There's a class of engineering stories where the most interesting part happens not in the laboratory, but in the gap between two countries and three decades. China is building the CiADS facility in Huizhou — the world's first subcritical reactor that physically cannot be made to explode, because the chain reaction is sustained from outside by a proton beam from an accelerator. Turn off the beam — the reaction dies in seconds, like a palm lifted from a string. The Western press presents this as yet another "Chinese miracle of 2027," but in this story the miracle is only half Chinese. The concept itself — the "energy amplifier" — was invented in 1993 by Italian Carlo Rubbia at CERN, and thirty years ago his Nobel name thundered at press conferences from Geneva to Zaragoza. Then Europe quietly buried the idea — and here's what's curious: not because it doesn't work.
I was caught by a line in a tech digest: "China in 2027 launches a reactor that eats radioactive waste and physically can't explode." Beautiful, but typical science-pop framing: as if the idea was born in China along with the blueprints. A search check revealed a tastier structure: the ADS (accelerator-driven system) concept is European, originating from CERN in 1993, with a loud death rattle of the project in Spanish Zaragoza in 2001. So China isn't inventing — China is resurrecting. And the question "why does the same physics die in debt in one country and become a national-scale construction project in another" is more interesting than any physics. That's how this longread was born.
An ordinary reactor lives on the edge: its chain reaction is self-sustaining, the neutron multiplication coefficient k equals one, and the entire engineering architecture — rods, boron, negative void coefficient — exists for one purpose: to keep that edge from crossing itself. The accident scenario of any classical nuclear plant is ultimately "k accidentally became greater than one where it shouldn't." Chernobyl in dry physics terms is precisely runaway supercriticality.
A subcritical system is fundamentally different: in its fuel k is deliberately less than one, say 0.97, and the chain reaction cannot sustain itself in principle. Each fission event produces on average 2.43 neutrons, but only 0.97 "offspring" reach the next fission event — the reaction is doomed to decay exponentially. To keep it smoldering, the missing neutrons are supplied by an external source: a proton beam from an accelerator hits a heavy target of lead-bismuth alloy, and in the act of spallation — "chipping" nucleons off the nucleus — each proton yields 20–30 neutrons (World Nuclear Association, 2026 update).
Hence the main consequence: accident criticality simply doesn't exist as a class in such a machine. There's no "reactor ran away" scenario — just as there's no "flashlight exploded from overload" scenario if you've removed the battery. Uncontrolled runaway requires k > 1, and the system is designed so k doesn't reach one without external feeding. Reactor control reduces to accelerator control: a volume knob instead of a sinking ship's helm. Coupled with a 10 MW beam, Rubbia's theoretical "amplifier" would produce about 1500 MW thermal — roughly 600 MW electric, of which only ~30 MW would be consumed by the accelerator itself. Twenty-fold amplification, and here's where the eternal trap begins, to which we'll return.
A separate delight — fuel. You can load into the blanket not just uranium, but thorium: cheap, abundant, not breeding plutonium. Or you can load what's a burden for conventional energy: minor actinides, neptunium, americium, curium, that very "tail" of spent fuel that's radiotoxic for hundreds of thousands of years. In a subcritical machine they can be burned and converted into short-lived isotopes — an occupation impossible in a classical reactor precisely because these isotopes have a poor fission-to-capture ratio, chain reaction can't be maintained on them. Waste transmutation is not a side effect but the second face of the same technology.
On November 24, 1993, Carlo Rubbia — 1984 Nobel laureate for discovering W and Z bosons, who had just left the position of CERN director general which he held since 1989 — stepped before the press and declared he had found a solution to all nuclear energy problems at once: safety, waste, fuel supply (physicist André Gsponer's recollections, La Gazette Nucléaire, 2003). That same month CERN registered the preprint "An energy amplifier for cleaner and inexhaustible nuclear energy production driven by a particle beam accelerator." The title spoke for itself: an energy amplifier — a device that takes a megawatt from the socket and returns hundreds.
