Note: This is a research note supplementing the book Unscarcity, now available for purchase. These notes expand on concepts from the main text. Start here or get the book.
Energy Sovereignty Is Compute Sovereignty
Or: Why a Search Company and a German Utility Just Bought a Piece of a Star
On July 7, 2026, Google and the German utility RWE did something that would have looked absurd a decade ago: they put money into a company that has never sold a single kilowatt-hour, cannot yet fuse atoms at commercial scale, and won’t have a demonstrator running until the early 2030s. The company is Proxima Fusion, a Munich startup building a stellarator, a fusion reactor that pins superheated plasma in place with a cage of twisted magnets. The round was €411 million (about $468 million), led by XTX Ventures and East X Ventures, and it valued a pre-revenue physics project at $2.7 billion.
Notice what Google and RWE bought. Not electricity. Not a power-purchase agreement promising megawatts in 2035. They bought equity — a piece of the company itself. RWE went further still, committing to help build a fusion plant on the site of the former Gundremmingen nuclear station in Bavaria. It’s a bet on where the money is headed next: once compute and intelligence both go cheap, value keeps sliding down to whatever’s still scarce, and few things are scarcer than a working reactor.
That distinction is the whole story, and almost everyone is missing it. We keep describing the collision between AI and energy as a supply problem: too many data centers, not enough electrons, someone go build more. But the smartest players have stopped acting like customers shopping for power. They’re acting like conquerors claiming territory. Because in the intelligence age, energy sovereignty is compute sovereignty — and whoever owns the reactor owns the compute.
The Difference Between Buying Power and Owning the Machine
Start with the boring-sounding thing that turns out to be everything: the difference between an offtake deal and an ownership stake.
A power-purchase agreement is a promise. Google signs a contract to buy 200 megawatts from Commonwealth Fusion Systems’ first ARC plant in Virginia, and in the early 2030s, if the physics cooperates, electrons show up and Google pays for them. That’s how Google backed CFS in 2025: a 200 MW offtake from a 400 MW plant. It’s how Microsoft backed Helion, contracting for power from a Central Washington plant due in 2028, a date Helion still keeps for “initial operations,” though in September 2026 it quietly dropped the promise that the plant would produce electricity by then. A promise is exactly what a PPA is. This is the model our Fusion Timeline has tracked all along, and the model our Electron Gap analysis assumed: hyperscalers as buyers of a scarce input.
Equity is a different verb. When Google increased its stake in CFS (it has been an investor since 2021), and when Google put money into TAE Technologies in 2022, and now into Proxima in 2026, it stopped being a customer and became an owner. Owners don’t get in line for electrons. They get a seat at the table where the roadmap is set, the magnet supply chain is allocated, and the first plants are sited. They get priority. In a world where the reactors that work will, at first, be few and precious, priority is the entire game.
Put bluntly: a PPA buys you power if the plant gets built. Equity buys you a say in whether, where, and for whom it gets built. One is a receipt. The other is a claim on the future.
Three Ways to Own a Sun
Once you see the shift from buying power to owning reactors, the global picture resolves into three competing ownership models — three answers to the question “who should hold the equity in the machine that ends energy scarcity?”
The American model: hyperscaler equity. In the United States, fusion is being privatized into the balance sheets of the companies that need it most. Google holds stakes across CFS, TAE, and now Proxima. Microsoft is wedded to Helion, which raised $465 million in June 2026 at a $15.5 billion valuation, extended the round to $500 million in September with pension and sovereign wealth investors. In June Helion also collected the two Washington State licenses it calls the first regulatory licenses ever granted for a fusion power plant. In March, Axios reported OpenAI in advanced talks to buy the equivalent of 5 gigawatts of Helion’s output by 2030, scaling to 50 by 2035; no agreement has been announced since. Nvidia and Siemens are building a digital twin of CFS’s SPARC machine, which had eight of its eighteen toroidal field magnets installed by late September and which CFS says will start operating within months, targeting net energy in 2027. And on July 30, 2026, CFS raised another $1 billion, taking its total to $4 billion, in a round that drew pension funds, sovereign wealth funds and infrastructure investors alongside the venture firms: the money that finances toll roads and transmission lines has decided a reactor is infrastructure, and it wants the equity too. In early October Type One Energy closed a $200 million Series B co-led by Breakthrough Energy Ventures, with Siemens Energy Ventures joining, for a plant at the Tennessee Valley Authority’s Bull Run coal site. American private fusion has absorbed the lion’s share of the sector’s money. Add up TechCrunch’s list of every startup that has raised over $100 million and roughly 85% of it sits in US companies, and the Fusion Industry Association’s July 2026 survey counted a record $4.48 billion raised in twelve months, $14.24 billion since 2021. The reactor and the data center are converging into a single, privately-owned stack. This is the world our Compute Clusters and Compute Landlords notes describe, extended one layer down: the landlords are no longer content to own the GPUs. They want to own the power plant too.
