Marin Ivezic
A superconducting quantum processor runs at roughly 10 millikelvin, about a hundred times colder than the space between stars. Reaching that temperature takes a dilution refrigerator, a multi-stage cooling system that circulates a mixture of two helium isotopes and holds millikelvin conditions continuously for months at a time. Only a handful of companies worldwide build them to research quality. The helium-3 they circulate barely exists on Earth, and almost all of the commercial supply comes from the radioactive decay of tritium held in nuclear weapons stockpiles.
None of that appears on a qubit roadmap.
Comparisons between quantum computing modalities usually run on three numbers: qubit count, gate fidelity, and coherence time. A modality is the physical system used to encode a qubit, and there are five in serious commercial development. Superconducting circuits. Ions held in electromagnetic traps. Neutral atoms held in optical tweezers. Photons. Electron spins in silicon. Each one draws on a distinct supplier base, and those modality-specific supply chains have been mapped in detail on PostQuantum.com.
Underneath those specifics runs a shared infrastructure that every architecture pays into. Four layers carry most of the weight. The control electronics that turn an algorithm into analog pulses. The cryogenics that keep quantum states alive long enough to compute with. The error correction hardware that converts noisy physical qubits into reliable logical ones. And the raw materials that all three depend on.
We treat these four as a diagnostic rather than as a market map. For each layer there’s a question that separates a vendor with an engineered path to scale from a vendor extrapolating a curve, and the answers are usually easier to check than the qubit specifications they sit behind. What follows is technology and supply analysis. It is not investment advice.
Control Electronics
Every qubit has to be told what to do, and the instruction always originates in classical electronics. A shaped microwave pulse drives a gate on a transmon, the superconducting circuit that most large gate-based machines use as a qubit. A laser pulse drives a transition in a trapped ion. A voltage pulse adjusts the coupling between two electron spins in silicon. A radio-frequency signal steers light onto a specific atom in a neutral-atom array.
In each case the same class of instrument does the work: arbitrary waveform generators that synthesise the pulse, digitisers that read the result, and controllers built on field-programmable gate arrays. An FPGA is a chip whose logic is configured after manufacture rather than fixed in the factory, which makes it fast enough for real-time signal processing and reprogrammable as the experiment changes. Almost every quantum control stack in commercial use is built on them.
This is the most modality-agnostic layer in the entire stack. A control platform designed for superconducting qubits can serve trapped-ion, spin, or neutral-atom systems with firmware changes and a different radio-frequency front end. The customer base grows with every quantum computer installed anywhere, whichever encoding wins.
Four suppliers dominate outside China. Zurich Instruments, a Swiss company acquired by the test-and-measurement group Rohde & Schwarz in 2021. Qblox, a spinout from QuTech in Delft, which has become the reference control stack for open-architecture systems. Quantum Machines in Israel, which sells the orchestration layer for national quantum centres as well as individual instruments. And Keysight Technologies, the largest publicly traded company with meaningful control-electronics exposure, bringing decades of microwave measurement engineering to the problem. Tabor Electronics and Swabian Instruments serve narrower segments, and QuantumCTek and Origin Quantum supply China’s domestic programme.
A software layer has grown alongside the hardware. Q-CTRL in Australia builds firmware that sits between the algorithm and the instrument, suppressing noise and tuning gate performance across several vendors’ hardware. QuantrolOx applies machine learning to automated qubit tuning, a problem that grows quickly for gate-defined quantum dots where each qubit needs its own voltage calibration. The commercial point is the same as for the hardware: these companies sell into every modality, so their revenue doesn’t depend on the outcome of the architecture debate.
The Wiring Bottleneck
Every control line running from a room-temperature rack down to a millikelvin processor carries heat into the cold stage. The cooling power available at the coldest stage of a dilution refrigerator is measured in hundreds of microwatts. At a few hundred qubits, careful cable engineering keeps the heat load inside that budget. At a hundred thousand qubits, and certainly at a million, it doesn’t, and no amount of engineering on the cables themselves closes the gap. There’s a second, cruder limit alongside it: a cryostat has a finite cross-section, and coaxial cables have diameters.
This is the wiring bottleneck, and across gate-based modalities it is arguably the single hardest constraint on scaling. It doesn’t show up in a fidelity number.
Control Electronics Inside the Cryostat
The answer being pursued across the industry is to move the electronics into the cold. Either to an intermediate stage at 1 to 4 kelvin, or, more ambitiously, to the millikelvin stage beside the qubits themselves.