In fairness, Gsponer's snark wasn't unjustified: "Rubbia reinvented warm water." Hybrid subcritical schemes had been discussed since the early years of reactor construction; Rubbia's contribution wasn't the formula but the packaging. He linked subcriticality with a specific neutron source — spallation on lead, with thorium cycle instead of uranium, and with the promise to burn others' waste. The result was a product with a ready press release: "all the pluses of nuclear, minimum minuses." True, at a parliamentary hearing in France in November 1995, held openly and chaired by Claude Birraux, Rubbia himself, to the hall's surprise, named the most promising applications of the amplifier as... military — plutonium and tritium production. Universal marketing has its skeletons.
But the main thing that made the idea beautiful precisely then: it didn't require belief in the unprecedented. Accelerators already existed, spallation was measured, subcritical assemblies were described. In 1995 under Rubbia's direction CERN ran the first experiment (FEAT) on thorium: a proton beam hit a lead block with a thorium blanket, and the measured neutron multiplication confirmed calculations. It was followed by a 164-page conceptual design of an industrial machine (CERN-AT-95-44). On paper by 1996 stood a technology with verified physics and only unclear engineering.
Then politics begins, and it's swift. June–July 1996: Rubbia and his Spanish colleague Juan Antonio Rubio circle Aragon — meetings with industrialists, reception with regional president Lanzuela, audience with Prime Minister Aznar in Madrid. In an El País interview (July 3, 1996) Rubbia lays out a proposition that sounds almost incredible today: "The cost of the full project — about three hundred million dollars, less than a satellite or military aircraft." A ton of plutonium, he says, equals three million tons of coal — and Spain has already accumulated twenty tons of this "unwanted" substance. A prototype in Zaragoza, four years construction: "If we start in 1998, by 2003 the machine will be out of childhood." New Scientist in April 1997 records a deal with 60 companies. The feeling of a startup taking off.
Then — 1997, creation of company LAESA (Laboratorio del Amplificador de Energía) with participation of the regional development institute, promise of a 100–250 MW prototype by 2002, publication of a separate CERN report with subtitle "The case of Spain." And then — silence. LAESA went bankrupt by the end of 2001, which the Spanish press reported almost in a whisper; contacts with Rubbia, according to Gsponer, broke off two years before the formal bankruptcy. European enthusiasm smeared across working groups and died in reports. The idea turned out to be a display window without a buyer.
Why? The naive answer "Europe fears nuclear" doesn't hold up — in France and Finland in those same years they were chewing through their reactor programs. The real reason is more prosaic and honest: in the amplifier chain there wasn't a single ready link at megawatt scale. An accelerator delivering 10 MW continuous beam didn't exist anywhere in the world (and the most powerful continuous machine is still about 2 MW). A liquid metal target surviving years under beam, also none. And the economics — the one with k on a fork of 0.95–0.98 — bit from both ends: push k toward one, and the danger stops differing from an ordinary reactor; drop lower — amplification gets eaten by the hungry accelerator. Gsponer formulated this lethally simply back in 1993: at k > 0.98 the amplifier isn't needed — danger is the same as an ordinary reactor, at k < 0.95 the amplifier doesn't pay off. What remained was a band of a few percent on which the installation had to be held for ten years of operation. Nobody dared.
The technical essence of the problem fits in one detail the size of a plate. The waveguide through which protons fly is deep vacuum. The target is molten lead-bismuth alloy at several hundred degrees. Between them must stand a window: a thin membrane, usually tungsten or steel, separating vacuum from liquid metal and passing the beam. And this membrane lives in hell: protons of a megawatt beam knock atoms out of it (radiation swelling), thermal shocks from beam pulsations work on fatigue, liquid metal corrosion erodes the surface, temperature jumps.
Calculations by Jacques Maillard and his graduate student Fabienne Bacher predicted: the window could fail in hours. The check happened on July 5, 2002: in a laboratory experiment the membrane under irradiation at full power collapsed after 36 hours. This doesn't mean the problem is unsolvable — it means it turned out to be that very long splinter because of which entire national programs disappear.
