Most of the published cost models for quantum hardware are written either by accelerator economists treating qubits like GPUs, or by lab physicists who have never had to depreciate a dilution fridge over a board-approved schedule. Neither is much help if you are actually trying to build, finance and operate a 100-physical-qubit superconducting machine in a regional Irish town. So this is the working version: the line items, the order-of-magnitude shape, and where the numbers genuinely move when you change an architectural decision.
What you are actually buying
A useful frame for quantum facility economics is to stop thinking "computer" and start thinking "small synchrotron with a software stack on top". The chip itself — the transmon array on a silicon substrate — is one of the cheapest physical components in the room. What you are paying for is the environment that lets that chip behave quantum-mechanically for long enough to be useful: a dilution refrigerator holding the package at sub-15 mK, a multi-stage shielding tower against magnetic and infrared noise, low-loss coaxial wiring with attenuators heat-sunk at every plate, cryogenic amplifiers, room-temperature control electronics, and the data-centre-grade plant that supports all of it.
So when people ask about quantum capex, they tend to fixate on the cryostat. That is an error of category. The cryostat is roughly a third of the hardware bill, and the hardware bill is roughly half of the build. The rest is the building, the power, the helium logistics, the control plane, the calibration software, and the people who can actually keep the system above 99% single-qubit fidelity for more than a fortnight at a time.
Capex: the line items that actually matter
For a 100-qubit superconducting transmon system on a heavy-hex or similar low-degree topology, the capital stack breaks roughly into six categories. I will not put euro figures on these — the spread between vendors is wide enough that any single number would be misleading — but the relative weights are stable across most architectures.
- Cryogenic platform: the dilution refrigerator itself, plus pulse tubes, compressors, gas-handling, vibration isolation. This is a long-lead item; order-to-install is typically twelve to eighteen months.
- QPU and packaging: the chip, the interposer, the sample holder, and the multi-chip-module assembly if you are going modular. Cheaper than people assume, but the yield story sits underneath it and the yield story is what determines how often you replace.
- Wiring and cryogenic RF chain: coaxial lines, attenuators, circulators, isolators, TWPAs or HEMT amplifiers. For a 100-qubit machine you are running into the low thousands of individual cryogenic components, and they are not commodity priced.
- Room-temperature control electronics: AWGs, digitisers, the FPGA-based control fabric, the timing distribution. This scales close to linearly with qubit count until you adopt cryo-CMOS, which changes the whole graph.
- Building and plant: vibration-isolated slab, EMI shielding, redundant power, chilled water, helium recovery, secure access. For a regional Irish site this is where local civil costs dominate over global hardware pricing.
- Software, calibration and integration: the control stack, the calibration loops, OpenQASM 3 frontends, Qiskit / PennyLane / Cirq compatibility shims, and the operator console. Often the most underbudgeted line in early plans.
The single biggest capex lever is not the qubit count — it is the wiring strategy. A naive one-coax-per-qubit approach blows up the fridge's cooling budget and the per-line cost long before you reach the surface-code regime. Any honest capex plan for a system above ~50 qubits has to engage with multiplexing, frequency-domain readout, and a credible cryo-CMOS migration path, because those decisions move millions, not thousands.
Opex: helium, electricity, and humans
Operating expenditure for a superconducting facility is unintuitive if your reference class is classical compute. A 100-qubit machine draws far less wall power than a single rack of modern AI accelerators — the QPU dissipates microwatts, and even the room-temperature control electronics for 100 channels are modest. The plant load is dominated by the pulse tubes, the compressors, and the building HVAC. For an Irish site, the climate genuinely helps: ambient cooling is cheap most of the year, and the grid carbon intensity is improving fast enough to matter for any customer doing climate workloads on the machine.
The three opex line items that actually swing the budget are:
- Helium: helium-3 in particular, used in the dilution mixture, is a strategic supply problem before it is a cost problem. A closed-cycle plant with proper recovery brings the marginal cost down sharply, but the recovery infrastructure is itself a capex item people forget.
