OpenAI Eyes Massive Fusion Power Deal in Strategic Shift Toward Energy Security
In a move that underscores the escalating energy demands of artificial intelligence, OpenAI is reportedly in advanced negotiations to purchase electricity from Helion Energy, a fusion startup backed by its own CEO, Sam Altman. According to Axios, the potential framework would grant OpenAI an initial 12.5% stake in Helion’s future power output, translating to a staggering 5 gigawatts (GW) by 2030, with possibilities scaling to 50 GW by 2035.
For context, a 5 GW allocation represents a monumental commitment—comparable to the output of several large nuclear reactors—and would position OpenAI as one of the world’s largest single power consumers. The 50 GW figure, if realized, would align the company’s projected energy footprint with national-scale infrastructure planning, highlighting a pivotal shift in strategy from focusing solely on computational hardware (chips) and data center logistics to securing the foundational energy supply itself.
Helion Energy: A High-Stakes Bet on Fusion
Helion has emerged as a frontrunner in the private fusion sector, attracting significant capital and attention. The company secured a $425 million Series F funding round in January 2025, valuing it at $5.425 billion post-money and pushing total investments beyond $1 billion. Its investor roster includes SoftBank Vision Fund 2, Mithril Capital, Dustin Moskovitz’s Good Ventures Foundation, and Sam Altman, who previously led a $500 million Series E round in 2021.
Critically, Altman has recused himself from OpenAI’s deal discussions with Helion and stepped down from the company’s board. This separation is essential, as Helion is not merely a vendor but a central pillar in Altman’s long-held thesis that achieving “abundant, cheap energy” via fusion is a non-negotiable prerequisite for scaling advanced AI.
The Fusion Caveat: A Milestone Not Yet Reached
The potential OpenAI-Helion agreement, however, remains highly conditional. The fundamental challenge is that commercial fusion power has not yet been achieved. Helion believes it is approaching “scientific breakeven”—the point where a fusion reaction generates more energy than is used to initiate it—but no private company has publicly demonstrated this milestone. Furthermore, major logistical questions, including the specific locations for Helion’s power plants, remain unresolved.
Helion’s progress, while promising, is still in the engineering and construction phase. The company signed what it called the world’s first fusion power purchase agreement (PPA) with Microsoft in 2023, targeting delivery of at least 50 megawatts (MW) from its first plant by 2028. In July 2025, Helion announced it had secured land and begun construction on that inaugural facility.
An Industry-Wide Race for Clean, Stable Power
OpenAI’s reported interest is part of a broader trend among tech giants securing long-term, clean energy sources to power data centers hungry for compute. Google has pursued a parallel path, signing a deal in June 2025 to purchase 200 MW from Commonwealth Fusion Systems (CFS), a Helion rival spun out of MIT. The Google-CFS agreement, for power from CFS’s proposed ARC plant in Virginia, was hailed by both parties as a landmark transaction for the fusion industry.
These binding commitments from Microsoft, Google, and potentially OpenAI, serve as both a financial catalyst and a credibility signal for fusion companies. They transform the technology from a scientific endeavor into a contracted future utility, accelerating development cycles and attracting further investment.
Looking Ahead: Ambition Meets Reality
The reported talks between OpenAI and Helion are a stark indicator of the AI industry’s trajectory: the next frontier is not just algorithmic, but infrastructural and energetic. While the proposed volumes are audacious, they are tethered to a technology still striving for its first commercial demonstration. The success of this potential deal, and others like it, will ultimately depend on Helion and its peers crossing the scientific and engineering finish lines on their aggressive timelines.
Disclosure: This article was edited by Estefano Gomez. For more information on how we create and review content, see our Editorial Policy.
