OpenAI is no longer just hunting for more GPUs – it’s quietly trying to lock down its own electricity supply. Talks with fusion startup Helion hint at a future where leading AI labs don’t just consume power, they reshape the global energy market.
This is not just another power purchase agreement. It’s a signal that the same people steering the AI revolution want a front‑row seat in the next energy revolution, too. In this piece, we’ll unpack what’s actually on the table, why the numbers look almost implausibly ambitious, how this fits into Sam Altman’s growing industrial ecosystem, and what it all means for Europe’s energy‑hungry AI ambitions.
The news in brief
According to TechCrunch, fusion startup Helion is in early talks to sell electricity to OpenAI. Axios first reported that a draft deal could give OpenAI rights to 12.5% of Helion’s future output: the equivalent of 5 gigawatts by 2030 and 50 gigawatts by 2035.
Helion claims each of its fusion reactors would generate around 50 megawatts. On the back of the Axios numbers, TechCrunch notes this would imply roughly 800 reactors installed by 2030 and another 7,200 by 2035 – assuming everything works.
Helion is already bound by a 2023 agreement to supply power to Microsoft, OpenAI’s key cloud partner, starting in 2028. The company is currently operating its Polaris prototype and recently reported plasma temperatures of about 150 million °C, aiming for around 200 million °C for commercial operation.
Both OpenAI and Helion have been heavily backed by Sam Altman. TechCrunch reports that Altman has stepped down as Helion’s board chair and recused himself from these talks to reduce conflicts of interest.
Why this matters
If you want to understand how far AI leaders are planning ahead, look at their energy strategy. A deal that reserves 12.5% of a fusion company’s entire future output is not about next year’s data‑center bill; it’s about locking in a structural advantage for the next decade.
Winners first. OpenAI gets a narrative and, potentially, a real asset: long‑term access to firm, low‑carbon power at a time when AI’s electricity appetite is exploding faster than grid planning processes can cope. If Helion even partially delivers, OpenAI can train and run models with less exposure to volatile energy prices and public backlash over carbon emissions. That’s a competitive moat no open‑source community or smaller lab can easily replicate.
Helion, in turn, gets something every deep‑tech hardware startup dreams of: an anchor customer willing to underwrite enormous capex with long‑term offtake. AI labs and hyperscalers are uniquely positioned to sign multi‑decade, multi‑gigawatt deals – the kind of demand signal that can unlock project finance at scale, much as long‑term contracts did for wind and solar in the 2010s.
The potential losers are more diffuse. If frontier AI labs secure priority access to future clean baseload, the rest of the economy could find itself bidding for what’s left – especially in regions already struggling with grid congestion. There’s also the governance question: should climate‑critical infrastructure be effectively pre‑booked by a small circle of private AI companies tied together by the same investor?
And then there’s execution risk. The numbers implied by the Axios / TechCrunch reporting are, frankly, staggering. Thousands of commercial fusion reactors within a decade would represent a deployment tempo the nuclear sector has never come close to. If Helion slips – and fusion timelines almost always do – OpenAI’s plan B will look a lot like everyone else’s: scramble for grid power and PR‑friendly renewables.
The bigger picture
Zoom out, and this story sits at the intersection of three mega‑trends: AI’s insatiable demand for compute, the electrification of everything, and the privatization of what used to be state‑driven infrastructure bets.
First, AI as a physical industry. For years, we talked about AI as “software eating the world”. Now we’re watching AI eat transformers, substations and gigawatts. Hyperscalers are already signing record‑breaking power purchase agreements for wind, solar and batteries. A move into fusion is simply the most extreme version of a broader pattern: cloud and AI companies becoming quasi‑utilities.
Second, frontier energy bets are converging around the same small set of actors. Altman has backed not only Helion but also Oklo, a company working on small fission reactors. According to TechCrunch’s recap of past reporting, he stepped down from leadership roles at both to ease potential conflicts as they explore deals with AI firms – but the strategic direction is clear. The AI–nuclear–fusion triangle is no accident; it’s an attempt to vertically integrate intelligence, energy and compute.
