AI companies have been gobbling up GPU clusters, data center leases, and semiconductor supply chains. Now OpenAI wants something far more ambitious: the sun's power source, on contract, delivered to its data centers within a decade.
OpenAI is in advanced talks to buy electricity from Helion Energy — a fusion power startup — with negotiations advancing rapidly since initial reports surfaced last week. The deal would start at 5 gigawatts of power by 2030 and could scale to 50 gigawatts by 2035. That's 12.5% of Helion's total projected output, enough to power tens of millions of homes, channeled entirely into running AI.
There's one problem. Sam Altman — CEO of OpenAI, the buyer — is also a major personal investor in Helion Energy, the seller. He previously served as Helion's board chair.
AI's Hunger for Power Is Real
OpenAI's Stargate project already envisions data centers consuming up to 10 gigawatts of electricity in total. The company's longer-term internal projections reportedly target 250 gigawatts by 2033 — roughly half of Europe's entire peak electricity load.
Traditional energy can't scale that fast. Solar and wind are intermittent. New nuclear plants take decades to permit and build. Fossil fuels are increasingly politically untenable.
Fusion is theoretically the answer: an inexhaustible source of clean energy generated by the same nuclear process that powers stars. The problem is that fusion has been "30 years away" for the last 70 years.
Helion's pitch is that they're different. Their seventh-generation prototype, Polaris, reached 150 million degrees Celsius in February 2026 using real deuterium-tritium fuel — ten times hotter than the core of the sun. That's real progress. But Helion has not yet demonstrated net energy gain — the fundamental threshold where fusion produces more power than it consumes.
The Numbers Are Hard to Ignore
5 gigawatts is the electrical consumption of approximately 3.7 million U.S. homes. OpenAI wants ten times that amount — 50 gigawatts — within a decade. As one energy analyst summarized: "AI's appetite for electricity is growing faster than any energy source can supply it. Fusion is a moonshot — but OpenAI is running out of conventional options."
To compare: the entire U.S. nuclear power fleet currently generates about 100 gigawatts. OpenAI alone wants half that, from a single company, using a technology that hasn't commercially worked yet.
Microsoft already has a separate power-purchase agreement with Helion for 50 megawatts by 2028. Google has inked deals with Commonwealth Fusion Systems, Helion's main competitor. The race to lock in clean energy is running parallel to the race for AI supremacy.
The Conflict Question
Altman has reportedly stepped down from Helion's board and recused himself from deal negotiations. But "I removed myself from the process" is not quite the same as having no skin in the game. As OpenAI moves toward a potential IPO — a process already generating internal friction over timelines — any deal where the CEO profits on both sides will receive serious scrutiny from investors, regulators, and governance watchdogs.
Altman has previously come under fire for similar dynamics — including his connection to semiconductor ventures benefiting from OpenAI's chip procurement decisions.
What It Means for AI Users
If fusion power works: stable electricity costs could keep AI subscription prices from rising, and reduce the environmental footprint that critics increasingly cite as AI's dirty secret.
If it doesn't work by 2030: OpenAI faces a power crunch at exactly the moment it needs maximum compute. That likely translates to higher prices, tighter rate limits, and deferred capability launches for users.
The deal signals something broader: AI companies no longer see themselves as pure software businesses. They're becoming energy companies, negotiating power contracts at nation-state scale. That transformation is accelerating faster than most observers expected — and the rules governing it are still being written.
