> ## Content Index > Fetch the complete content index at: https://nexi.fund/llms.txt > Use this file to discover other available public pages before exploring further. # Fusion's Procurement Race: AI Data Centers Are Buying Power a Decade Early - URL: https://nexi.fund/fusion-power-ai-data-centers-2026/ - Published: 2026-08-04T09:30:07.000Z - Updated: 2026-08-04T09:30:07.000Z - Description: Hyperscalers are signing fusion power contracts a decade before delivery. Helion's $465M Series G at a $15.5B valuation shows the money is now following the contracts โ€” and the bottleneck has shifted from plasma physics to magnet factories. - Author: Nexi.fund Labs - Tags: Energy & Climate, #mode-6, #hook-thesis, #track-F AI's power problem stopped being a footnote in energy news. It is now writing the term sheets of the fusion industry. ๐ŸŽฏ Fusion is being pulled out of the lab by a single customer class: hyperscale data center operators who are signing power contracts a decade before delivery. Helion just raised $465 million at a $15.5 billion valuation, nearly tripling its worth in five months, on the strength of a 50 GW procurement pipeline that does not yet exist. The binding constraint has moved from plasma physics to factory output. The industry's biggest bets are now on magnets, not reactors. $465M Series G, June 2026 โ†‘ 9% vs January Series F #### Helion's latest round Led by Thrive Capital, the round funds reactor manufacturing, not new physics. Total invested in the company now exceeds $1.5 billion. ยท *Helion, 2026* $15.5B post-money value โ†‘ 3ร— vs $5.245B #### Valuation jump in five months The January Series F valued Helion at $5.245 billion. June's round nearly tripled that on the back of data center power demand. ยท *Goodwin, DCD, 2026* 50 GW OpenAI target, 2035 #### Reported fusion offtake Talks reported in March would give OpenAI 12.5% of its output, ramping from 5 GW in 2030\. Neither company has confirmed. ยท *DCD, 2026* ## The procurement race is outrunning the technology Every company on this list is buying electricity from a reactor that has not generated a single watt yet. Microsoft signed a 50 MW power purchase agreement (PPA) with Helion in 2023, the first fusion offtake with a data center operator in the industry's history, with delivery scheduled for 2028\. Google committed to 200 MW from Commonwealth Fusion Systems (CFS), whose ARC plant is still years from first power. Reports in March put OpenAI in talks to take as much as 50 GW from the company by 2035. The order of operations is what changed. Fusion companies used to raise money to build a reactor, then look for a customer. Now hyperscalers hand over the contract first, and the financing follows. The money trail confirms it. Helion raised $425 million in January at a $5.245 billion valuation. In June it closed $465 million more at $15.5 billion, a near-tripling in five months, with Thrive Capital leading. Total capital in the company sits above $1.5 billion. Google has backed CFS and separately funded TAE Technologies past a $150 million round. Thea Energy, a Princeton spinout, pulled in a $100 million Series B in May. Fusion funding across the sector has climbed from roughly $1.7 billion in 2020 to more than $15 billion by the end of 2025, and the current surge is driven almost entirely by AI load. This is not a physics story anymore. It is a supply-chain story with a physics dependency. ## What 50 gigawatts actually requires Scale is where the promise and the risk separate. Helion's Orion plant, the first commercial reactor under construction in Washington state, is a 50 MW machine. Delivering 5 GW by 2030 means roughly 800 of those reactors in service. The 50 GW OpenAI target implies a build-out more than an order of magnitude beyond that by 2035. That is a manufacturing ramp with no precedent in energy history. Fusion reactors have only ever been built one at a time, as research instruments. Nobody has assembled one on a production line. It is building a 166,000-square-foot factory to try, which is the real use for the Series G money, not the prototype milestones the company announced in February, when its Polaris machine demonstrated measurable deuterium-tritium fusion and plasma temperatures of 150 million degrees Celsius. The physics milestone matters. The factory is the test that actually fails. โš ๏ธ **The timeline risk is the asset class risk** No fusion company has yet delivered net-positive electricity to a grid. Oak Ridge National Laboratory's director has pushed its own pilot-plant forecast from the early 2040s to the mid-2030s, still years behind its 2028 delivery promise. The gap between a signed PPA and delivered power is now the widest in any energy market. That gap is exactly what a sophisticated principal should watch. A PPA signed in 2026 for power in 2035 carries counterparty risk, technology risk, and permitting risk all in one instrument. The hyperscalers accept it because their compute build-out is already committed for the next decade, and every alternative (gas, renewables, fission) has its own binding constraint. Fusion is the only source that answers all of them at once. Which is why the buyers keep signing. ## The magnet factory is the new frontier Thea Energy's $20 million ARPA-E SCALEUP award is a bet on manufacturing, not physics. The Princeton spinout is using it to build the first domestic production lines of modular high-temperature superconducting (HTS) magnets, the components that confine plasma in most next-generation designs. The award landed weeks after Thea's $100 million Series B and the design review of its Helios power plant, a stellarator built around planar coil magnets that can be mass-produced instead of hand-wound. This is the quietest signal in the whole story, and it may be the most important. Every fusion architecture competing for data center contracts (Helion's magneto-inertial design, CFS's ARC tokamak, Thea's stellarator) depends on