The dominant approach is cryo-CMOS, meaning complementary metal-oxide-semiconductor circuits, the standard transistor technology of the entire chip industry, redesigned to work at cryogenic temperatures where threshold voltages shift and noise behaves differently. Intel’s Horse Ridge programme generates qubit control signals from a chip operating at 4 kelvin. SemiQon in Finland has developed transistors specifically optimised for sub-kelvin operation. A different route runs through superconducting digital logic: SEEQC’s single flux quantum technology operates at millikelvin temperatures and aims to eliminate most of the external wiring rather than shorten it.
Cryogenic control is where several modality-specific supply chains converge. Whether the processor holds superconducting circuits or silicon spins, the classical electronics next to it face identical physics and identical thermal constraints. Companies that master cryogenic circuit design serve all of them, which is why semiconductor process expertise and quantum system integration keep pairing up in partnership announcements.
The question for this layer. What is the plan when control lines outnumber the cryostat’s thermal budget, and is the control stack a commercial platform with an upgrade path or an in-house build that has to be rewritten to scale?
Cryogenics
Coherence is the interval over which a qubit holds its quantum state before the surrounding environment scrambles it. Thermal energy, the random jostling of atoms at ordinary temperatures, is the most reliable way to destroy it. Cooling the system until thermal energy falls well below the quantum energy scales being manipulated is the most direct defence available.
How cold depends entirely on the modality, and the differences are strategically significant rather than merely technical.
| Temperature regime | Which systems need it | Cooling technology | Supply characteristics |
|---|---|---|---|
| 10–20 millikelvin | Superconducting processors, most silicon spin systems today | Dilution refrigerator | Very few qualified suppliers, helium-3 dependent, lead times measured in quarters |
| 1–4 kelvin | Superconducting nanowire photon detectors, cryogenic control electronics, hot-qubit spin designs | Pulse tube and Gifford-McMahon cryocoolers | Broad competitive supplier base inherited from MRI and semiconductor manufacturing |
| Ambient | Trapped-ion and neutral-atom apparatus, most photonic circuits | None for the qubits themselves | Atoms are laser-cooled to microkelvin while the chamber around them stays at room temperature |
A silicon spin design that operates at 1 kelvin rather than 20 millikelvin, what the field calls a hot qubit and what Diraq among others is pursuing, steps out of the most constrained tier of the cryogenic supply chain entirely and into a market served by industrial cryocooler vendors. That is a supply-chain advantage, and it arrives independently of anything happening to the qubits.
For the systems that can’t make that step, the supplier list is short. Bluefors in Finland is the dominant builder for quantum applications, with its platform underneath IBM’s large systems and manufacturing expanded into the United States. Oxford Instruments offers a publicly traded alternative serving both quantum and condensed-matter physics. Leiden Cryogenics, CryoConcept, and FormFactor round out the established Western base. Entrants are arriving from two directions at once: Maybell Quantum with denser systems in smaller footprints, and Kiutra in Germany with solid-state magnetic cooling that removes helium-3 from the design. China has built out roughly ten domestic manufacturers as part of a deliberate effort to remove Western dependencies from its programme.
Two Isotopes, Two Different Problems
Helium-3 is the working fluid. It doesn’t occur in useful quantities on Earth, and the commercial supply comes overwhelmingly from tritium decay in weapons stockpiles, which means production volume is set by nuclear stewardship policy rather than by demand. Neutron detection, medical imaging, and quantum computing now compete for the same fixed output. Maybell Quantum has signed a supply agreement with Interlune, a company planning to mine helium-3 on the Moon. Whatever one makes of lunar mining as a business, that contract tells you how the buyers rate the terrestrial supply.
Helium-4 is a different problem wearing the same name. It’s a commodity gas, used in cryocoolers, MRI scanners, semiconductor fabrication, and as a pre-coolant in dilution refrigerators. It’s far more abundant than helium-3 and it has still gone through repeated shortages, because healthcare, aerospace, and chip manufacturing all want it and production is geographically concentrated, with Qatar a major source. Quantum is a small buyer in that market and gets priced accordingly.
Three responses are in play, and two of them can already be bought. Helium-free cooling architectures and closed-loop recovery and recirculation are both available today, each with performance trade-offs against a conventional dilution refrigerator, while extraterrestrial supply remains years from deployment.
The question for this layer. What happens to this system’s thermal budget and helium demand at 10 times the current qubit count, and has the operator secured supply or simply assumed it?
Error Correction Hardware
Quantum error correction is widely described as a software problem. Real-time error correction, the kind fault tolerance actually requires, is a hardware problem, and it’s becoming its own supply chain layer.
Start with the vocabulary, because the two terms get used interchangeably in marketing and they mean very different things. A physical qubit is one piece of hardware: one transmon, one trapped ion, one electron spin. A logical qubit is one reliable qubit assembled from many noisy physical ones through an error-correcting code. Announced physical qubit counts and demonstrated logical qubit counts are separated by roughly two orders of magnitude on current hardware.
