Three Chinese state institutes hold more fusion patent families than every venture-backed fusion company combined. Remove Chinese-only filings and that lead almost entirely disappears: just 2.4 percent of their families ever publish outside China.
Commonwealth Fusion Systems closed another $1 billion on July 30, taking it to $4 billion raised since 2018. The Fusion Industry Association’s July report counts $14.24 billion of private investment across 56 companies, with $4.48 billion of it arriving in the twelve months to July, a record year and 69 percent above the one before.
Run the patent landscape against those names and they are not where the depth is. Three Chinese state institutes hold more fusion patent families than every venture-backed fusion company in the world combined, several times over.
Then remove Chinese-only filings, and that lead almost entirely disappears.
The landscape is 5,388 extended patent families published since 2011: everything classified as a fusion reactor, the magnetic-confinement and plasma-heating classes, and anything calling itself a tokamak, stellarator, inertial confinement system or fusion reactor in its title, abstract or claims. Building it took four passes, because “fusion” in patent text usually means sensor fusion, spinal fusion or fusion proteins, and “plasma confinement” is standard vocabulary for the rings inside a semiconductor etch chamber. Left unconditioned, that one phrase put Lam Research in the top ten fusion patent holders.
Filings grew from 324 applications in 2011 to 931 in 2023, a 2.9x rise over twelve years. Steady, not explosive. The AI landscapes we have charted this year run 10x to 20x over the same span. Fusion looks like a mature industrial field that woke up rather than a new one being born, which fits a technology whose core concepts date to the 1950s and whose bottleneck has always been engineering.
The 2025 and 2026 bars are partial. Applications publish about 18 months after filing, so the right edge of the chart will keep rising for another year and a half.
Rank the landscape by patent families and ultimate owner:
Add up the fifteen Western startups we checked, from CFS down to Avalanche Energy, and they hold 188 families between them. The four Chinese state institutions hold 1,406.
The company that raised $4 billion holds 15 families in this landscape. Some of the foundational CFS magnet work sits under MIT, where the technology originated, so the honest CFS number is somewhat higher. It is not 700.
The fission-era incumbents are not in this fight either. General Atomics, which runs the DIII-D tokamak for the Department of Energy, shows 2 families. Mitsubishi Heavy Industries 6, Toshiba 9, Hitachi 2. ITER, the multi-decade international project in France, holds none, because it is a construction consortium whose members patent in their own names.
It is tempting to set $14.24 billion of private capital against 1,406 Chinese families and call it a contradiction. It is not one, because the two figures do not measure comparable things.
The FIA number counts disclosed private investment in companies that choose to report it. China’s fusion program is state-funded, and its budget is not published on those terms. The public estimates that exist are indirect: Neutron Bytes, citing the head of the Department of Energy’s fusion office, puts China’s annual fusion spending near $1.5 billion against a US Fusion Energy Sciences budget of $806 million for 2026, and reports that China mobilized at least $6.5 billion in public funds between 2023 and 2025, with some estimates running far higher. Beijing also capitalized a national fusion company in July 2025 with $2.1 billion in registered capital. Those are reported figures from a secondary source, not audited disclosures, and they should be read that way.
So the right reading is not that Western money produced Chinese patents. It is that private capital is visible and concentrated in a dozen Western companies, state capital is large and largely unlit, and patent filings are one of the few signals both sides emit in a comparable format. That is why the filing data is worth reading closely, and also why it should not be turned into a funding scoreboard.
Here is the finding that changes what the depth means.
Take the same landscape and keep only families with at least one publication outside China. That leaves 2,112 of 5,396 families, about 39 percent. Re-rank the owners on that slice and the Chinese Academy of Sciences, holder of 773 families, is left with 30. Southwestern Institute of Physics drops from 458 to 2. Fusion New Energy (Anhui) holds 170 families and not one of them publishes outside China.
Across the four Chinese state institutions, 32 of 1,345 families ever leave China. That is 2.4 percent.
Every Western and allied holder in the set sits at 100 percent: Tokamak Energy 74 of 74, TAE 24 of 24, General Fusion 20 of 20, CFS 15 of 15, First Light 14 of 14, Helion 8 of 8, MIT 19 of 19, UKAEA 31 of 31, Rosatom 57 of 57, the US labs 53 of 53.
On that international slice, Tokamak Energy is the largest corporate filer, followed by TAE. In European filings specifically, no Chinese entity appears in the top ten at all. In US filings the Chinese Academy of Sciences scrapes in near the bottom of the top ten.
A patent only excludes competitors in the country that granted it. A Chinese patent is enforceable in China. It does nothing to stop a company building or selling a reactor in the United States, Europe, Japan or Korea. So the 1,406-family position is formidable inside the world’s largest energy market and close to inert outside it. Read as a global blocking position, it barely exists. Read as a home-market position, it is close to unassailable, and any Western company that eventually wants to sell into China will be negotiating with institutions that got there first.
There is a second, simpler reading. Patenting abroad is expensive: translations, national-phase fees, local counsel, renewals in every jurisdiction. You do it when you intend to sell or license abroad. A program building reactors for its own grid has little reason to pay it.
