Marin Ivezic
In 2025, Bluefors agreed to buy helium-3 from Interlune, a Seattle company that intends to extract the isotope from lunar soil and has not yet flown a mission. The volume and term the two companies announced are up to 10,000 litres a year, from 2028 through 2037.
Bluefors, based in Finland, builds the refrigerators that cool most of the world’s superconducting quantum computers. Helium-3 is what makes that cooling work, and there is almost none of it on Earth. So the company with the clearest forward view of demand for these machines looked at terrestrial supply, looked at its own order book, and went shopping off-planet.
Read as a demand signal rather than as a space story, the contract says something specific and testable. Bluefors expects to build enough refrigerators over the next decade that current helium-3 production won’t cover them.
Most attention in quantum computing goes to the machine builders and to the contest between modalities, the competing physical approaches to building a qubit. That contest is real and genuinely unresolved. For anyone allocating capital or planning a procurement, though, it’s the second question rather than the first. The first question is which parts of a superconducting machine come from a handful of firms, and which of those firms would still have a business if superconducting lost.
This is an analysis of technology and market structure. It isn’t investment advice.
What Is Actually in the Chandelier
The gold-tiered structure in every quantum computing photograph is a dilution refrigerator with its vacuum cans removed. The quantum processor is the small chip at the very bottom. Everything above it exists to keep that chip cold, quiet, and connected to the outside world.
A working superconducting system needs seven things.
A quantum processing unit (QPU). A chip patterned with superconducting circuits on a silicon or sapphire substrate. Each qubit is built around a Josephson junction, which is two superconductors separated by an insulating barrier a few atoms thick.
A dilution refrigerator. The cooling system that brings the chip to roughly 10–15 millikelvin, or thousandths of a degree above absolute zero. That is colder than deep space. It gets there by mixing two isotopes of helium and exploiting the way they separate at low temperature.
Cryogenic wiring and microwave components. Hundreds of coaxial cables running from room temperature down to the coldest stage, with attenuators, filters, isolators, circulators, and amplifiers at every level.
Control electronics. Hardware that generates precisely shaped microwave pulses, typically in the 4–8 GHz band, to manipulate each qubit and read its state back out.
Classical computing. Calibration, scheduling, and the error-correction decoding that has to keep pace with the machine in real time.
Software. Compilers, error mitigation, and the layer that turns an algorithm into a pulse sequence.
The building. Reinforced floors for cryostats that weigh several hundred kilograms, chilled water, power conditioning, electromagnetic shielding, vibration isolation, and helium handling.
Every one of those layers has its own suppliers, and they behave very differently from one another.
Three Questions Before You Look at Any Layer
We teach this screen before we teach the map, because the map changes every quarter and the screen doesn’t. Three questions, applied to each layer in turn.
How many suppliers can actually deliver at volume? Not how many exist. How many could ship twenty systems next year without breaking. In several of these layers, the honest answer is one or two.
What happens to this layer if superconducting stalls? Some components are specific to superconducting qubits and would be stranded by a shift to neutral atoms or photonics. Others get bought by every modality, and a few get bought by markets that have nothing to do with quantum computing at all.
Is this layer a market, or is it somebody’s internal department? IBM fabricates its own chips and builds its own control stack. That’s a supply chain, but it isn’t a market. You can’t buy into it except by buying IBM. A layer only becomes investable when somebody sells it to third parties.
If you take one thing from this article, take the second question. Concentration is what people notice first, and it’s the least durable of the three.
Cryogenics: The Hardest Chokepoint
No layer is more concentrated than the dilution refrigerator. Every superconducting quantum computer running today depends on one, and research from ICV Tank puts Bluefors and Oxford Instruments together above 70% of the market for quantum applications.
Bluefors is the leader by a distance. Founded in 2008, it moved early to adapt its systems for qubit researchers rather than for general low-temperature physics, and the company reports having delivered more than 1,500 dilution refrigerators and more than 15,000 cryocoolers. Its KIDE platform is the backbone of IBM’s Quantum System Two. Google’s Willow chip runs in a Bluefors system. The company also bought Cryomech, a cryocooler manufacturer in Syracuse, New York, which secured an upstream component and gave it a North American manufacturing base at the same time.
Oxford Instruments, in the United Kingdom, is the second pillar, with decades of cryogenic instrumentation behind it. Leiden Cryogenics in the Netherlands, CryoConcept in France, and FormFactor each hold a niche. Then there are the challengers. Maybell Quantum, in Denver, is building compact systems aimed at higher qubit density per unit of floor space. Kiutra, in Munich, has taken a different route entirely, using magnetic cooling in solid-state materials to reach low temperatures without helium-3 at all, and raised money in late 2025 on exactly that supply-resilience argument.
