OpenAI in Talks With Helion to Secure Fusion Energy

OpenAI is negotiating a power supply agreement with Helion Energy that would secure up to 50 gigawatts of fusion-generated electricity by 2035, a move that signals artificial intelligence's transformation from software platform to energy-intensive industrial system requiring dedicated baseload capacity at unprecedented scale [1].

The talks, confirmed by sources to POWER magazine in March 2026, would grant OpenAI access to 5 GW from Helion's facilities by 2030, escalating to 50 GW within a decade [1]. To contextualize: 50 GW represents roughly 5 percent of total U.S. installed generating capacity and would position Helion among the nation's largest power producers — if the company can execute.

"What makes this deal significant is that it reframes AI as an energy-intensive industrial system rather than just a software platform," Siddardha Vangala, senior AI Platform Engineer and Enterprise AI Systems Architect with MasTec Advanced Technologies, told POWER [1]. The statement crystallizes what capital markets have been slow to price: AI's scaling constraints are no longer algorithmic but infrastructural.

The Fusion Premium: Why OpenAI Is Paying 17.5% Guaranteed Returns

OpenAI is simultaneously pursuing an aggressive capital raise through private equity channels, offering preferred equity stakes with guaranteed minimum returns of 17.5 percent — well above market norms for technology growth equity [1]. The dual strategy reveals acute pressure: the company needs both capital and power, and it's willing to pay premium rates for both.

The guaranteed return structure is notable. Typical venture growth rounds carry no return floor; the 17.5 percent minimum suggests OpenAI is targeting institutional investors with fixed-income return expectations — pension funds, insurance capital, infrastructure vehicles — rather than pure venture capital. OpenAI is also dangling early access to new AI models as additional inducement [1].

The linkage between these capital raises and the Helion discussions is direct: training and running frontier AI models requires exponential compute, and compute requires power. Data center demand is rising so rapidly that companies are now securing dedicated generation years in advance [1]. OpenAI's CEO Sam Altman, who led Helion's $500 million Series E funding round in 2021, has positioned himself at the nexus of both challenges [1]. Helion closed an additional $425 million funding round in January 2025 [1].

Altman stepped down from Helion's board of directors in March 2026 but retains what has been described as a "sizable" — though undisclosed — financial stake in the company [1]. He has recused himself from direct negotiations between OpenAI and Helion [1]. The governance maneuver addresses conflict-of-interest concerns while preserving Altman's exposure to both sides of the transaction.

Other Helion investors include Softbank, Mithril Capital (led by PayPal founder Peter Thiel), and Meta, including Facebook co-founder Dustin Moskovitz [1]. The roster signals Silicon Valley's view that fusion represents critical path infrastructure for the next generation of compute-intensive applications.

Helion's Technical Milestone: 150 Million Degrees and the Path to Commercial Viability

In February 2026, Helion announced that its Polaris prototype achieved plasma temperatures of 150 million degrees Celsius and became the first privately developed fusion energy machine to demonstrate measurable deuterium-tritium fusion [1]. The company described the milestones as "significant breakthroughs" toward commercially viable fusion energy [1].

Polaris is Helion's seventh-generation nuclear fusion prototype, employing a Field-Reversed Configuration plasma generator focused on a smaller, pulsed, non-thermal approach to achieve commercial power generation [1]. The architecture differs from tokamak designs pursued by Commonwealth Fusion Systems and other competitors.

Google signed agreements with Commonwealth Fusion Systems in 2025, establishing a parallel procurement path for tech giants seeking carbon-free baseload power [1]. Microsoft in 2023 executed what was considered the first power purchase agreement tied to fusion energy, contracting with Helion to buy electricity beginning in 2028 [1]. The Microsoft deal established precedent for long-dated commitments to unproven technology — OpenAI's discussions represent an order-of-magnitude scale-up of that model.

Jason Mann, CEO and co-founder of growth marketing firm Stock, framed the infrastructure challenge bluntly: "The execution for Helion to provide 5 GW by 2030, 50 GW by 2035 to OpenAI is an enormous undertaking, and risk, for the company. It's a risk because fusion technology at these scales is new and unproven" [1].

The Competitive Landscape: Energy as Moat

Edward Tian, CEO of AI firm GPTZero, identified energy access as the binding constraint on AI advancement: "The biggest bottleneck for AI today is not the capabilities of the model itself, but instead computing and the energy to run those models" [1]. The insight explains why OpenAI, Anthropic, and other frontier AI labs are now competing not just on algorithms but on power supply agreements.

Anthropic is reportedly pursuing similar joint venture structures with private equity to raise capital for infrastructure [1]. The convergence suggests a broader pattern: AI companies are becoming quasi-utilities, vertically integrating generation capacity to secure competitive advantage.

Mann noted downstream implications: "What this means for consumers and the end user, is basically cheaper AI. But where consumers will benefit most is if Helion can actually provide fusion energy at a larger scale because then theoretically it should be cheaper energy not just for AI but for everyone else as well" [1]. The statement assumes successful commercialization at scale — a significant conditional.