the same scarce component. The company that industrializes HTS magnet production does not need to win the reactor race to win the market. It can sell to all the losers too. #### Why magnets became the bottleneck Superconducting magnets are the single largest material cost in a modern fusion plant. They have historically been built as one-off custom units taking months each. Thea's design standardizes four magnet types for its twelve large coils and one type for the more than 300 smaller ones, turning the whole system into a factory product. **Confirmation criteria:** the ARPA-E program is explicitly designed to move past awards toward market adoption. Follow whether the HTS production lines ship modules to customers outside Thea's own plant. The same magnets have applications beyond fusion: compact systems for medical imaging, high-performance transport, and grid equipment. That hedges the thesis: even if the reactor timeline slips, the manufacturing capability retains a market. Investors in the fusion trade are effectively buying exposure to a superconducting-components supply chain that the energy transition needs regardless. ## The unit economics nobody has seen yet Every fusion PPA signed so far is priced against a number that does not exist: the levelized cost of energy (LCOE) of a reactor that has never produced power. The buyers are underwriting cost curves the same way early cloud customers underwrote unproven infrastructure, on the assumption that scale will do the work. There is no public tariff in any of the announced deals. Microsoft's 50 MW contract, Google's 200 MW commitment, and the reported OpenAI talks all price electricity at terms that will only make sense if fusion reaches a cost per megawatt-hour that competes with baseload gas or fission. Hyperscalers can afford to pay a premium for priority access (they have been doing it for grid power and water), but a data center operator's cost structure only tolerates that premium on the marginal load, not the whole fleet. The interesting part is what the investors are actually buying. A $15.5 billion valuation for a company with zero revenue is not a bet on electricity prices. It is a bet on scarcity: whoever delivers first, at any cost, locks in the decade-long demand from AI build-out. First-mover pricing power in fusion is the whole game, because the second mover competes against a grid that will have gotten cheaper in the meantime. โœ… **What the 2028 delivery date really signals** If it delivers first electrons on schedule, it resets the entire industry's cost benchmark and validates every earlier PPA. If it slips, the valuation story transfers to whoever has the shortest credible path, which is why the manufacturing race, not the physics race, is where the money is concentrating. For a principal evaluating private exposure here, the discipline is the same as in any pre-revenue technology: the balance sheet is not the company. The contract book is. Watch whether the 50 MW Microsoft offtake is joined by signed, disclosed terms from a second hyperscaler, and whether any customer ever publishes a price. Until one does, the $15.5 billion is a claim on an unproven cost curve. ## How the fusion offtake deals compare | Deal | Buyer | Seller | Size | Delivery | | ----------------- | ----------- | --------- | ------- | ----------- | | **Microsoft PPA** | โœ” Microsoft | Helion | 50 MW | 2028 | | **Google PPA** | โœ” Google | CFS (ARC) | 200 MW | early 2030s | | **OpenAI talks** | โ— OpenAI | Helion | 5โ€“50 GW | 2030โ€“2035 | Reported and confirmed fusion offtake for data centers, mid-2026\. OpenAI deal unconfirmed. The shape of these contracts is the same shape as the rest of the hyperscale energy build-out: physical delivery PPAs in place of spot-market exposure. The buyer is not hedging price. The buyer is hedging existence. ๐Ÿ“Š **Key signals to track** Whether it delivers first electrons to the grid from Orion in 2028, the date every other 2030s commitment is indexed to. Whether the OpenAI-Helion talks convert from reports into a signed contract, and at what capacity. Whether Thea's HTS magnet lines ship modules to outside customers, proving the component supply chain has real demand. Whether the next fusion funding round is led by energy investors or by more hyperscaler strategic capital. The thesis here is straightforward: fusion's economics were always going to be a manufacturing problem wearing a physics disguise. What changed in 2026 is that the biggest technology buyers on earth decided the disguise was worth paying for, in advance, on terms nobody in the industry had ever seen. The contracts are real. The reactor is the bet. [ Helion Raises $465 Million Series G Funding Round to Meet Surging Global Demand for Power Primary-source announcement of the round, its $15.5 billion valuation, and the manufacturing capacity it funds. Helion Energy ](https://www.helionenergy.com/newsroom/helion-raises-465-million-series-g-funding-round-to-meet-surging-global-demand-for-power?ref=nexi.fund) The company's own framing: the money is for production lines, not plasma research. [ Microsoft-backed fusion firm Helion raises $465m in Series G funding round Data Center Dynamics maps the full procurement picture: Microsoft's 50 MW PPA, Google's 200 MW deal with CFS, and the OpenAI talks. Data Center Dynamics ](https://www.datacenterdynamics.com/en/news/microsoft-backed-fusion-firm-helion-raises-465m-in-series-g-funding-round/?ref=nexi.fund) The clearest single source on the PPA race and the 2028 timeline risk. [ Thea Energy lands $20M ARPA-E grant to scale fusion magnet production Dealroom covers the ARPA-E SCALEUP award and why modular HTS magnet manufacturing is the industry's next competitive frontier. Dealroom ](https://app.dealroom.co/news/note/thea-energy-lands-20m-arpa-e-grant-to-scale-fusion-magnet-production?ref=nexi.fund) The manufacturing-bottleneck angle that connects the reactor race to the supply chain.