Humanity did have one honest test at megawatt level: MEGAPIE — a consortium of nine institutes (from CEA and PSI to JAERI and LANL), which built a liquid Pb-Bi target at 1 MW for the Swiss source SINQ. From August 14 to December 21, 2006 it honestly ran the beam — four months, with electromagnetic pumps, heat exchanger and shutdown procedures. Four months versus promised "years without replacement" — an order of magnitude gap, and that's at half the power. The Chinese know this better than us: their thermohydraulic target prototype for CiADS by August 2025 had run only 340 hours continuously, and that's already an achievement, officially reported as the country's first specimen. Tellingly, in parallel IMP is testing an alternative — a windowless target altogether: gravity-driven dense granular target, where instead of a liquid bath — a stream of dense granules under gravity. Bypass the splinter rather than pull it out.
Now to concrete numbers of what's growing in Huizhou — and it looks less futuristic than the headlines. CiADS is a facility of national scientific scale, not a power plant. Approved in July 2021, three main components: a 350-meter superconducting proton linac accelerating protons to 250 MeV up to 2.5 MW beam power (space laid for upgrade to 1.5 GeV); a liquid metal Pb-Bi spallation target at the same 2.5 MW; and a subcritical reactor of thermal power ~7.5 MW. Total about 10 MW thermal — by energy standards a crumb, on the order of one percent of thermal analogs of modern light-water units. This is a research stand where for the first time in the world the "accelerator + target + reactor" chain will be run together, at full power, for more than 24 hours straight.
Progress is real, but by no means "built in a year." As of August 2025 (Institute of Modern Physics report, IMP CAS) the entire room-temperature front-end of the linac is ready: ion source with closed-loop current stabilization, RFQ resonator with quality factor above 13,000, thermohydraulic target prototype after 340 hours of continuous work accepted by commission. The reactor nuclear island passed overall design, non-nuclear integral stands are being built, China's Ministry of Ecology approved the first phase of environmental assessment. By February 2026 (Science and Technology Daily report from the site) installation went underground — a tunnel at 13 meters depth where installers move, in their own words, "like in slow motion" to avoid raising a speck in the clean zone. Annual plan-2026: mount all 32 superconducting accelerating sections in six months, with a peak of eight parallel assemblies. The reporter isn't far from the truth when calling this "a transition from single product to industrial standard."
An important key to understanding the Chinese approach lies in two numbers from the recent past. In 2011 IMP launched the CAFe program — a small prototype linac only 38 meters long with one goal: bring continuous beam current to 10 mA. This took ten years and in February 2021 ended with a world record: 10 mA continuously, 205 kW, 20 MeV, with machine availability 93–96% on long runs (CERN Courier, April 2021). Ten years on "merely" reliability. It's precisely this piece — not flash but stubborn dependability — that 1990s Europe lacked to believe in the amplifier.
An honest analysis must end not with a forecast but with a list of thin spots. First: the claimed effects — "uranium utilization a hundred times higher, waste lifetime a thousand times shorter" — are program goals, not achieved results; even optimistic OECD/NEA estimates (2002) required multiple fuel reloading and commitment to maintain transmutation industry for a hundred years to get an order-of-magnitude reduction in radiotoxicity. Second: a full "accelerator + reactor" chain hasn't worked in the world for a single day — there were only laboratory stands: Japanese KUCA since 2009 and Chinese zero-power Venus II and Venus III (2017 and 2019), and even they don't approach megawatt class. Third: the target. MEGAPIE gave four months at 1 MW; CiADS must hold 2.5 MW for years — and even their first phase beam permit is modest, 500 MeV and only 25 kW, clearly with aim "first feel it out, then push." Fourth: economics. The amplifier pays off only with huge volumes of burned actinides, and for that you need an entire reprocessing (partitioning) industry — nobody has it today. And fifth, most honest: MYRRHA in Belgium — the European twin project, with €1.6 billion budget and state support since 2018 — stopped in May 2026: the federal government froze signing the accelerator building contract for "nuclear strategy reassessment." While one half of the story builds, the second — hesitates again. Thirty years later the dispute isn't about physics but about whether the government believes in its own long-term commitment.