How to Read China’s Quantum Supply Chain

Somebody will eventually hand you a quantum vendor claim and ask whether it holds. It might be a supplier questionnaire, a due-diligence pack, a slide in a partner briefing, or a line in a research report your board has already read. The claim will usually take one of three shapes: this system is domestically produced, this component is no longer available from that country, or this capability is now sanction-proof.

All three are supply chain questions rather than physics questions, and the method for answering them is closer to procurement analysis than to quantum mechanics. What follows is that method, worked through the hardest available case: China’s quantum hardware supply chain between 2021 and 2026, where the claims are loud, the verification is thin, and the stakes for anyone doing vendor assessment are real.

We are using China because it is the most instructive example, not because it is the only supply chain that needs reading. The same questions apply to a European photonics supplier or an American cryogenics vendor. The difference is that in the China case, both the boosters and the skeptics have written a great deal, which gives us plenty of claims to test.

The component nobody thought China could build

Start with the case that made the argument famous.

A dilution refrigerator, usually shortened to DR, is the tall gold chandelier you see in every photograph of a superconducting quantum computer. Its job is to cool the processor to a few thousandths of a degree above absolute zero, a range measured in millikelvin (mK). Superconducting qubits only behave like qubits down there. Above that range, thermal noise swamps the quantum states you are trying to compute with.

For most of the last two decades, three companies built almost all of them: Bluefors in Finland, Oxford Instruments in the UK, and Leiden Cryogenics in the Netherlands. That concentration is what made the component look like a chokepoint. Restrict three suppliers in three allied countries, the reasoning went, and you constrain an entire national program in superconducting quantum computing.

The reasoning was correct about the dependency. It was wrong about how long the dependency would last.

In 2021, China had no domestic dilution refrigerator manufacturer. By 2025, six had announced or delivered systems. QuantumCTek’s ez-Q Fridge reached a base temperature near 10 mK on the company’s published specification, with deliveries starting in mid-2023. Origin Quantum’s SL1000, built in Hefei from June 2024, is specified by the company for systems above 100 qubits and takes 840 coaxial cables. CSSC Pengli in Nanjing came at it from the other direction, converting cryogenic capability originally built for MRI machines. CETC showed an XS1000 at the November 2025 Hefei conference. Hefei Zhileng, founded in 2023, published a 7.45 mK base temperature. A 2024 spinout from the Chinese Academy of Sciences Institute of Physics, CASQI, reported a system reaching 6.6 mK.

That is roughly the time it takes to complete a doctorate, and it went from nothing to a domestic industry.

China built them, and the more useful detail for an assessor is that the export controls telegraphed themselves for six years before they arrived. Quantum was flagged as an emerging technology under the Export Control Reform Act of 2018. Comprehensive Commerce Control List additions did not land until September 2024. Chinese cryogenics firms started their quantum refrigeration projects in 2018, in public, in response to signals that were also public.

When you assess whether a restriction will bite, the question is never only whether the target can build the thing. It is how much warning they had, and whether the underlying industrial base was adjacent to something they already built. Cryogenics for MRI is adjacent to cryogenics for qubits. That adjacency is the whole story of CSSC Pengli.

Read the spec sheet the way an engineer would

You have a Chinese DR spec sheet and a Bluefors spec sheet in front of you. What actually distinguishes them?

Two headline numbers dominate the marketing. Base temperature is how cold the system gets with nothing in it. Cooling power at 100 mK is how much heat the system can remove while sitting at that temperature, quoted in microwatts. The second number is the one that matters for scaling, because every coaxial cable running down into the fridge carries heat with it, and a machine with hundreds of qubits needs hundreds of cables.

On both numbers, the Chinese systems now sit in the same range as the Western ones. Cooling power at 100 mK, on published manufacturer specifications, runs from around 450 µW on the ez-Q, through 1,000 µW on the Origin SL1000, to a reported 2,000 µW on the CASQI system. Bluefors guarantees more than 1,000 µW on its XLD1000sl; the Oxford Instruments Proteox is specified above 850 µW.

So the numbers converge. And the numbers are not the assessment.

Four things a spec sheet does not tell you, all of which decide whether a machine is usable:

Vibration. Mechanical vibration destroys qubit coherence, and isolating a cryocooler from the sample stage is decades of accumulated engineering. We are not aware of published vibration performance data from any Chinese manufacturer. That’s an absence of evidence rather than evidence of a problem, but for a procurement decision, an absence of evidence in a load-bearing area is itself a finding.

Loaded versus unloaded performance. Most published base temperatures describe an empty fridge. What matters is the temperature under real experimental heat load, with hundreds of cables installed and qubits running. The gap between the two can be large, and it’s the gap where reputations are made.