The surface code, the most studied approach, spreads one logical qubit’s information across a two-dimensional patch of physical qubits. You never measure the data qubits, because measuring a quantum state destroys it. Instead you repeatedly measure ancillary syndrome qubits, which report whether an error has occurred nearby without revealing the encoded information. A decoder, a classical algorithm, reads that stream of syndrome data and works out the most likely pattern of errors to correct.
Now the timing. In superconducting hardware, a syndrome round takes on the order of a microsecond, and the rounds repeat continuously for as long as the computation runs. If the decoder is slower than the syndrome stream, the backlog grows without bound and the machine’s effective clock speed collapses. So the decoder needs sub-microsecond latency, sustained, while processing syndrome data from every logical patch on the chip. Craig Gidney’s 2025 resource estimate for factoring RSA-2048 sits under a million physical qubits, an order of magnitude below the 20 million in the 2019 Gidney and Ekera analysis, and the syndrome data rate scales with whichever figure holds. Conventional processors don’t keep up. The decoding has to run on FPGAs, on purpose-built ASICs, or eventually on superconducting logic inside the cryostat.
Riverlane in Cambridge is the leading independent supplier in this layer, selling decoder hardware and its surrounding stack to manufacturers across several modalities rather than betting on one. Google Quantum AI builds its decoding internally and tightly coupled to its own hardware; its 2024 Willow result, published in Nature that December, showed the logical error rate falling as the code distance grew from three to five to seven, the first convincing demonstration of a surface code operating below threshold. IBM has moved toward quantum low-density parity-check codes, a family that promises far fewer physical qubits per logical qubit at the cost of a different and harder decoder architecture. Quantinuum has demonstrated error correction on trapped ions, where lower native error rates and all-to-all connectivity may allow simpler schemes.
Decoder hardware optimised for surface codes is not decoder hardware optimised for qLDPC codes, and a supplier’s addressable market depends on which family the industry standardises around. Nobody knows yet which one that will be.
The FPGA Dependency
Underneath all of it sits a dependency almost nobody prices. Riverlane’s decoders run on FPGAs. Qblox’s control instruments use them. Quantum Machines’ platform is built on them. The FPGAs involved are the highest-performance parts available, and there are two suppliers of consequence: AMD, which acquired Xilinx, and Intel, which acquired Altera. Those same parts are wanted by defence programmes, telecommunications infrastructure, and AI networking, and allocation conflicts are a routine feature of procurement rather than an occasional shock.
As volumes grow, production systems will likely move from FPGAs to ASICs, application-specific integrated circuits designed for one job and fabricated in quantity. That transition trades a two-supplier allocation risk for a foundry access and non-recurring engineering cost problem. It removes the dependency; it does not remove the exposure.
One more constraint has nothing to do with silicon. Riverlane’s annual error correction report has flagged the shortage of people who can do this work as the field’s binding limit, and the observation is echoed across national laboratory programmes. Decoder engineering sits at the intersection of quantum information theory, digital design, and real-time systems, and very few universities produce that combination. For anyone assessing a roadmap, hiring plans in this discipline are a legitimate leading indicator, and the same shortage shows up as opportunity on the other side of the table, which is what QuantumCareers.com tracks.
The question for this layer. Whose decoder, running on what hardware, at what measured latency, and does the roadmap distinguish clearly between what has been demonstrated and what has been announced?
Critical Materials
Quantum computers are not built from commodity inputs. They demand substances that are scarce, difficult to purify to the required grade, geopolitically concentrated, or all three. The purity requirement is the part most often underestimated: a material that is abundant at industrial grade can have three qualified suppliers at quantum grade.
Helium-3, covered above, is the clearest case. Fixed supply, growing demand, no commercial-scale alternative source in production.
Silicon-28. Natural silicon contains about 4.7 percent silicon-29, an isotope with a nuclear spin that scrambles the coherence of a nearby electron spin qubit. Enrichment above 99.9 percent silicon-28 extends coherence times by orders of magnitude, which is why every serious silicon spin programme depends on it. The supply chain is young. ASP Isotopes has been building enrichment capacity in South Africa and Silex Systems has a laser separation programme in Australia, while historical supply routes ran through Russia and are now constrained. Nobody has demonstrated the tonne-scale throughput that widespread deployment would need.
Niobium. The workhorse superconductor, used in processor wiring and in the niobium-titanium cabling inside cryostats. The US Geological Survey has consistently placed Brazil’s share of global production near 90 percent. Niobium isn’t geologically scarce, but processing it to the parts-per-billion purity that quantum devices tolerate is a specialised capability held by few firms.