This layer keeps getting compared to ASML, and the comparison holds partway. ASML is the standard example of a single supplier gating an entire industry: no extreme ultraviolet lithography machine, no leading-edge chip, and the whole semiconductor roadmap runs at the pace one Dutch company can build. Dilution refrigerators occupy a structurally similar position in superconducting quantum computing. The difference is the moat. ASML’s position rests on twenty years of physics, thousands of patents, and a supplier network nobody can reconstruct. A dilution refrigerator is a difficult, precise, deeply experience-dependent piece of engineering, and it is not that. The barrier is manufacturing capacity and accumulated know-how rather than an unreproducible IP position, which is precisely why more than ten Chinese manufacturers appeared within a few years once the demand was visible, as Xinhua reported in November 2025.
That distinction changes what you’re underwriting. A supplier whose advantage is capacity and experience earns very well while demand outruns supply and gets competed down afterwards. A supplier whose advantage is unreproducible physics doesn’t.
A niche instrumentation business, small enough that two firms supply most of it, is gating a field that has absorbed tens of billions in public and private funding. That is an unstable arrangement. Either the market grows by an order of magnitude, or capacity becomes the industry’s schedule.
Helium-3: A Constraint, Not a Cycle
The two helium isotopes are easy to conflate and they aren’t the same product.
Helium-4 is the ordinary stuff. It’s extracted industrially as a byproduct of natural gas production, with Qatar and the United States dominating supply, and it goes into MRI scanners, semiconductor fabs, and rocket propellant purging as well as cryogenics. It’s tight and expensive, and quantum computing competes for it with much larger buyers.
Helium-3 is a different substance with a different supply chain. It barely occurs naturally on Earth. The main terrestrial source is the radioactive decay of tritium, which is itself a product of nuclear weapons programmes, and world supply is measured in kilograms rather than tonnes, set in effect by decisions taken decades ago in nuclear stockpile management. Prices run into thousands of dollars per litre. Each dilution refrigerator charges a few dozen litres and holds it in a closed circuit, so this is a stock problem rather than a consumption problem, but the stock has to be found before the machine can be built.
Three responses are underway. Interlune is pursuing lunar extraction, on the basis that solar wind has been depositing helium-3 in the top few metres of regolith for billions of years. Kiutra is removing the requirement. Bluefors and others are building recovery systems that capture and recirculate rather than vent.
For the screen, helium-3 scores unusually. Supplier count is close to one, in the sense that a national isotope programme is not a competitive market. Modality exposure is high, since this constraint follows superconducting and not much else. And it is barely a market at all, which is why the interesting positions are the companies removing the dependency rather than the ones supplying it.
The Processor: A Narrow Materials Palette
Superconducting qubits use surprisingly few materials, at extraordinary purity. Most designs use aluminium, which becomes superconducting below a critical temperature of about 1.2 kelvin, deposited on high-resistivity silicon or sapphire. The Josephson junction is usually an aluminium oxide barrier grown between two aluminium layers by controlled oxidation. Systems needing more complex multi-layer wiring, including D-Wave’s annealers, use niobium, which superconducts below 9.2 kelvin. Tantalum has drawn attention for extending coherence times, and work at Argonne and elsewhere has revived niobium for gate-based designs.
The NATO Transatlantic Quantum Community study of May 2025 flagged niobium, tantalum, titanium, lithium niobate, silicon-28, and helium-3 as materials with exposure to unstable or non-allied sources. Niobium is the sharpest of those. The United States Geological Survey puts Brazil at close to 90% of world production, and niobium goes into both processors and the superconducting cables inside the cryostat.
Fabrication borrows tools from semiconductors and then diverges. The highest-coherence qubits come out of comparatively simple processes, often a single metal layer rather than the dozen-plus stack in a logic chip, because the constraint is surface defects rather than integration density. The standard junction technique, double-angle evaporation, needs a tilting evaporator and doesn’t fit modern CMOS lines cleanly, although IMEC in Belgium has shown compatible alternatives approaching state-of-the-art performance.
Most hardware companies fabricate in-house. IBM, Google, and Rigetti all run their own lines, and IQM in Finland and Oxford Quantum Circuits in the United Kingdom maintain their own capability. That keeps the layer internal and therefore unbuyable.
Two companies are trying to externalise it. QuantWare, in Delft, sells superconducting processors as components to anyone building a machine and is constructing an industrial-scale fabrication facility to do it at volume. That’s the Intel model applied to quantum: one firm makes the processor, others assemble the computer. SEEQC, spun out of HYPRES and operating in New York, London, and Naples, runs a multi-layer superconductor foundry and builds single flux quantum (SFQ) chips, a form of superconducting digital logic that can run inside the refrigerator alongside the qubits.