Reliability over years, not months. A quantum error correction experiment may run continuously for weeks. Western DRs have decade-long operating records across hundreds of installations. The Chinese systems have two or three years. There’s no shortcut through that; the data accumulates at the speed of calendar time.

Software and automation. Remote monitoring, automated cooldown sequences, fault handling. It’s unglamorous, and it determines whether a system needs a specialist physicist babysitting it or can be run by a facility engineer.

The generalizable rule: when two spec sheets converge on the headline metrics, the systems differ in whatever the spec sheet omits. Ask what is not printed. For cryogenics that means vibration, load, uptime, and software. For a laser it means linewidth drift over months. For a control system it means calibration time. The specific list changes; the move does not.

Components within the components

The most common failure in supply chain assessment is stopping one level too early. A system can be assembled domestically out of parts that are not.

Three dependencies sit underneath the Chinese dilution refrigerator industry, and each behaves differently.

Pulse tube cryocoolers. Every modern “dry” DR uses a pulse tube cryocooler to do the first stage of cooling, taking the system down to around 4 kelvin before the dilution cycle takes over below 1 K. The global supply is concentrated in Cryomech, an American firm now owned by Bluefors, and Sumitomo in Japan. Chinese producers make cryocoolers for general industrial use; the sub-4 K performance a DR needs is a different product. Six companies assembling complete refrigerators does not resolve a shared dependency on the stage that sits underneath all six.

Parametric amplifiers. Reading out many qubits at once requires amplification with almost no added noise, which in practice means a traveling wave parametric amplifier, or TWPA. Chinese groups have published research-grade Josephson parametric amplifiers. A commercial domestic TWPA supplier doesn’t appear to exist. Global supply runs through a handful of firms, mostly in the Netherlands and Sweden, plus academic pipelines.

Helium-3. This one is different in kind. The dilution cooling cycle works by mixing two helium isotopes, and helium-3 is the scarce one. It is produced almost entirely as a byproduct of tritium decay in nuclear weapons programs, which caps global supply at a level set by other countries’ weapons stockpiles rather than by market demand. China has essentially no domestic production.

Distinguish those three carefully, because they call for different responses. Pulse tube cryocoolers are an engineering problem: hard, expensive, and solvable by a determined industrial program on a multi-year timeline. TWPAs are a commercialization problem: the research exists, the company doesn’t yet. Helium-3 is a physics and geopolitics problem, and no amount of factory construction fixes it. The only real escape is a different cooling technology, which is why Chinese groups are publishing on adiabatic demagnetization refrigeration using rare-earth alloys. Those remain laboratory demonstrations.

When you assess any supply chain claim, sort the dependencies into those three buckets. Engineering problems fall to money and time. Commercialization problems fall to a single successful company. Physics problems do not fall at all.

What an 80% localization claim means

Origin Quantum states that its Wukong quantum computer is 80% localized. The figure is repeatedly cited, including by analysts at the Center for Strategic and International Studies (CSIS) quoting the company’s chief scientist, and it is almost certainly accurate as a count of components.

The 80% appears to cover the processor chips, the Tianji control electronics, the Origin Pilot operating system, the QPanda programming framework, the SL-series refrigerator, microwave interconnect modules, and cryogenic thermometry. That is a genuinely impressive list, and dismissing it would be as much an error as accepting it.

The remaining 20% is what the assessment has to examine. It appears to include the pulse tube cryocoolers, helium-3, parametric amplifiers, high-end test instrumentation, some specialized cryogenic cabling, and precision fabrication equipment, including lithography tools imported from Germany.

Read those two lists next to each other, and the missing fifth contains the items without which the other 80% will not switch on.

This generalizes directly to vendor questionnaires. A localization or country-of-origin percentage counts parts; it does not weight them. Two questions convert the number into information. Which specific components sit in the unlocalized remainder? And does the system function without them? A supplier who can answer both is telling you something. A supplier who offers only the percentage is telling you they counted.

Where China is genuinely ahead

An assessment that only finds weaknesses is not an assessment; it is a prior with citations. Several Chinese quantum components are competitive, and at least one category is world-leading.

Narrow-linewidth lasers are the clearest case. CSIS’s January 2026 report puts Shanghai Precilaser’s exports to Harvard at roughly a hundred systems. When MIT applied for duty-free import of Precilaser equipment, US federal records noted that no instrument of the same general category is manufactured in the United States. The same CSIS analysts put Chinese laser pricing at around a third of German, Swiss, and American competitors. Western laboratories buy from Precilaser because the alternative costs three times as much and, on the MIT filing, sometimes does not exist domestically at all.

Control electronics are strong. Origin Quantum says its Tianji 4.0, launched in May 2025, supports more than 500 qubits and integrates RF generation, digital control, DC sources, and synchronization in one stack, with automated calibration intended to let a general engineer operate the system.

