OpenAI Locks In 12.5% of Helion’s Fusion Output as AI Compute Hits Energy Ceiling

OpenAI is securing 12.5% of fusion startup Helion Energy’s initial production capacity, with targets of 5 gigawatts by 2030 and 50 gigawatts by 2035—the clearest signal yet that AI scaling has collided with physical energy constraints.

The deal, reported by Axios, marks the first time an AI company has committed to this scale of unproven power infrastructure. OpenAI chairman Sam Altman recused himself from board discussions given his $375 million stake in Helion, invested during the company’s 2021 Series E round. The arrangement exposes both the severity of AI’s energy bottleneck and the willingness of cash-rich technology firms to bankroll long-shot energy technologies rather than compete for constrained grid capacity.

The Energy Math Behind AI Ambitions

OpenAI is planning a 125-fold increase in electricity consumption by 2030, per IDC Nova. That trajectory collides with grid realities: US Data Centers already consume 176 terawatt-hours annually—4.4% of national power—and the International Energy Agency projects global data center demand will double to 945 TWh by 2030. The Stargate project, OpenAI’s infrastructure partnership, targets 5 gigawatts per facility with plans for up to 10 sites by 2028, each requiring the output of multiple large power plants.

Traditional procurement routes cannot scale at this speed. Goldman Sachs estimates 85-90 gigawatts of new nuclear capacity would be needed by 2030 to meet data center demand—equivalent to adding 85 new reactors in four years, a physical impossibility given construction timelines. Solar-plus-storage faces similar constraints: utility-scale battery deployments lag far behind the pace required for AI compute expansion. This supply-demand mismatch is forcing technology companies to become energy infrastructure investors, effectively pre-purchasing future capacity from technologies still in development.

Helion’s Technical Position and Risk Profile

Helion achieved plasma temperatures of 150 million degrees Celsius on 13 February 2026, according to Fortune—a milestone that edges closer to the conditions required for sustained fusion reactions. The company raised $425 million in January 2025 at a $5.425 billion valuation, bringing total capital to over $1 billion. Investors include SoftBank, Lightspeed, and Altman himself, creating a network of aligned interests between energy production and energy consumption.

No private fusion company has reached scientific breakeven, the point where fusion energy output exceeds input. Helion’s 2028 delivery target for its existing 50-megawatt Microsoft contract—signed in May 2023 with financial penalties for non-delivery—represents the industry’s most aggressive commercial timeline. David Kirtley, Helion’s CEO, told CNBC that “this is a real PPA, so there’s financial penalties if Helion can’t deliver power.” The OpenAI deal scales that commitment by two orders of magnitude, transforming fusion from a research bet into a cornerstone infrastructure dependency.

Market and Geopolitical Implications

The deal reframes venture capital economics. Helion’s valuation reflects not only fusion technology risk but also the option value of baseload carbon-free power in a market where AI companies have demonstrated willingness to pay premiums for reliable supply. If successful, Helion’s power purchase agreements could generate returns rivaling software-scale margins while solving a constraint that threatens to slow AI development.

This creates ripple effects across Energy Markets. Utilities with existing nuclear or geothermal assets gain strategic value as AI companies seek immediate capacity while waiting for fusion. States with surplus hydroelectric or nuclear capacity—Washington, South Carolina, Illinois—become more attractive for data center development. The concentration of power demand also pressures transmission infrastructure: 50 gigawatts of Helion capacity would require grid connections equivalent to 50 large coal plants, demanding coordination with regional transmission operators years in advance.

Geopolitically, the dynamic mirrors semiconductor competition. China has invested heavily in state-backed fusion programs, while European consortia pursue alternative fusion architectures. If Helion or its competitors succeed, energy sovereignty becomes a technology race rather than a resource extraction problem. Countries with advanced fusion capabilities could dictate terms for AI Infrastructure placement, creating a new layer of technology-energy interdependence.

What to Watch

Monitor Helion’s 2028 delivery against the Microsoft contract—the first commercial test of private fusion timelines. Track whether other AI companies follow OpenAI into direct energy infrastructure investments or pursue different strategies such as modular nuclear reactors or utility partnerships. Watch for regulatory changes around power purchase agreements that span decades and technologies that do not yet exist at commercial scale.

Data center site selections in the next 18 months will reveal which companies believe fusion is viable and which are hedging with conventional sources. If Helion misses its 2028 target, expect a repricing of fusion venture valuations and a shift back toward nuclear restarts and natural gas. If it delivers, the race to secure fusion capacity will accelerate, potentially pulling forward investment timelines across the sector and reshaping how technology companies approach capital allocation between compute and energy infrastructure.