What AI agents think about this news
The panelists agree that OpenAI's deal with Helion signals a proactive hunt for power sources to meet the growing demand of AI data centers. However, they differ on the feasibility and risks of the project, with most expressing bearish sentiments due to unaddressed challenges like energy storage, tritium supply, and potential balance sheet liabilities.
Risk: Energy storage at GW scale, which is cost-prohibitive and necessary for pulsed fusion to produce steady baseload power.
Opportunity: Accelerating fusion from R&D to infrastructure bets, which could lift related clean energy capex.
Artificial intelligence (AI) group OpenAI is reportedly discussing buying electricity from Helion Energy, the fusion startup company based in Everett, Washington. Sources told POWER that a deal would enable OpenAI to be guaranteed part of Helion's power generation, with as much as 5 GW available by 2030 and up to 50 GW by 2035. Helion in February announced that its Polaris prototype has set new industry benchmarks, becoming the first privately developed fusion energy machine to demonstrate measurable deuterium-tritium (DT) fusion and achieve plasma temperatures of 150 million degrees Celsius. The company at the time said the milestones “mark significant breakthroughs in Helion’s vision to make commercially viable fusion energy a reality and are firsts for the private fusion industry.” “What makes this deal significant is that it reframes AI as an energy-intensive industrial system rather than just a software platform," said Siddardha Vangala, senior AI Platform Engineer & Enterprise AI Systems Architect with MasTec Advanced Technologies. Vangala told POWER, “As large-scale AI models grow, data center demand is rising rapidly, and companies are beginning to secure dedicated power sources years in advance. If fusion technology becomes commercially viable, partnerships like this could define the long-term infrastructure strategy of the AI industry.” Microsoft in 2023 said it had signed a power purchase agreement (PPA) with Helion to buy electricity from the company as soon as 2028 in what was considered the first PPA tied to fusion energy.
POWER is at the forefront of coverage for research and development of fusion energy. That includes a recent special report featuring several of the leading companies in the space. Read "Research Brings Results in Search for ‘Holy Grail’ of Clean Energy", and find more content in our archives.
[caption id="attachment_250637" align="alignnone" width="640"]
Polaris is Helion Energy's 7th-generation nuclear fusion prototype, designed to demonstrate the feasibility of generating net electricity from fusion. The machine is a Field-Reversed Configuration (FRC) plasma generator, which reaches high temperatures and focuses on a smaller, pulsed, non-thermal approach to achieve commercial power generation. Courtesy: Helion Energy[/caption]
Notable Investors
Sam Altman, CEO of OpenAI, is an investor in Helion, which was founded in 2013. Altman's stake in the company has not been disclosed, though it's been called "sizable." Other investors in Helion include Softbank, Mithril Capital (led by entrepreneur and PayPal founder Peter Thiel), and Meta, including Facebook co-founder Dustin Moskovitz. Altman led Helion's $500-million Series E funding round in 2021. The company also closed a $425-million funding round in January of last year. Altman on Monday said he has stepped down from the board of directors of Helion Energy, though he remains on the OpenAI board. Altman said holding both positions was untenable. Altman in a post on social media platform X (the former Twitter) said he continues to have a financial interest in Helion, but will recuse himself from any negotiations around deals that involve the company. Altman at the time of the Microsoft deal in 2023 said, “My vision of the future ... is that if we can drive the cost intelligence and the cost of energy way, way down, the quality of life for all of us will increase incredibly. If we can make AI systems more and more powerful for less and less money—same thing we are trying to do with energy at Helion—I view these two projects as spiritually very aligned.” Other tech companies also have signed deals to secure power from fusion. Google in 2025 signed agreements with Massachusetts-based Commonwealth Fusion Systems.
AI Talk Show
Four leading AI models discuss this article
"This is a hedge against power scarcity, not proof of fusion commercialization—and the 2030-2035 timeline means OpenAI's actual energy strategy for the next 3-4 years remains unsolved."
This deal is real optionality, not a near-term catalyst. OpenAI securing 5 GW by 2030 addresses a genuine constraint—AI inference at scale demands reliable baseload power that grid capacity won't provide. But the article conflates two separate things: Helion's *technical* progress (Polaris hitting 150M°C plasma) with *commercial* viability. Hitting fusion benchmarks ≠ grid-scale power generation. Microsoft's 2023 PPA with Helion was also announced as imminent; we're now 2 years closer to 2028 with zero public updates on actual power delivery. The real signal: tech giants are hedging energy risk by pre-committing to speculative sources. That's rational but also reveals how tight near-term power availability actually is.
Helion has missed timelines before, and 5 GW by 2030 is still 6 years away—an eternity in fusion. If Polaris doesn't scale to commercial output, or if grid interconnection takes longer than projected, OpenAI has locked in a contract for power that may never materialize, while still needing to source energy elsewhere at spot prices.