The Shanghai Institute of Microsystem has published photon-number-resolving single-photon detection at high rates.

High-density microwave connectivity modules were localized by Origin Quantum in May 2024, replacing a part previously sourced from Japan. The announcement came days after a major Entity List expansion, which is either coincidence or the fastest press release in the industry.

The procurement consequence is direct and often overlooked. If you’re assessing supply chain concentration for a Western quantum program, the lasers point the dependency in the opposite direction from the one everyone discusses. Concentration risk is a property of components, and the country of assembly is a poor proxy for it.

The controls, and what they did

For readers who need the regulatory picture rather than the component picture, the outline is short.

The Entity List is the US mechanism that blocks unlicensed supply to a named organization. The first quantum-specific additions came in November 2021, covering three entities. May 2024 brought 22 more in a single action, reaching across university laboratories, national institutes, and commercial firms. March 2025 moved upstream to seven supply chain entities, including cryogenics manufacturers and, for the first time, scientific equipment distributors.

The Bureau of Industry and Security (BIS) set the technical thresholds in its September 2024 Interim Final Rule. The published criteria include quantum computers at or above 34 physical qubits with associated gate error conditions, cryogenic cooling systems delivering 600 µW or more at or below 100 mK sustained beyond 48 hours, two-stage pulse tube cryocoolers rated below 4 K, and parametric signal amplifiers. Allied jurisdictions moved in parallel through 2024 and 2025, and a separate outbound investment regime took effect in January 2025, restricting US investment in Chinese quantum computing, military quantum sensing, and quantum communications.

Two observations for anyone building a compliance position. The thresholds are specific numbers, which means product designs can sit near them; the ez-Q’s 450 µW rating sits below the 600 µW control line, and we would not draw a conclusion from a single data point about whether that was designed or incidental. And the controls now reach distributors, which means an intermediary in your supply chain can be the restricted party even when neither the manufacturer nor the end user is.

Does the semiconductor parallel hold?

The standard argument runs: Huawei was cut off in 2019, written off, and shipped a 7 nm-class phone chip in 2023. DeepSeek was constrained to export-limited hardware and trained a frontier-class model anyway. Restrictions produce the capability they were meant to prevent. Quantum next.

The mechanism is real. Restrictions solved the Chinese suppliers’ hardest commercial problem, which was persuading domestic customers to buy an unproven local product instead of an established Western one. Once the Western option disappears, the local product wins by default and gets the volume it needs to improve.

But the analogy needs two corrections before you rely on it.

Quantum is easier to substitute than semiconductors. There is no equivalent of EUV lithography: no single tool, from a single supplier, that gates everything downstream. The supply chain is shallow and dispersed. Capital requirements per facility are far lower. Global volumes are measured in hundreds of systems rather than trillions of chips, so a factory serving a national program does not need to be enormous.

And quantum is harder in ways semiconductors are not. Helium-3 is a physics constraint with no engineering path around it. The technology sits at the research frontier, so there is no established recipe to copy, only papers to interpret. Semiconductors had a legacy market of older nodes that generated revenue while capability caught up. There is no commercially useful older, simpler quantum computer to sell in the meantime.

The shallowness favors the challenger. Less accumulated know-how to replicate also means less accumulated know-how protecting the incumbent, so the catch-up window in quantum is shorter than the semiconductor precedent suggests, and a sourcing plan that assumes a decade of Western hardware advantage is planning on the wrong number.

The photonics hedge

Watch what a constrained program invests in, because the investment pattern reveals the strategy more reliably than the announcements.

Chinese investment in photonic quantum computing has accelerated sharply since 2024. QBoson announced groundbreaking on a photonic manufacturing facility in August 2025 and a large Series B in April 2026 led by state-linked investors. Guizhen Silicon Quantum, a spinout from the University of Science and Technology of China (USTC), released a programmable optical quantum computer in November 2025 built on a CMOS-compatible silicon photonics process. TuringQ has raised substantial funding for integrated photonic chips.

The strategic appeal is obvious once you’ve read the cryogenics section. Photons work at room temperature. No dilution refrigerator, no pulse tube cryocooler, no helium-3. The manufacturing can in principle run through existing semiconductor foundries.

Two cautions belong in any assessment of these claims.

First, distinguish machine types. QBoson’s systems are coherent Ising machines: special-purpose optimization hardware, not general-purpose gate-based quantum computers. Boson sampling experiments are likewise demonstrations of a specific sampling problem. Both are legitimate physics. Neither is a machine that runs arbitrary quantum algorithms, and “qubit” in a headline can obscure the difference.