Split the landscape by subject and the sector’s attention sits on the parts of the machine that take the beating, not on plasma physics.
The divertor is the exhaust: the component that must survive a heat flux comparable to a rocket nozzle for years. It is a materials and manufacturing problem, and it is where the Chinese institutes are most prolific. The newest filing in our exemplar set is a Hefei divertor design, filed in June and published September 11: tungsten-armored copper blocks on a cooling tube with an insulating layer whose conductivity varies around the circumference so the corners of the tungsten stay within their limit. Component-level, manufacturable, out the door in three months, and filed only in China.
The Western filings point somewhere else. CFS published a PCT application in May on a power management system coupling a pulsed fusion device to a utility grid through fast-discharge storage. Helion’s 2024 Canadian filing covers recovering electricity directly from an expanding plasma into the same coils that compressed it, the core of its no-steam-turbine design. Tokamak Energy published a two-stage tantalum and tungsten neutron shield in July, working around the central-column constraint that limits its compact spherical machine. Those are power-plant problems, filed internationally.
Two programs, two theories of where the hard part is, and two entirely different filing footprints.
One caution for anyone reading a fusion portfolio claim. China’s share of this landscape is 69.4 percent of families but only 36.4 percent of applications and 37.8 percent of publications. The family basis flatters a filer who files once per invention; the application basis flatters a filer who takes each invention into many countries.
The same effect shows up between companies. TAE holds 544 applications across 24 families, roughly 23 filings per invention. Tokamak Energy runs 617 across 74. The Chinese Academy of Sciences runs 801 across 773: one filing per invention, nearly all domestic.
Neither strategy is wrong. They encode different commercial plans. But if a deck claims the largest fusion patent portfolio in the world, ask which basis it counted and which countries it counted in. Those two questions move the ranking more than anything else on the page.
For founders and supply-chain companies: the crowded lanes are plasma-facing components and magnets, and the Chinese institutes are prolific in both, though mostly at home. The lanes with commercial pull and comparatively few families sit between the machine and the grid: pulsed-power conditioning, direct energy conversion, tritium handling at plant scale, and predictive plasma control. Princeton’s PACMAN control system, tested at DIII-D this month, predicts a tearing-mode instability 200 milliseconds ahead and issues commands in 20. That whole category is 314 families today.
For investors: separate filing strategy from depth, and ask where the filings are. A 773-family portfolio that lives in one country is a different asset from a 74-family portfolio filed across a dozen. Both can be the right answer. Only one of them constrains a competitor in your market.
For anyone planning to build, license or partner in China: the counterparties are not startups. They are the Chinese Academy of Sciences and CNNC’s Southwestern Institute of Physics. The Hefei BEST tokamak is in final assembly, targeting completion in 2027 and a net-gain demonstration by 2030, and the domestic IP position around it is already built.
Fusion’s disclosed private capital has gone almost entirely to a dozen Western companies. Its patent families have gone, by a wide margin, to a handful of Chinese state institutes. Both of those statements are true, and neither causes the other.
What the filing map actually shows is two programs optimizing for different markets: one filing many domestic inventions around machines it intends to build at home, the other filing fewer inventions across many countries around plants it intends to sell. If you are building in the fusion supply chain, the question is not how crowded the field looks in total. It is which lane you are in, and whether your filings reach the countries you actually plan to sell to.
Methodology: Patent data from Minesoft Origin. The landscape was searched September 14, 2026 and the international slice re-run September 16, 2026; the landscape grew from 5,388 to 5,396 families between the two runs as new publications entered the database, and each figure below is labeled with the run it came from. Landscape definition: CPC G21B (fusion reactors), CPC H05H1/02 through H05H1/22 (plasma confinement and heating), or fusion-specific keywords in title, abstract or claims with “fusion energy,” “fusion power” and “plasma confinement” conditioned on a nuclear-context term, and data, sensor, image, spinal and protein fusion excluded from the keyword arm; publication date 2011 or later. The landscape is 5,388 extended families, 10,482 applications and 14,669 publications, all describing the same set. Owner, sub-area and international-reach figures are extended-family counts from one query per entity and are not deduplicated across entities. The “outside China” slice is the landscape excluding publications with country code CN, which counts families having at least one publication outside China; 462 families publish on both sides. The filings-by-year chart uses application-level counts. The applications-per-family ratio and the 2.4 percent reach figure are derived; 32 of 1,345 families is 2.4 percent. Country shares are publication-level. Owner attribution follows Minesoft’s ultimate-owner rollup. Patent reads are title, abstract and first claim only; nothing here is a statement of claim scope, validity or freedom to operate. Funding figures from Commonwealth Fusion Systems (July 30, 2026), the Fusion Industry Association 2026 report (July 13, 2026), and Neutron Bytes (December 14, 2025) for public-spending comparisons, which are that outlet’s reported and attributed figures rather than audited disclosures.
Work with ipCapital Group
From invention to monetization, our team has guided 2,000+ engagements across the full IP lifecycle. Start with a free 30-minute discovery call.
Written by
Seth Cronin