question of who evaluated this deal independently is a legitimate governance inquiry that has not been answered.

The Energy Math

Altman has spoken publicly about OpenAI's power needs. The number he's used is approximately 250 GW. If the Helion deal delivers 5 GW by 2030, that is 2% of his stated requirement. He is not betting on fusion alone.

The money side is also now on-record. CNBC reported OpenAI expected roughly a $5 billion loss on $3.7 billion in 2024 revenue (Sept. 27, 2024), and Reuters later reported OpenAI had raised its projected cumulative cash burn through 2029 to $115 billion, citing The Information (Sept. 5, 2025). That does not prove Helion works. It does explain why OpenAI is trying to lock in power anywhere power might plausibly exist.

OpenAI's Stargate infrastructure buildout involves nuclear power agreements, conventional gas, and solar. Helion is one line in a much longer energy ledger. That context matters: this is a real bet on a speculative technology inside a diversified energy strategy, not a single-source gamble on fusion.

But 50 GW by 2035 — if ever realized — would be a different story entirely. That is not a hedge. That is a foundational energy position. And it requires Helion to achieve commercial viability, build out manufacturing capacity, and deliver electricity at a scale the entire global fusion industry has not collectively approached.

What to Watch

The deal has conditions. The most important ones are technical: Helion needs to cross scientific breakeven and demonstrate a path from prototype to manufacturing. The Orion plant in Malaga is supposed to deliver for Microsoft by 2028 — that is the proof-of-concept event this entire stack depends on. If Orion slips, the OpenAI deal's 2030 targets shift with it.

Site selection for OpenAI-associated capacity is also unresolved. For a fusion plant, that involves permitting, transmission interconnection, water access, and regulatory approvals that can take years on their own.

The real test of this deal's credibility is not the headline number. It is whether the Orion plant for Microsoft produces electricity on schedule. Watch that.

Primary sources: Axios (reporting, paywalled); Helion Energy (Polaris milestone announcement, Feb 13, 2026); MIT Technology Review (expert commentary, 2023); Reuters on OpenAI projected burn (https://www.reuters.com/technology/openai-expects-business-burn-115-billion-through-2029-information-reports-2025-09-06/); CNBC on OpenAI 2024 loss projection (https://www.cnbc.com/2024/09/27/openai-sees-5-billion-loss-this-year-on-3point7-billion-in-revenue.html); Helion DOE validation statement; Chelan County permit records.