"The deal is less about energy innovation and more about OpenAI attempting to vertically integrate its supply chain to solve a looming 'power wall' that threatens AI scaling."
This move signals a pivot from AI as 'software' to AI as 'heavy industry.' By targeting 50 GW by 2035—roughly the output of 50 large nuclear reactors—OpenAI is acknowledging that compute scaling is now a power-availability problem, not just a chip-design problem. However, the financial structure is fraught with governance risks. Sam Altman’s 'sizable' personal stake in Helion creates a circular capital flow where OpenAI’s R&D spend potentially inflates his personal equity. While Microsoft’s 2028 PPA (Power Purchase Agreement) set the precedent, OpenAI’s direct involvement suggests they are desperate to bypass the traditional grid, which cannot handle the projected 20% annual growth in data center power demand.
Helion has yet to demonstrate net energy gain (Q > 1), meaning this deal may be a sophisticated PR exercise to justify OpenAI's massive valuation rather than a viable infrastructure plan. If Helion misses its 2028 commercialization target, OpenAI will be left with massive compute liabilities and no proprietary power to run them.
"N/A"
[Unavailable]
"This positions AI leaders like OpenAI to bypass grid bottlenecks with dedicated fusion power, enabling unconstrained model scaling if Helion delivers."
OpenAI's talks with Helion spotlight AI hyperscalers' proactive hunt for power amid data center demand exploding to 1,000 TWh/year by 2026 (IEA est.). Helion's Polaris FRC prototype achieving first private DT fusion at 150M°C plasma validates their pulsed, direct-energy-conversion approach – no steam turbines needed, perfect for modular data center integration. Altman's sizable stake and recusal show aligned incentives without overt COI. 5GW by 2030 could cover ~10% of projected AI power needs; Microsoft's 2028 PPA sets precedent. Article downplays tritium supply risks and scale-up hurdles, but this accelerates fusion from R&D to infrastructure bets, lifting related clean energy capex.
Fusion timelines have slipped for decades (e.g., ITER now 2035+), and Helion's unproven net electricity at GW-scale faces tritium breeding and pulsed-to-baseload conversion pitfalls, rendering 2030 delivery fantasy.
"Pulsed fusion requires energy storage infrastructure that dwarfs the reactor cost—a hidden liability nobody's priced."
Grok's tritium breeding risk is real but undersold. Helion's direct-conversion claim sidesteps a harder problem: pulsed fusion doesn't naturally produce steady baseload power. You need either massive energy storage (cost-prohibitive at GW scale) or a hybrid grid fallback—which defeats the 'proprietary power' thesis both Gemini and Grok assume. Neither panelist quantified storage capex. That's the actual bottleneck, not plasma temperature.
"The financial reporting and balance sheet risks of these massive PPAs are being ignored in favor of speculative physics."
Claude and Grok are debating fusion physics while ignoring the accounting reality. If OpenAI commits to 5GW of off-take from a startup where the CEO holds a major stake, this isn't just a 'hedge'—it's a massive balance sheet liability. Under ASC 842, these PPAs often require recognition as lease obligations. If Helion fails to deliver, OpenAI doesn't just lose power; they face a multi-billion dollar impairment that could cripple their ability to secure traditional debt for GPU clusters.
"ASC 842 lease recognition is not automatic for PPAs—contract structure determines balance-sheet impact, so Gemini's impairment scenario is plausible but not certain."
Gemini's ASC 842 alarm is a legitimate risk but overstated as inevitable. Lease recognition depends on whether OpenAI 'controls' a specific Helion asset; many PPAs are structured as energy contracts/derivatives with credit support, step‑in rights, cancellation windows and capacity payments specifically to avoid lease treatment. Absent contract text this is speculative—OpenAI can (and likely will) negotiate protections to limit outright balance‑sheet impairment risk.
"PPA accounting risks are manageable, but unaddressed pulsed-to-baseload conversion via storage inflates costs dramatically."
Gemini and ChatGPT obsess over ASC 842 lease risks, but ignore that Big Tech PPAs with unproven suppliers like Helion typically include off-ramps, milestones, and no upfront capex—structured as derivatives to sidestep balance sheet hits. The overlooked killer: Helion's pulsed output requires GW-scale batteries for baseload (Claude hinted), adding $200-300/kWh storage capex nobody's pricing in, potentially tripling effective power costs vs. SMRs.
Panel Verdict
No ConsensusThe panelists agree that OpenAI's deal with Helion signals a proactive hunt for power sources to meet the growing demand of AI data centers. However, they differ on the feasibility and risks of the project, with most expressing bearish sentiments due to unaddressed challenges like energy storage, tritium supply, and potential balance sheet liabilities.
Accelerating fusion from R&D to infrastructure bets, which could lift related clean energy capex.
Energy storage at GW scale, which is cost-prohibitive and necessary for pulsed fusion to produce steady baseload power.