The race to dominate artificial intelligence is beginning to collide with a more fundamental constraint: power. As data centers expand and computing loads surge, the question is shifting from how fast AI can scale to whether the energy system can keep up.
That tension is now drawing some of the industry’s most ambitious bets into view.
At the center is Helion, which is in early discussions to supply electricity to OpenAI in what could become one of the most consequential energy deals tied to the AI boom. The outline, reported by Axios, suggests OpenAI could secure 12.5% of Helion’s future output—about 5 gigawatts by 2030 and 50 gigawatts by 2035.
Register for Tekedia Mini-MBA edition 20 (June 8 – Sept 5, 2026).
Register for Tekedia AI in Business Masterclass.
Join Tekedia Capital Syndicate and co-invest in great global startups.
Register for Tekedia AI Lab.
Those figures are striking not just for their size, but for what they imply. Helion has said each of its reactors would generate roughly 50 megawatts. Meeting the proposed demand would require the company to build about 800 reactors within five years and more than 7,000 additional units over the following half decade. For an industry that has yet to deliver a single commercial fusion plant, the scale borders on an industrial moonshot.
This implies that AI companies are no longer content to rely on utilities and existing grids. They are moving to secure dedicated power sources, often years in advance, and increasingly from unconventional suppliers. Microsoft, OpenAI’s key partner, signed a separate agreement with Helion in 2023 for electricity deliveries beginning in 2028, effectively underwriting the company’s first commercial ambitions.
The common thread in these moves is urgency. Training large AI models and running inference at scale require vast amounts of electricity, and that demand is rising faster than grid expansion in many regions. Securing power is becoming a strategic priority, on par with access to chips and data.
Helion’s approach sets it apart from much of the fusion field. Most developers are pursuing designs that convert fusion-generated heat into steam to drive turbines. Helion is attempting direct conversion, using magnetic fields to compress plasma until fusion occurs, with the resulting energy fed back into the system to generate electricity. The design promises higher efficiency and fewer moving parts, but it also carries unproven engineering risks.
Progress has been incremental. The company’s Polaris prototype has reached plasma temperatures of 150 million degrees Celsius, nearing the 200 million threshold it believes is necessary for sustained fusion reactions. That remains a technical milestone rather than a commercial one. Bridging the gap between experimental conditions and reliable, grid-scale output has eluded the industry for decades.
Even if the physics holds, the manufacturing challenge looms just as large. Producing thousands of reactors would require a supply chain that does not yet exist, along with regulatory approvals and capital commitments on a scale rarely seen outside conventional energy megaprojects. Helion raised $425 million last year from investors including Sam Altman, Mithril, Lightspeed, and SoftBank, but that sum would cover only a fraction of the cost implied by mass deployment.
Altman’s role ties the narrative together because he has backed Helion while leading OpenAI’s expansion, and though he has stepped away from formal decision-making in the discussions, the overlap highlights how closely the future of AI is becoming linked to the future of energy. His earlier move to step down as chair of Oklo, another advanced energy company exploring partnerships with AI firms, points in the same direction.
There is also a competitive dimension. If Helion succeeds, it could leapfrog other fusion developers targeting the early 2030s for commercial deployment. Securing anchor customers such as OpenAI and Microsoft would give it both financial backing and a guaranteed market—two factors that have historically been missing in fusion’s long development cycle.
However, the timelines assume rapid technical validation, regulatory clearance, and industrial scaling, all within a decade. Any delay in one area could cascade across the entire plan. For OpenAI, the arrangement would secure long-term energy supply, but it would also tie part of its infrastructure to a technology that has yet to prove itself outside the lab.
What is clear so far is that the relationship between computing and energy is tightening. The expansion of AI is beginning to reshape power markets, drawing capital toward new generation technologies and accelerating timelines that once seemed distant.
Fusion, long treated as a distant prospect, is being pulled forward by that demand.