Tantalum. Increasingly used in superconducting qubits, where several groups have reported longer coherence than aluminium equivalents. Supply is concentrated in Central Africa, with the governance and traceability problems that region’s mineral trade is known for, and quantum-grade purity narrows the list further.
Rare earth elements. Erbium and ytterbium underpin the optical amplifiers and detectors used in photonic computing and quantum networking; europium and neodymium appear in quantum memory research. The mining is not the constraint. Refining is, and the International Energy Agency puts China’s share of global rare earth refining capacity in the region of 85 to 90 percent.
Sapphire and specialty substrates. Superconducting processors are fabricated on high-resistivity silicon or on single-crystal sapphire, which offers better dielectric properties at higher cost. The crystal growers serving that market built their capacity for LEDs and semiconductors; quantum is a rounding error in their order books, which affects both priority and format availability.
The pattern across all six is consistent, and it differs from the pattern in the other three layers. The risk is that whole processing capabilities are concentrated in facilities and jurisdictions never chosen for this purpose, not that one company dominates a market the way Bluefors dominates dilution refrigerators.
The question for this layer. Which inputs in the bill of materials have fewer than three qualified suppliers at the required purity, and has the company signed long-term supply agreements or is it buying spot?
Where the Dependencies Converge
Mapping modality against layer produces a picture that is more useful than any single-technology analysis.
| Modality | Dilution refrigerator | Room-temperature control | Real-time decoding | Distinctive material exposure |
|---|---|---|---|---|
| Superconducting | Essential | Essential | Essential, highest throughput demand | Helium-3, niobium, tantalum, sapphire |
| Silicon spin | Essential for most current designs, avoidable for hot-qubit approaches | Essential | Essential | Enriched silicon-28 |
| Trapped ion | Not required | Essential | Essential, potentially lower throughput | Ultra-high vacuum components, precision lasers |
| Neutral atom | Not required | Essential | Essential | High-power laser sources, spatial light modulators |
| Photonic | Required for superconducting nanowire detectors | Essential | Essential, distinct architecture | Rare earths, silicon photonics fabrication capacity |
Two readings come out of the table. First, control electronics and real-time decoding are universal. Every architecture pursuing fault tolerance needs both, which is why we regard those two layers as the closest thing to a modality-independent position in the industry. Second, the modalities that escape the dilution refrigerator carry the heaviest modality-specific burdens instead. Trapped-ion and neutral-atom systems trade cryogenics for precision optics, ultra-high vacuum engineering, and laser supply chains that have their own concentration problems. Every architecture relocates the supply-chain problem to a different set of suppliers.
Using the Four Questions
Put together, the four layer questions form a short diagnostic that can be run against any vendor briefing, in any modality, without needing to adjudicate the physics.
- Thermal budget. What cooling power is available at the coldest stage, how much of it does the current wiring consume, and what changes at 10 times the qubit count?
- Control scaling. Commercial platform or in-house build, how many channels does it support today, and what is the upgrade path?
- Decoding. Who supplies the decoder, on what hardware, at what measured latency, and against which code family?
- Materials. Which inputs have fewer than three qualified suppliers, and are they secured contractually?
A vendor with an engineered path to scale answers these in specifics and volunteers the constraints. A vendor extrapolating a curve answers in adjectives. We find the distinction shows up faster in these four questions than in any discussion of gate fidelity, because the supply-side answers are checkable against public information in a way that internal benchmark numbers are not.
Supply constraints genuinely slow the arrival of large fault-tolerant machines, and it’s tempting to treat them as a reason to relax about cryptographic migration. That inference doesn’t hold. A cryptographically relevant quantum computer, meaning one capable of breaking currently deployed public-key cryptography, requires the whole enabling stack to scale, and industrial capacity is harder to forecast than physics. Migration deadlines are set by two things that have nothing to do with dilution refrigerator lead times: how long your data must stay confidential, and what your regulators and clients require and when. Those dates are already published. The hardware timeline is a variable; the compliance timeline is not.
Building the Judgment
Reading a hardware roadmap against its supply chain is a skill, and it isn’t taught in either the physics curriculum or the standard technology due-diligence playbook. It requires enough quantum engineering to know why 10 millikelvin is different from 1 kelvin, enough procurement literacy to know what a two-supplier dependency does to a delivery schedule, and enough discipline to keep asking what has been demonstrated rather than announced.
That is the kind of assessment our programs are built around, and it’s why we teach the hardware layers alongside the market structure rather than separately. Participants who work in investment, procurement, or technology strategy generally arrive able to read the qubit numbers and leave able to read what supports them.
The full program catalog is at quantumacademy.com/. For the modality-by-modality supply chain detail behind this framework, the underlying analysis is on PostQuantum.com.