Second, photonic claims have a specific quantum-washing failure mode. At least one widely reported Chinese “photonic quantum chip” turned out to be an interferometer mesh operating with bright classical light, with no single-photon sources, no entanglement, and no quantum gates. The due-diligence questions are concrete: what is the photon source, is entanglement demonstrated and measured, and what gate operations are implemented?

Was the photonics push a response to export controls? Probably not in origin. USTC’s photonics work predates the controls by years. The commercial acceleration after 2024 looks like hedging: fund domestic cryogenics and fund a technology path that needs no cryogenics at all, and see which one arrives.

The counter-pressure

Supply chain pressure runs in both directions, and the return fire lands squarely on quantum.

China’s April 2025 controls restricted seven heavy rare earth elements under case-by-case licensing. An October 2025 escalation added five more, including holmium, erbium, thulium, europium, and ytterbium, and introduced extraterritorial reach for the first time. A partial one-year suspension of the October measures followed diplomatic talks in late 2025; the April measures were not suspended.

Those specific elements are not an arbitrary selection. Ytterbium is the working ion in leading trapped-ion quantum computers. Europium is the standard material for solid-state quantum memory, with published spin coherence lifetimes measured in hours. Erbium emits at 1,550 nm, the telecom band, which makes it central to quantum networking over existing fiber. Gadolinium gallium garnet is a key material for adiabatic demagnetization refrigeration, which is one of the candidate routes around helium-3. China processes the large majority of the world’s rare earth supply.

For most industries, rare earth coercion is a blunt instrument: it raises prices, funds alternative supply, and loses potency with each use. Quantum is the unusual case where it bites harder. The volumes required are tiny and the purity requirements are extreme, which makes building alternative supply slow and commercially unattractive. Nobody opens a separation facility to serve a market measured in kilograms.

What the components do not tell you

Everything above concerns hardware supply. The competitive question is different, and it is worth ending on the distinction because it is the one most easily lost.

Building a dilution refrigerator is necessary and not sufficient. What determines whether a quantum computer is useful is error correction: whether many noisy physical qubits can be combined into a smaller number of reliable logical qubits that stay coherent long enough to finish a computation. That is the benchmark the field has converged on, and it is where the comparison currently favors Western systems. Google demonstrated below-threshold error correction on its Willow processor in December 2024. Chinese groups have run surface-code error correction and have not publicly matched that particular milestone.

Which is why raw qubit counts deserve suspicion in either direction. A 500-qubit processor with poor gate fidelity is worth less than a smaller machine with demonstrated logical qubits. Physical qubit counts are a manufacturing statistic. Logical qubits are a capability statistic. Any vendor comparison that leads with the first and omits the second is incomplete, wherever the vendor is headquartered.

There is also a verification gap that applies to the whole picture. The US-China Economic and Security Review Commission noted in November 2025 that Chinese quantum results often lack independent verification, blurring genuine progress and political signalling. CSIS made a similar point in January 2026: the systems are reported to rival international competitors, and third-party verification has not been done. Treat that as a statement about available evidence rather than a verdict about capability. Absent verification, the honest position is a wider confidence interval, not a lower estimate.

The questions worth asking

Compressed into a form that survives contact with a supplier questionnaire:

1. What is the dependency underneath the dependency? Domestic assembly of a system that relies on one foreign sub-component has moved the risk, not removed it.
2. Which bucket is each constraint in? Engineering problems fall to money and time. Commercialization problems fall to one successful company. Physics problems do not fall.
3. What does the spec sheet omit? When headline numbers converge, the systems differ in reliability, tolerance to real operating conditions, and software.
4. What does the localization percentage count, and what does it weight? Ask which components sit in the remainder and whether the system runs without them.
5. Is the machine general-purpose or special-purpose? Optimization hardware and sampling demonstrations are not gate-based quantum computers.
6. Has anyone independent verified this? No verification means a wider uncertainty band, in both directions.
7. Which direction does the concentration actually run? For lasers, the answer may not be the one the headlines suggest.

None of those questions requires you to evaluate a quantum algorithm. They require you to read a supply chain claim the way you would read any other engineering procurement claim: by asking what sits underneath it, what the numbers omit, and who has checked.

Building the assessment skill

The pattern in this article, distinguishing announced from demonstrated, spec sheet from operating performance, physical qubit count from logical capability, is the core analytical skill for anyone whose job touches quantum procurement, vendor risk, or technology strategy. It is a structured skepticism that happens to need enough physics to know which questions land.

Quantum Academy’s certification programs are built around exactly that skill set, for professionals who have to make defensible assessments rather than run experiments. You can review the full portfolio and choose the program that matches your role at quantumacademy.com/.

For deeper technical background on the hardware and error correction questions raised here, PostQuantum.com carries longer-form analysis. If your interest is the migration and inventory side of quantum risk rather than the hardware side, pqcframework.org covers that methodology.