Parallel developments in energy infrastructure underscore the sector's momentum. In March 2026, KBR was awarded the Front-End Engineering Design contract for the EcoLog Terminal in Amsterdam, the world's first commercial-scale facility designed to import liquid hydrogen and export liquid CO2 [3]. Initial annual throughput is planned at 200,000 tons of liquid hydrogen and 1.8 million tons of liquid CO2, with potential expansion to 600,000 tons and 4.25 million tons respectively, targeted for end of 2030 [3]. While distinct from fusion, the project reflects capital markets' willingness to fund first-of-a-kind energy infrastructure when anchored by credible offtake commitments.

Similarly, Datacor's March 2026 acquisition of GoldSim Technology Group, a developer of dynamic simulation software for complex systems under uncertainty, deepens engineering simulation portfolios across energy and environmental sectors [2]. Tom Jackson, president of Datacor, stated: "This acquisition reflects our continued investment in industry-specific solutions and our dedication to helping engineers apply the power of simulation across a broader range of their work" [2]. The deal signals rising demand for risk assessment tools as energy infrastructure projects increase in complexity and scale.

The Plocamium View

The OpenAI-Helion discussions represent the coalescence of three structural trends: AI's industrialization, the financialization of energy infrastructure, and the emerging market for dedicated compute power.

Our thesis: this deal, if consummated, will establish fusion energy not as a future technology but as a present-day asset class. The 17.5 percent guaranteed return OpenAI is offering to PE investors creates a benchmark for how capital markets should price infrastructure exposure to AI compute demand. That return profile — well above investment-grade debt but below typical venture equity — reflects infrastructure-like risk characteristics with technology growth upside.

The scale is what matters. At 50 GW, Helion would need to deploy fusion capacity equivalent to 50 large nuclear reactors or 167 utility-scale gas combined-cycle plants. The capital requirement for such buildout would conservatively exceed $100 billion, assuming $2 per watt installed cost — a figure that implies significant cost reduction from current demonstration-scale economics. This is infrastructure finance at the scale of interstate highway systems or telecommunications backbone networks.

OpenAI's willingness to lock in power supply a decade forward reveals how energy access has become the primary competitive moat in AI. Algorithmic advantages erode through publication and talent mobility; proprietary training datasets face regulatory and ethical constraints; but dedicated, carbon-free, baseload power represents durable competitive advantage. The company that secures generation capacity today locks in cost structure and regulatory positioning for the next cycle of model development.

The governance structure Altman has adopted — off Helion's board but retaining financial interest — establishes a template for how founder-investors navigate conflicts when portfolio companies become counterparties. Expect similar recusal frameworks as venture capital increasingly funds both infrastructure and the applications that depend on that infrastructure.

Risk factors are material. Helion has demonstrated fusion in a prototype; scaling to multi-gigawatt commercial deployment by 2030 requires solving engineering, regulatory, supply chain, and capital formation challenges simultaneously. The Microsoft PPA, for 2028 delivery, represents the near-term execution test. If Helion misses that milestone, confidence in the 2030 and 2035 commitments will collapse, and OpenAI will face a scramble for alternative supply — likely natural gas with carbon offsets, undermining the environmental narrative both companies have cultivated.

The Bottom Line

OpenAI's pursuit of 50 GW of fusion power by 2035 is the most ambitious energy procurement commitment by a technology company in history. It transforms AI from a software business into an industrial power consumer requiring utility-scale baseload generation. The deal structure — guaranteed equity returns, board governance adjustments, decade-forward supply commitments — establishes templates for how infrastructure and technology capital will intersect as compute demand scales.

For institutional investors, the opportunity set is clear: infrastructure vehicles that can underwrite multi-decade power supply to AI offtakers now carry premium return profiles with technology exposure. The risk is execution: Helion must deliver commercial fusion at scale within four years to validate the model. If successful, every major AI lab will pursue similar dedicated generation agreements, creating a multi-hundred-billion-dollar market for carbon-free baseload power. If Helion stumbles, the entire premise of fusion-as-infrastructure collapses, and AI companies revert to competing for scarce grid capacity at whatever price utilities demand.

The next 48 months will determine whether fusion becomes the foundation of the AI industrial complex or a cautionary tale of premature infrastructure bets. Capital is voting with conviction: the scale of commitments suggests the former.

---

References [1] Proctor, D. (2026, March 23). OpenAI in Talks With Helion to Secure Fusion Energy. POWER Magazine. https://www.powermag.com/openai-in-talks-with-helion-to-secure-fusion-energy/ [2] Bailey, M. (2026, March 23). Datacor acquires dynamic simulation provider GoldSim Technology Group. Chemical Engineering. https://www.chemengonline.com/datacor-acquires-dynamic-simulation-provider-goldsim-technology-group/ [3] Bailey, M. (2026, March 23). KBR awarded FEED contract for Port of Amsterdam liquid-hydrogen terminal. Chemical Engineering. https://www.chemengonline.com/kbr-awarded-feed-contract-for-port-of-amsterdam-liquid-hydrogen-terminal/