The European model: utility, state, and hyperscaler in a blend. Proxima is the tell. Its cap table mixes a sovereign-adjacent utility (RWE), a hyperscaler (Google), venture capital (XTX, East X), and public money (the EU’s EIC Fund, Germany’s KfW Capital). Europe, having spent the 2010s learning what energy dependence costs when a pipeline gets turned off, is treating fusion as strategic infrastructure to be co-owned, not outsourced. RWE building on a decommissioned nuclear site isn’t nostalgia; it’s a bid to keep the generating asset, and the sovereignty that comes with it, on European soil. Two months after the round, Proxima showed how far down the stack that logic runs. On September 9, 2026 it signed a memorandum with Lower Saxony for a roughly €140 million factory for fusion-grade high-temperature superconducting tape, the magnet material its stellarator is wound from, with the state putting in €21 million and the site still to be chosen. Its demonstration plant needs 20,000 kilometers of the stuff and a commercial plant twice that, and today most of the world’s supply is made in China and Japan. Owning the reactor is not enough if someone else owns the tape.
The Chinese model: the state owns the sun outright. China skipped the debate about who should hold the equity by having the government hold all of it. By the Special Competitive Studies Project’s conservative count, Beijing has mobilized at least $6.5 billion for commercial fusion since the start of 2023, almost three times what Washington appropriated for its fusion science program over the same period. Its BEST tokamak in Hefei, more than half assembled as of October 2026, is due to finish construction by the end of 2027 and aims to demonstrate fusion electricity around 2030. “Controllable nuclear fusion” is written into the 15th Five-Year Plan as a national growth frontier. When the EAST “artificial sun” held plasma at 100 million degrees for 1,066 seconds in January 2025, it wasn’t a startup chasing a valuation. It was a state building an asset.
Three models, one contest. And the prize isn’t clean energy for its own sake. It’s the compute that clean, cheap, sovereign energy unlocks.
The Contrarian Part: Fusion Doesn’t End Geography. It Creates It.
Here’s where this note has to argue with its own neighbors.
Elsewhere on this site, we’ve made the case that abundant energy dissolves the old logic of territorial conflict. Our Geopolitics of Abundance note argues that when solar and fusion make resources plentiful, resource wars become economically irrational and nations reorganize as Network States. Our fusion coverage celebrates energy “decoupled from geography.” Both are right — eventually. That’s the destination.
But the road there runs the other way. During the build-out decade (call it now through the late 2030s), fusion is the most geographic technology on Earth. The first reactors will be scarce, staggeringly expensive, and physically located somewhere: Chesterfield County, central Washington, Bavaria, Hefei. Whoever owns those first machines, and wherever they sit, gets first claim on the compute frontier that runs on them. A frontier AI lab in the 2030s won’t ask “is energy abundant globally?” It will ask “can I get five gigawatts, behind the meter, next to my GPUs, on friendly soil, before my rival does?”
That is a sovereignty question, not a supply question. And it’s why the fight over who pays for AI’s electricity, the ratepayer battle we cover in Who Pays for AI’s Electricity?, is only the opening skirmish. The deeper contest is over who owns the generator. Cost allocation decides who foots the bill this year. Reactor equity decides who leads in artificial intelligence for a generation.
Fusion will end geography. But not before it makes geography, briefly and intensely, destiny.
Why the Book Saw This Coming (and What It Warns)
Call this moment the Terawatt pivot: the point where energy becomes the substrate of everything, because once energy is nearly free, food, water, compute, and manufacturing follow. The book’s Foundation, the layer of housing, food, healthcare, and energy delivered to each person in a connected Commons, is built on that substrate. No cheap terawatts, no Foundation.
Which is exactly why the ownership question matters so much. The book’s optimistic case is that abundant energy becomes a commons, delivered like the Free Zones deliver housing: on by default, for everyone. But look at the transition actually unfolding. The substrate of the abundance economy is being pre-purchased, right now, as equity — by four hyperscalers, a handful of utilities, and two governments. If the machines that end scarcity are owned by the incumbents of the scarcity era, abundance doesn’t automatically become a commons. It becomes a product, metered and gated by whoever holds the cap table.
