Hydrogen-Boron Fusion Reaches 100M Reactions a Second in China

What ENN actually reported

On September 28, ENN Group announced that its EXL-50U, a compact spherical tokamak, had produced hydrogen-boron fusion reactions. The company is based in Langfang, Hebei province.1 ENN says no commercial fusion company had done this before on a machine it built itself.1 The headline figure is a reaction rate above 100 million fusions per second.2 Chinese-language accounts also describe the result as the first clean, neutron-free fusion with advanced fuel on any Chinese magnetic confinement device.910

The method matters as much as the figure. The team did not try to heat the whole plasma to the extreme temperatures that proton-boron fusion normally needs. Instead, it paired high-energy neutral beam injection with radio-frequency waves to push protons into the energy band where the reaction is most likely, which physicists call the "first resonance peak."1 That produced a large population of "fast protons" that hit boron nuclei and pushed the total reaction rate past the 100-million mark.2 Fuel was loaded as diborane, a hydrogen-boron compound, together with hydrogen gas.17

ENN says more than ten fusion experts from international institutes and universities reviewed the data. According to the company, they agreed unanimously that the alpha particles produced by the reaction had been measured effectively and repeatably, and called the work a "significant and valuable contribution" to the field.1 Alpha particles are the signature product of the reaction, which makes them the key evidence.7

Why boron, and why it is so hard

Most mainstream fusion programs, including ITER in France, use deuterium-tritium fuel.1 That fuel has real drawbacks. Tritium is scarce, radioactive and expensive, and the reaction releases high-energy neutrons that damage reactor walls and make nearby components radioactive.2 About 80% of the energy from a deuterium-tritium reaction leaves as an uncharged neutron, which magnetic fields cannot contain.7

In the proton-boron reaction, a proton fuses with a boron-11 nucleus and splits into three helium nuclei, releasing about 8.68 MeV.10 The main reaction gives off no neutrons, but side reactions still produce a small number.7 Boron-11 makes up about 80% of natural boron, which is already mined in bulk for glass, detergents and fertilizer.7 Because the products are charged, future plants might turn their motion straight into electricity without a steam turbine, though that is still largely theoretical at power-plant scale.7

The drawback is physics. Boron has five protons, so it repels an incoming proton far more strongly than hydrogen isotopes repel each other.7 One report puts ENN's long-term operating target at around 1.8 billion degrees Fahrenheit, against ion temperatures of about 72 million degrees so far on the EXL-50U. That is a gap of roughly 25 times.18

Keeping the number in proportion

Coverage agrees on the basic facts but differs on what they mean. One analysis worked out that 100 million reactions per second, at about 8.7 MeV each, comes to roughly 0.14 milliwatts of fusion power. That is a tiny fraction of what the machine uses to heat and confine its plasma.7 The experiment was never designed to produce net energy. Its aim was to show measurable, verifiable proton-boron fusion in a magnetically confined plasma.7

Some summaries go further than the evidence. One described the result as a self-sustaining chain reaction "at a commercially meaningful scale."4 Another said a laser was used to generate the plasma.3 Neither claim matches the technical accounts, which describe neutral beams, radio-frequency heating and fast protons acting as a catalyst.12 The milliwatt-level output alone rules out the "commercially meaningful" reading.

The word "first" also needs care. In March 2023, Japan's National Institute for Fusion Science and TAE Technologies of the US carried out the first hydrogen-boron fusion experiment in a magnetically confined plasma.10 ENN's claim is narrower: the first commercial company to do it on its own device. That is accurate, but it is a smaller claim than some headlines suggested. TAE and Germany's Marvel Fusion are also pursuing this fuel.1

Temperature figures also need care. One timeline says the EXL-50U's electron temperature passed 100 million degrees Celsius in early 2026.10 Yet ENN's chief engineer, Yang Yuanming, marked the new result by promising to get the plasma to 100 million degrees "as soon as possible."1 The likeliest explanation is the difference between electron and ion temperatures. Proton-boron fusion depends on very hot ions, and ENN's own roadmap calls for ions much hotter than electrons.20 Readers should treat the two milestones as measuring different things.