Third, this echoes an earlier era. In the 20th century, oil majors used to own everything from wells to refineries to shipping. Today’s tech giants are assembling something analogous across data centers, undersea cables, chip design and, increasingly, power. The Helion talks signal that at least some AI labs believe relying purely on public grids and short‑term market contracts will not be enough to sustain their growth.
Compared to competitors, this move raises the bar. Google, Meta, Amazon and others have aggressive renewable strategies, but none (publicly) are staking claims on specific fusion output at this scale. If OpenAI can say to regulators and enterprise customers, “our models run on dedicated, zero‑carbon fusion,” that’s not just ESG marketing – it’s a procurement weapon.
The flip side: if Helion misses its milestones, the whole fusion‑powered‑AI narrative risks becoming the next Hyperloop – a distraction that lets companies over‑promise future green power while still burning today’s fossil‑heavy mix.
The European / regional angle
For Europe, this story lands in the middle of several uncomfortable debates: how to power AI growth, how to handle nuclear‑adjacent technologies, and how much dependence on US‑centric tech ecosystems is acceptable.
European data centers are already colliding with local grids. Ireland has effectively paused new connections in parts of Dublin; the Netherlands has pushed back on hyperscaler sites; Germany faces tight supply in industrial regions. Add EU AI Act compliance – which will likely push more companies toward large, regulated providers – and you get further concentration of compute, and therefore power demand, in a handful of hubs.
If Helion ever delivers commercially, its first priority will almost certainly be North American sites tied to US hyperscalers. European users of OpenAI or Microsoft Azure would benefit indirectly from lower lifecycle emissions, but the physical resilience – the ability to keep servers running during local shortages – would not automatically transfer.
There’s also a strategic autonomy angle. The EU has poured billions into ITER, the experimental fusion reactor in France, and supports a growing ecosystem of European fusion startups. Yet it risks watching US private players, bankrolled by Silicon Valley AI money, reach commercial deployment first and then sell the technology back to Europe under their terms.
Regulators in Brussels will also look at this through the lens of the Digital Markets Act and upcoming grid‑access reforms. If a handful of gatekeeper platforms start cornering long‑term clean‑power contracts, does that entrench their market power in AI and cloud even further? Expect these questions to surface as part of broader discussions about data‑center siting, capacity auctions and the role of nuclear and fusion in the EU’s green taxonomy.
Looking ahead
The most important thing to understand is that very little about this deal is real yet. The talks are early, the technology is pre‑commercial, and the 2030 and 2035 numbers are more like stretch targets than bankable forecasts.
Still, they tell us what to watch.
On the technical side, the key milestone is not a press release about new temperatures but a sustained net‑electricity‑producing shot, independently validated, followed by a clear roadmap to a first commercial plant. Until that happens, Helion’s gigawatt projections remain speculative.
On the commercial side, look for three signals:
- Contract structure. Does OpenAI commit to take‑or‑pay terms, prepayments, or even equity‑linked arrangements that effectively finance Helion’s build‑out?
- Regulatory treatment. How US and EU regulators classify and oversee fusion – separate from fission – will define permitting timelines and public acceptance.
- Imitation. If this talks turn into a real agreement, expect a fast follower effect: other frontier AI labs and cloud providers rushing to sign their own nuclear or fusion deals.
For readers, the practical timeline is long. Nothing about your OpenAI usage in 2026 will be powered by fusion. But the power‑supply decisions taken now will shape where future data centers are built, which regions attract AI investment, and who controls the levers of both digital and energy infrastructure in the 2030s.
The risk is that hype gets ahead of engineering, and policymakers assume that “AI will be green because fusion” without fixing today’s grid, permitting and investment gaps. The opportunity is that AI’s hunger for power finally forces governments and investors to treat clean firm generation as urgent infrastructure, not a science project.
The bottom line
The OpenAI–Helion talks are less about a specific contract and more about a worldview: that whoever controls compute and clean power will control the next industrial cycle. The numbers on the table look wildly ambitious, and history suggests fusion will slip. But the direction of travel is clear – AI labs are becoming energy players.
The open question for Europe and the rest of the world is simple: do we want the future’s most critical clean‑power assets effectively bundled with a small club of AI giants, or do we build alternative models before it’s too late?