This is the capture risk the book spends its second half designing against. It’s why Transition Trusts exist — to route dying fortunes into public infrastructure rather than private moats. It’s why the Sovereign EXIT Protocol tries to convert nation-states from rivals racing for the reactor into founding stewards of a shared grid. And it’s why the framework insists on predistribution — structuring ownership before the concentration hardens, not taxing it afterward.
The Proxima deal isn’t a villain. Google and RWE funding a European stellarator is genuinely good news for the timeline. But it’s a signal worth reading honestly: the race to own the sun has started, the equity is being claimed while it’s cheap, and nobody has yet answered the question the book keeps asking. When energy becomes the thing everything else is made of, is it a commons — or is it the most valuable private asset in history?
Whoever owns the reactor owns the compute. The only question left is whether “whoever” is a shareholder, a state, or all of us.
Further Reading
- The Electron Gap - Why energy, not chips, is the real bottleneck on AI, and why “energy independence” increasingly means “AI independence.”
- Fusion Timeline 2024-2030 - The companies, the milestones, and the money behind the commercial fusion race.
- The Geopolitics of Abundance - The counter-thesis: how abundant energy dissolves the logic of territorial war (true at the destination, not on the road there).
- Compute Clusters - The $800B GPU arms race the energy race sits beneath.
- Compute Landlords - When the builders of compute become its rentiers: the ownership-is-control logic, one layer up from the reactor.
- Who Pays for AI’s Electricity? - The ratepayer fight over the AI buildout, the skirmish before the sovereignty war.
- Who Vetoes the AI Buildout? - The siting fight: who gets to veto a data center, the third front alongside cost and ownership.
- Transition Trusts - How to fund the reactors without letting them become private moats.
- The Foundation - Why the whole abundance stack rests on cheap energy.
- The Commoditization of Intelligence - Why the cheap, plentiful supplier - not the smartest lab - captures the market, and can ration it.
- Robots That Build Abundance - The same ownership question one layer further down: when robots build the energy machines faster and cheaper, who keeps the compounding surplus?
References
- Proxima Fusion Raises €411 Million to Build Europe’s Commercial Fusion Champion - Company announcement, July 2026
- Google backs Proxima Fusion in $468 million round - CNBC
- Germany’s Proxima Fusion secures $2.7B valuation as Google backs Europe’s reactor race - PitchBook
- Google and Commonwealth Fusion Systems Sign Strategic Partnership - 200 MW PPA plus increased equity stake
- Every fusion startup that has raised over $100M - TechCrunch, sector funding landscape
- Commonwealth Fusion Systems raises another $1B, reaching $4B total - TechCrunch, July 30, 2026
- Fusion industry attracts record annual funding of $4.48bn, raising total to $14.24bn - Fusion Industry Association, July 2026
- Proxima Fusion bets €140M on a critical fusion ingredient dominated by Asian suppliers - TechCrunch, September 10, 2026
- Helion hedges on how soon it expects to sell electricity from fusion - Axios, September 8, 2026
- Nuclear fusion included in the 15th Five-Year Plan’s key future growth frontiers - Global Times, April 2026
- Cash, scale and speed: why China’s $6.5 billion fusion push matters - Special Competitive Studies Project, on the at-least-$6.5-billion estimate since 2023
- BEST device scheduled for completion by the end of 2027 - Chinese Academy of Sciences, October 2025, and China’s artificial sun advances - Interesting Engineering, October 2026
- Scoop: OpenAI bets on Altman-backed fusion startup - Axios, March 23, 2026, the 5-to-50-gigawatt talks
- Helion clears key regulatory milestone - Helion, June 16, 2026, the two Washington State licenses
- Helion funding round grows to $500M - GeekWire, September 15, 2026
- Proxima Fusion and Lower Saxony sign MoU for the first large-scale fusion-grade HTS tape facility - Proxima Fusion, September 2026
- Eight installed toroidal field magnets of the 18 total - CFS on X, September 22, 2026, and CFS on starting SPARC operations in coming months - CFS blog, August 28, 2026
- Type One Energy raises $200 million Series B - NucNet, October 9, 2026
- Unscarcity, Chapter 1 (The Foundation) and Chapter 10 (Geopolitics of Abundance)
Wealth used to mean owning the oil field. Now it means owning the reactor - because the reactor makes the electrons that make the intelligence that makes everything else. Debate the ownership question at unscarcity.ai/forum.