Control Electronics: The Layer That Travels
If the refrigerator is the body, the control electronics are the nervous system. Each qubit needs shaped microwave pulses for gate operations and a separate microwave measurement for readout, and the demands scale worse than linearly with qubit count. More channels, tighter synchronisation between them, and, for error correction, a feedback loop that has to close in microseconds.
The suppliers here are mostly European and mostly specialists. Zurich Instruments, now part of Rohde & Schwarz, leads on high-fidelity control and readout. Qblox, in the Netherlands, supplies the modular stacks that most open-architecture builds use and was selected by the Department of Energy and Fermilab to manufacture and distribute the QICK (Quantum Instrumentation Control Kit) control platform in the United States. Quantum Machines, in Israel, has positioned itself as an infrastructure company rather than an instrument vendor, and its orchestration layer runs Israel’s national quantum computing centre. Keysight Technologies is the largest publicly traded firm with meaningful exposure here, and its control system was selected for the Fujitsu and RIKEN 256-qubit machine. Tabor Electronics and Swabian Instruments hold further niches.
On the screen, this layer scores best of any in the stack. Supplier count is moderate rather than critical. It’s plainly a market, with several vendors selling to everyone. And modality exposure is low, because superconducting qubits, silicon spin qubits, and parts of the trapped-ion and neutral-atom stacks all need precise microwave and radio-frequency control. A firm here is selling into the whole field rather than into one bet.
The disruption to watch is internal to the layer. If cryogenic control chips like SEEQC’s SFQ approach mature, a large share of the room-temperature rack moves inside the refrigerator, which rewrites both this layer and the wiring layer below it.
Cold Wiring: The Least Glamorous Constraint
The cables are the part nobody photographs deliberately, and they may be the part that sets the schedule.
Every qubit needs several lines. A drive line carries the pulses that perform gates. A flux line tunes the qubit’s frequency. Readout lines carry the measurement signal back up. An analysis in EPJ Quantum Technology put a 50-qubit system at roughly 124 radio-frequency lines once pump lines for the amplifiers are included. Each line runs through every temperature stage of the refrigerator, with attenuators along the way to strip thermal noise, and each one carries heat downward into a cold space with a cooling budget measured in microwatts.
That is the wall. Triple the cable count and you triple the heat load, and at some point the refrigerator simply cannot hold temperature. Higher-density interconnects, vertical routing through the chip package, and moving control logic into the cold are all attempts to get past it.
The suppliers are small and specialised. Delft Circuits builds flexible cabling designed for high-density qubit wiring. Radiall, in France, has developed non-magnetic cryogenic connectors and anchors the QRYOLink consortium, an eight-million-euro programme aimed at cabling for systems far larger than anything running now. Rosenberger, in Germany, makes multichannel cryogenic connectors and assemblies in stainless steel, cupronickel, niobium-titanium, and beryllium-copper. On amplification, traveling wave parametric amplifiers (TWPAs) sit at the coldest stage and boost readout signals while adding close to the minimum noise physics allows, and they remain closer to research components than to catalogue parts. Low Noise Factory, in Sweden, supplies the high electron mobility transistor (HEMT) amplifiers that take over at the 4-kelvin stage.
This layer is fragmented rather than concentrated, which makes it hard to invest in and easy to underestimate. It’s also where sovereignty money is going first in Europe, because cabling is tractable in a way that fabrication isn’t.
Modularity Turned Supply Chains Into Markets
The structural change underneath all of this is quantum open architecture, or QOA: the idea that a quantum computer should be assembled from interoperable components bought from specialists, rather than built as a single proprietary stack.
The Tuna-5 system in Delft was the proof of concept, built almost entirely from Dutch suppliers, with QuantWare processors, Qblox control electronics, Delft Circuits cabling, and integration by Orange Quantum Systems, all inside a Bluefors cryostat. The Q-PAC system took the same approach commercially in the United States, assembled in months rather than years from QuantWare, Qblox, Maybell, and Q-CTRL components, with Arrow Electronics handling distribution.
For the third question on our screen, this is the whole ballgame. In the vertically integrated model, every layer of the supply chain is internal to two or three enormous companies and none of it is purchasable. Under QOA, each layer becomes a separate business with its own customers, margins, and competitive dynamics. The reason a supply chain map is worth drawing at all in 2026, and wasn’t in 2019, is that modularity created the markets the map describes.
It also changes what a national programme has to build. Italy’s largest quantum installation runs on QuantWare processors. Israel’s national centre pairs QuantWare chips with Quantum Machines orchestration. Neither country had to fund a full-stack champion.