A schedule that is both ahead and behind

ENN's timeline gives a mixed picture. In 2024 the company moved the goal of a first hydrogen-boron reaction from its next machine, Helong-2, to the upgraded EXL-50U.10 Hitting that goal before the end of 2026 has been presented as beating the schedule.8 ENN's 2024 roadmap paper, however, expected Helong-2 to be finished in 2026.14 The current plan is for equipment installation by June 2027 and the start of experiments by the end of 2027.24 ENN has broken ground on the roughly 120-acre site and is aiming for its first hydrogen-boron electricity by 2030.28 It wants a demonstration reactor, with low-cost power, before 2035.28

The best case for ENN is that it builds hardware fast. EXL-50 took less than ten months to build, and the upgrade to EXL-50U took six.14 It has also earned some outside recognition. In April 2026 it joined physics work coordinated through ITER, only the second private fusion company to do so after Commonwealth Fusion Systems.18 Even so, the 2030 electricity target is a goal that has not been demonstrated, and the gap in ion temperature is the obstacle that will decide whether it is met.518

The AI-for-science thread

ENN lists fusion artificial intelligence among the five key technologies in its commercial plan, alongside reaction rates, high-parameter engineering, plasma temperature and direct conversion of alpha-particle energy into electricity.5 It has also set up a Fusion AI Lab with Nankai University.23 The company has not said AI was behind the September result. The connection is about where the field is heading.

The outside evidence for that direction is getting stronger. Princeton researchers recently tested PACMAN, an AI control framework, on the DIII-D tokamak in San Diego. Its control loop runs in about 20 milliseconds, and in one experiment it predicted a damaging tearing-mode instability around 200 milliseconds in advance and steered the plasma away from it.30 For a proton-boron program that depends on holding a fragile population of fast particles at the right energies, that kind of real-time control is likely to matter more over time. That is an inference, but it follows from the physics ENN describes.

Fusion's result and the October wave of AI proofs

The fusion announcement came just as the debate over AI-generated mathematics peaked, and the two stories share a theme. On October 6, OpenAI published 722 manuscripts grouped into 372 families of results on long-standing problems.31 They included a claimed "quasi-Riemann hypothesis": a proof that the Riemann zeta function has no zeros where the real part is greater than 7/8.32 A day later, OpenAI withdrew three manuscripts after a sign error broke a foundational argument, and revised fourteen others.34 As of October 13, about 42% of the catalogue had a completed formal proof in Lean, a language whose proofs a computer can check.34 Terence Tao, who before the release had called the coming flood of results "toxic waste," later revised that to "maybe treated waste."38

China's AI-for-science effort is producing similar results. Shanghai AI Laboratory said a team working with its Shusheng-S2 model had settled one case of the 31-year-old Ghys conjecture. A Peking University team then checked every logical step with more than 550,000 lines of Lean code.35 The proof can be checked by anyone, but the model behind it is not public, and its benchmark claims come only from the lab.35 Elsewhere, DeepMind reports that its AlphaProof Nexus system solved 9 of 353 open Erdős problems.37

All of these stories depend on how results are checked. ENN's claim rests on repeatable alpha-particle measurements and an expert review it organized itself.1 The strongest AI math claims rest on formal verification by machine.35 In both fields, producing an impressive result is getting easier, and the hard part is getting independent confirmation.34

The verdict

ENN's result is a real and carefully documented scientific milestone. It is not an energy breakthrough. It shows that a commercial company can produce measurable, repeatable proton-boron fusion in a tokamak with a targeted heating method. But the output is measured in milliwatts, and the plasma is still far too cool for a power plant. The most useful next step would be for ENN to publish its alpha-particle data for outside groups to scrutinize. Until then, the EXL-50U result is a verified first step toward a power-plant goal that has not been proven.