China Built A 582-Tonne Magnet. It's Kind of a Big Deal.
china
That is not a projection. That is not a roadmap. That is not a "targeted deployment window" or a "path to commercialization" or any of the other phrases that energy executives use when they want to describe something that does not exist. It is a physical object. It is twenty-one meters long, twelve meters wide, shaped like a capital D, and it is sitting in a building in Anhui province right now, having passed expert acceptance and full-parameter testing.
NEW: 🇨🇳 China's nuclear fusion research has completed a massive magnet weighing 582 tons. The goal is an "unlimited energy" artificial sun. They plan to start the first power generation by 2030. A safe, CO2-zero ultimate energy race.
I want to be extremely clear about why this matters, because the coverage has almost entirely missed it.
The magnet is a toroidal field coil, the thing that generates the magnetic cage holding a hundred-million-degree plasma away from the walls of a reactor that would otherwise be instantly destroyed by it. According to the Chinese Academy of Sciences, it has roughly 1.3 times the volume and three times the stored energy of the equivalent coil at ITER, the thirty-five-nation megaproject in southern France. ASIPP researcher Wu Yu told CCTV that "16 of these coils will eventually be assembled" into a full magnetic field, each running at 100 kiloamperes and producing 6.5 tesla at the center. Your hospital MRI runs at one and a half to three.
Institute director Song Yuntao said the coil took six years. He also said the specialty steel, the insulation and the superconducting wire were all made in China. Every critical technology, per Xinhua, developed domestically.
Hold onto that last part. We are coming back to it.
Meanwhile, in the country that invented the venture capital press release
Here is what the American fusion industry has produced in roughly the same period.
In May 2023, Helion Energy signed a power purchase agreement with Microsoft to deliver at least 50 megawatts of fusion electricity by 2028. CEO David Kirtley described it to The Verge as a "binding agreement that has financial penalties" if the company cannot build a working fusion system. Nobody has ever seen those penalties. Nobody outside the two companies knows what they are. The terms were not disclosed. Helion has since raised over a billion dollars at a $5.4 billion valuation, is chaired by Sam Altman, and has broken ground on a plant in Washington state that is supposed to power Microsoft data centers.
Google has signed a 200 MW offtake with Commonwealth Fusion Systems, whose SPARC machine was going to demonstrate net energy in 2025, and then 2026, and whose commercial ARC plant is meant to start construction in 2027 or 2028.
Do you understand what is happening here? These are not energy contracts. These are narrative instruments. A hyperscaler with an enormous and politically inconvenient electricity problem signs a piece of paper with a fusion startup, the startup gets a valuation and a headline, the hyperscaler gets to appear in a story about clean baseload power instead of a story about how much coal its data centers are burning, and the actual electricity arrives at a date that is always four to five years out and always will be.
Data Center Dynamics put it about as well as anyone has: the real question is not whether Helion can deliver by 2028, but "why on earth is Microsoft pretending to believe that it can?"
Here is the state of play in 2026, stated plainly by an industry tracker rather than by me: no commercial fusion plant has produced electricity for a grid, and "no company is within twelve months of doing so".
Not one. Not close. And yet the funding keeps arriving, because the thing being funded is not a power plant. It is the option on a power plant, which is a much better product, because options never have to work.
Now, the part where I am also mean to China
I am not here to tell you that Hefei has solved fusion, because it has not, and the failure modes are not magnetic.
An ASIPP researcher told Global Times that the completed magnet testing represents "about 80 percent of the task". Eighty percent, in engineering, is the number you announce right before you discover where the remaining hundred percent of your problems were hiding. Sixteen coils must be built, not one. They must be assembled onto a Dewar base that will carry 6,500 tonnes at tolerances that do not care how heavy the individual pieces were.
And then there is tritium, which is the thing nobody wants to talk about at conferences.
Daniel Jassby spent his career as a plasma physicist at Princeton Plasma Physics Laboratory and has spent his retirement being the most inconvenient man in the field. His position on deuterium-tritium fusion, given to Science, is not nuanced: "This makes deuterium-tritium fusion reactors impossible." His argument is that tritium does not occur naturally in useful quantities, that the world's usable supply comes as a byproduct of Canadian CANDU fission reactors, that a reactor must therefore breed its own fuel in a lithium blanket, and that nobody has ever demonstrated this in a closed loop on a working machine. Writing in the Bulletin of the Atomic Scientists, he described ITER as a "havoc-wreaking neutron source" rather than a power producer. UCLA's Mohamed Abdou, who is considerably more sympathetic to fusion, has made essentially the same structural point about fuel supply from the other direction.
BEST, the Burning Plasma Experimental Superconducting Tokamak that these magnets are being built for, is scheduled for completion by the end of 2027 and is meant to demonstrate fusion electricity around 2030. It intends to try tritium breeding. Trying is not succeeding. And nearly all the performance data we have comes from institutional announcements and state media, which is not an accusation of fraud, it is a statement about what kind of evidence we are working with. So no, I do not think a Chinese fusion plant is going to be selling electrons in 2030.
But here is the thing
Ask yourself why ITER is late? It is not late because the physics got harder. It is late because it was designed as a diplomatic instrument, funded in-kind by thirty-five countries, with components manufactured across four continents and shipped to Provence to be fitted together by people who did not build them. The current baseline has research operations starting in 2034 and deuterium-tritium operation in 2039, with roughly five billion euros in additional cost. When director-general Pietro Barabaschi announced the delay, he described the abandoned first-plasma milestone as "rather symbolic", which was honest and also one of the most quietly brutal things anyone has said about fifteen years of work.
Fusion has never really been a physics problem. It is an industrial problem wearing a physics costume. And industrial problems are won by whoever can specify a superconducting cable on Monday and have a domestic supplier quoting on Tuesday without needing a treaty to do it.
That is the actual story in Hefei, and it is why the line about domestic production matters more than the tonnage. China has designated fusion one of eight priority frontier technologies in its five-year plan. It is expanding a fusion technology center at Tianfu, in Chengdu, explicitly to build out supply chain and industrial replication capacity, which is a genuinely strange thing to spend money on unless you expect to eventually need to build these machines repeatedly.
Meanwhile the Western fusion sector's most celebrated achievements of the last three years are: a power purchase agreement with undisclosed terms, a valuation, a groundbreaking ceremony, and a series of slipped dates.
I am not saying China will get there first. I am saying that one of these systems produces 582-tonne objects and the other produces funding rounds, and if you have spent any time watching what China is doing to the AI industry, you should recognize the pattern and be just a bit worried. I am.
Watch coil number two. Watch coil number seven. Watch whether Hefei can build sixteen of these on a schedule, and then start again for the next machine. That is the only metric that means anything, because fusion does not become real when a reactor works. It becomes real when reactors become manufacturable.
Dr. Kai Nakamura makes quantum computing and frontier physics legible — separating the genuinely near-term from the perennially five-years-away.