Sovereignty Is Demand That Doesn’t Depend on Quantum Working
Export controls now restrict advanced cryogenic systems and quantum control hardware to certain destinations, and Chinese quantum firms have been added to the United States Entity List. Governments on both sides are funding domestic capacity in response.
The NATO study scored the supply chain across supplier count, research leadership, scaling capability, intellectual property position, and vulnerability, and it kept returning to two findings. Semiconductor manufacturing runs underneath everything, since processors, field-programmable gate arrays, and application-specific integrated circuits all come out of the same fabs, and the study recommended coordinating quantum supply chain policy with existing semiconductor programmes rather than building parallel ones. And rare earth processing, with more than 90% of high-purity capacity outside NATO territory, is a dependency nobody can fix quickly.
China is answering with vertical integration rather than modularity. Origin Quantum builds the processor, the cryogenics, the control system, the operating system, and the cloud platform in-house, and SpinQ has assembled a comparable end-to-end capability. That’s a coherent strategy for export markets, because a country buying a quantum programme rather than a component gets one supplier and one contract.
The consequence for anyone reading this commercially is straightforward. Sovereignty spending creates demand for cryogenics, control electronics, cabling, and fabrication capacity on a policy timetable rather than a commercial one. It pays out whether or not a quantum advantage arrives on schedule, and it favours suppliers with production inside allied jurisdictions.
The Map, Scored
| Layer | Suppliers able to deliver at volume | Survives a modality shift | Buyable today |
|---|---|---|---|
| Dilution refrigerators | Two dominant, several challengers | Partly. Neutral atoms and photonics need far less cooling; spin qubits need the same | Yes, mostly private |
| Helium-3 | Effectively state programmes | No | Barely; the plays are substitution and recycling |
| QPU fabrication | Mostly captive to hardware firms | No | Only through QuantWare and foundry models |
| Room-temperature control electronics | Four to six credible | Largely yes | Yes, private and public |
| Cryogenic wiring and microwave parts | Fragmented, many small specialists | Partly | Yes, but no scale player |
| Cryogenic control chips | One to two serious efforts | No | Private, early |
| Error correction and control software | Several | Yes, hardware-agnostic by design | Yes |
| Facilities, integration, distribution | Emerging | Yes | Yes, through incumbents |
Two patterns fall out. The layers with the most concentration are also the layers most exposed to the modality question, and the layers that travel best across modalities are the least concentrated. That is not a coincidence. Concentration in this industry comes from solving a problem specific to one physical approach, and specificity is exactly what a modality shift destroys.
What Could Break the Chain
Helium supply. Helium-4 disruption raises costs across cryogenics, fabrication, and healthcare simultaneously. Helium-3 disruption stops new superconducting installations outright.
Cryogenic capacity. If system demand grows faster than two or three manufacturers can expand, lead times become the industry’s real roadmap regardless of what the qubit-count charts say.
A modality result. A decisive scaling demonstration from neutral atoms, photonics, or topological qubits strands the superconducting-specific layers. Refrigerators, microwave control, and junction fabrication would find smaller markets; software, error correction, and hybrid classical integration would not.
Export control escalation. A faster split into parallel Western and Chinese supply chains removes scale economies from both.
People. The population that can design, fabricate, calibrate, and operate these systems is small enough to count, and it constrains every layer at once. That is the one input no amount of capital shortens quickly.
Questions Worth Asking
If you’re doing diligence on a hardware company, or negotiating access to a machine, the useful questions aren’t about qubit counts.
Who makes the cryostat, and what is the current quoted lead time for a replacement or an additional unit? Is the control stack bought or built, and if bought, from how many possible vendors? Is the processor fabricated in-house, in a partner facility, or purchased, and what is the yield on the current design? Which materials in the build have single-country sourcing? If this company’s cryogenics supplier prioritised a larger customer for two quarters, what would the schedule do?
The answers tell you more about a delivery date than any roadmap slide, and they take about twenty minutes to collect.
Where We Teach This
Supply chain structure is one of the harder things to learn from the outside, because the useful detail is spread across company announcements, procurement documents, and standards work rather than collected anywhere. We build it into our hardware and geopolitics material as a repeatable exercise: take a modality, decompose the machine, score each layer on supplier count, modality exposure, and market structure, and then argue the result against someone who scored it differently.
The hardware and geopolitics material at quantumacademy.com/ runs that exercise in sequence, across modalities, rather than leaving it to be assembled from articles. For the migration-side counterpart, where the same diligence habits apply to cryptography rather than cryogenics, pqcframework.org is the companion reference.
The industry needs to get from hundreds of physical qubits to the millions that fault tolerance demands. The physics is hard. Securing the helium, building the refrigerators, fabricating the processors, and manufacturing the control racks to go with them is a different kind of hard, and it’s the kind that shows up on a delivery schedule.