How the build is funded — capex, opex, grants, equity

The honest cost of a 100-qubit superconducting machine

A 100-physical-qubit transmon system is not a single line item. It is a stack — cryogenics, control electronics, fab-sourced devices, RF infrastructure, shielding, classical compute, and the building that holds it all to vibration and EMI tolerances. Before any conversation about grants, equity, or EuroHPC participation, the capex needs to be itemised honestly so that funders, procurement leads, and journalists understand what they are actually being asked to back.

The dominant capex blocks for Ireland Quantum 100 fall into roughly five buckets. First, the dilution refrigerator — a wet or dry-dilution cryostat with sub-15 mK base temperature, multiple plates (50 K, 4 K, still, cold plate, mixing chamber), and enough wiring real estate for ~400+ coax lines to feed a 100-qubit device with room for flux, drive, and readout per qubit plus Purcell filters and quantum-limited amplifiers. Second, the control stack: arbitrary waveform generators with sub-ns timing, RF up/down conversion, TWPA/JPA pump lines, and an FPGA-class real-time controller capable of mid-circuit measurement and feed-forward — non-negotiable for any honest surface-code roadmap. Third, the quantum processing unit itself: heavy-hex transmon devices, sourced as foundry runs, characterised, and re-spun. Fourth, the shielded room and services — mu-metal, vibration isolation, helium recovery, redundant power, low-loss cabling. Fifth, the classical HPC envelope for compilation, calibration, and hybrid variational workloads.

Capex versus opex — and why the ratio matters

For most superconducting platforms the capex-to-annual-opex ratio settles somewhere in the region of 4:1 to 6:1 over a five-year window, but the opex line is unforgiving and frequently underestimated. The single largest recurring cost is cryogenics consumables and power. A dilution fridge running continuously draws tens of kilowatts at the wall once the pulse-tube compressors, helium handling, control racks, and HVAC are added. Liquid helium has become structurally expensive and supply-fragile since 2022; closed-cycle dry-dilution systems mitigate but do not eliminate this.

The second opex block is recalibration and device replacement. Transmon qubits drift. T1 and T2 coherence times degrade with two-level-system defects, and individual qubits on a heavy-hex lattice can fail. Realistic operations assume a device re-spin or replacement cycle on the order of 12–24 months, and a continuous calibration overhead measured in compute-hours per day. Third is staff — cryogenic engineers, RF engineers, calibration scientists, compiler and middleware engineers for the Qiskit / PennyLane / Cirq / OpenQASM 3 toolchains, and a customer-facing applications team for the climate cohort. Fourth is software licensing and cloud egress for the hybrid workloads that pair the QPU with classical solvers.

Anyone modelling quantum investment in Ireland on capex alone will get the economics wrong by a factor of two within thirty-six months.

The grant stack: EuroHPC, IDA, SFI and Horizon Europe

Ireland's position inside the European quantum funding landscape is genuinely useful, and the routes are well-defined even if the awards are competitive. EuroHPC JU is the most structurally relevant instrument: its quantum-computing pillar co-funds the procurement and hosting of quantum systems integrated with national HPC sites, with member states matching Union contribution. EuroHPC Ireland participation is the cleanest path to non-dilutive capex co-funding for a sovereign machine, and the hosting-entity model fits a Tipperary-based site that intends to expose access to European researchers.

IDA Ireland supports capital-intensive, export-oriented technology infrastructure, and IDA quantum funding conversations sit alongside the agency's broader semiconductor and deep-tech remit. Science Foundation Ireland (now within Research Ireland) funds the academic-collaboration layer — the algorithms, the error-correction theory, the materials work that feeds back into device design. Horizon Europe Cluster 4 and the Quantum Flagship successor programmes fund consortium-scale R&D where a hosting site can be a partner rather than the prime.

None of these are grants you write in a weekend. EuroHPC hosting bids run to hundreds of pages of technical and financial annexes, with binding commitments on uptime, user access, and reporting. The engineering team carries that load alongside the build.

Equity, and why it is the smaller line

Equity is the most-discussed and least-dominant part of the stack. For a sovereign-compute facility with a climate-priority cohort, equity primarily funds the company — IMPT.io and the operating entity around Ireland Quantum 100 — rather than the machine itself. The machine is a hosted national asset whose capex is best matched against grant and state co-funding, because dilutive capital priced against a five-to-seven-year hardware depreciation curve is poor capital for the cap table.

Where equity does its real work is on the software, applications, and commercial-access layers: the compiler stack, the climate-workload application engineering, the customer onboarding, the integration with IMPT's offset-stack supplier evaluation, and the years of payroll before per-shot revenue becomes meaningful. That is a venture-shaped risk, and it is appropriately funded by venture-shaped money.

Where the climate workloads pay back

Quantum investment in Ireland has to justify itself against real workloads, not slideware. The climate cohort is where the economics close. Variational quantum eigensolver (VQE) and quantum phase estimation routines on 50–100 logical-adjacent qubits — even pre-fault-tolerance, with aggressive error mitigation — are credible for small active-space chemistry: amine and metal-organic-framework binding sites for direct air capture, transition-metal catalysts for green ammonia, electrolyte decomposition pathways in lithium-ion and sodium-ion cells, and photovoltaic absorber screening. None of these need a million qubits to be useful; they need a clean, well-calibrated 100-qubit device with mid-circuit measurement and a mature classical co-processor.

That is what justifies the funding mix. EuroHPC and IDA money is matched against a machine whose first-cohort outputs feed directly into