- Electricity: predictable, hedgeable, and on an Irish site increasingly low-carbon. Less interesting than people think, until you start running 24/7 calibration sweeps.
- People: the binding constraint. A working machine needs cryogenic engineers, RF engineers, control-software engineers, calibration scientists, and a small applications team. In Ireland this is a tight market, and the salary curve is the real quantum opex line that compounds.
There is a fourth, quieter cost: scheduled warm-up cycles. Every time you open the fridge to swap a chip or repair a line, you lose roughly a week of availability. A facility that warms up six times a year has materially different unit economics from one that warms up twice. This is why packaging and modularity decisions belong in the opex conversation, not just the capex one.
Depreciation: how to actually book this
Depreciation is where quantum facility economics gets genuinely awkward, because the useful life of the components diverges by an order of magnitude. The building shell is a 25-to-40-year asset. The dilution refrigerator is realistically a 10-to-15-year asset with major service intervals. The control electronics are on a 5-to-7-year refresh cycle, faster if cryo-CMOS lands. The QPU itself — the actual chip — should be depreciated aggressively, because you will replace it as fabrication improves and as the fidelity floor under the surface code rises.
A defensible schedule treats the facility as a portfolio of assets with separate lives, not a single "quantum computer" booked over five years. That matters for two reasons. First, it gives the board an honest picture of when reinvestment cycles hit. Second, it stops the QPU refresh from being treated as an emergency rather than a planned event — which it should be, because chip improvement is the single largest source of capability upside in the next three years.
The other depreciation conversation worth having is software. Calibration code, compiler passes, error-mitigation routines — these are real assets with real maintenance costs, and treating them as pure opex understates what the facility is actually worth.
Why an Irish site changes the arithmetic
The Ireland quantum cost story has three structural advantages and one structural disadvantage. The advantages: low-carbon grid moving in the right direction, ambient-cooling-friendly climate, and an R&D tax credit regime that genuinely engages with deep-tech capex. The disadvantage: every piece of long-lead hardware ships in, and the local supply chain for cryogenic components is thin.
That last point matters for spares strategy. If a HEMT amplifier fails on a Friday and the nearest stocked replacement is in central Europe, the warm-up-and-repair calendar gets ugly. A serious Irish facility plan carries deeper on-site spares than a comparable site in Munich or Delft would, and that working-capital line needs to be in the model from day one.
The other Irish-specific factor is the customer mix. A sovereign machine sited here is not competing with hyperscaler cloud quantum on price-per-shot — that race is unwinnable and uninteresting. It is competing on data residency, on dedicated allocation for climate workloads, and on the ability to co-locate chemistry simulations with the downstream decision-making they feed. The economics only close if you take that seriously, which is why the climate-workload cohort sits at the centre of the operating plan rather than at the edge.
Where the numbers move most
If you are modelling this seriously, four decisions dominate the spreadsheet:
- Wiring architecture: multiplexed readout and a credible cryo-CMOS path versus brute-force coax. This is the single largest capex and opex lever above 50 qubits.
- Helium recovery: closed-loop or not. Changes opex profile and supply-risk exposure simultaneously.
- QPU refresh cadence: planned annual swaps versus reactive replacement. Changes availability, depreciation, and the ability to ride the fabrication-improvement curve.
- Operator headcount model: in-house calibration team versus vendor-supported. The honest answer for a sovereign facility is in-house, and that has to be funded properly.
Get those four right and the rest of the model is rounding error. Get any one of them wrong and the facility either runs out of cash or runs out of uptime, and the failure modes look identical from the outside.
Where to start this week
If you are thinking about funding, partnering with, or buying time on a facility like this, the most useful thing you can do this week is read the published cryogenic-component bills of materials from the academic groups that have built 50-to-100-qubit systems, and price the wiring stack yourself. It is the fastest way to develop a real intuition for where the money goes.