Sovereign quantum and Irish data residency

When a researcher in Cork runs a chemistry simulation on a quantum processor in California, three jurisdictions touch that workload before the result comes back: the country where the circuit was authored, the country where the cloud broker resides, and the country where the cryostat physically sits. Each of those touch-points is a place where the problem definition, the ansatz, the parameter sweep, and ultimately the answer are visible to someone other than the team that paid for the work. That is the entire reason Ireland Quantum 100 is being built in Tipperary rather than rented from somewhere else.

What "sovereign quantum" actually means

The phrase gets thrown around loosely. For the purposes of this article — and for how we are building Ireland Quantum 100 — sovereign quantum means four concrete things, not slogans:

• The physical qubits, the dilution refrigerator, the control electronics, and the classical co-processor sit on Irish soil under Irish operational control.
• The job-submission API, the queue, the calibration data, and the result store are hosted on infrastructure governed by EU law and operated by an Irish entity.
• No portion of the workload — circuit IR, transpiled gates, measurement shots, or post-processed results — leaves the EU at any point in its lifecycle, including telemetry and support tooling.
• The operator can be subpoenaed only under Irish and EU process. There is no parallel obligation to a foreign government that could compel disclosure of customer workloads.

That last point is the one that gets glossed over in marketing decks. A machine physically located in Frankfurt but operated by a US-incorporated entity is not sovereign in any meaningful sense — the operator remains subject to extraterritorial discovery orders regardless of where the steel sits. Sovereignty is a legal property of the operator, not just a geographical property of the hardware.

Why quantum workloads are unusually sensitive

People underestimate how much information leaks through a quantum job. A typical workflow on a superconducting transmon machine looks like this: you compile a problem to OpenQASM 3, the platform transpiles it against a heavy-hex coupling map, schedules it onto the queue, runs some thousands of shots, and returns a histogram. On the face of it, the histogram is just numbers.

But the circuit itself is the intellectual property. A variational quantum eigensolver ansatz for a novel MOF used in direct-air-capture tells a competitor exactly which chemistry you are pursuing. A QAOA cost Hamiltonian for a grid-balancing problem reveals the topology and constraints of the network you are trying to optimise. Even the parameter sweep schedule — how you are walking the energy landscape — is competitively meaningful. None of this is recoverable from a finished paper; it lives in the job submission.

Classical HPC has the same problem in principle, but the industry has had thirty years to build defensible patterns around it. Quantum cloud is currently where classical cloud was around 2008: the dominant providers are foreign-headquartered, the terms of service permit broad telemetry, and the assumption is that early customers are doing toy problems where it doesn't matter. That assumption stops being safe the moment you run anything you would patent.

The architecture we are building, and why each layer matters for residency

Ireland Quantum 100 is a 100-physical-qubit superconducting transmon machine. The hardware stack is the conventional one for this class of device, and the residency story is layered onto it deliberately:

The cold stack

A dilution refrigerator running below 15 mK, housing the transmon chip with a heavy-hex coupling topology. The fridge itself is a piece of physics equipment — it doesn't process data — but the choice of where it sits dictates which jurisdiction the qubits are physically resident in. Ours sits in Tipperary.

The control plane

Room-temperature electronics generating the microwave pulses that drive single-qubit rotations and two-qubit gates, plus the readout chain that digitises dispersive measurements. This is where calibration data lives — T1, T2, gate fidelities, crosstalk maps. Calibration data is more sensitive than people realise: it tells you which qubits a customer is being routed onto and, by inference, the size class of problem they are running.

The classical co-processor and SDK surface

Qiskit, PennyLane, and Cirq users submit jobs through a gateway that compiles to our native gate set and schedules onto the device. The gateway, the compiler, the queue manager, and the result store all run on Irish-hosted infrastructure. Telemetry, logging, and operational metrics never traverse a non-EU endpoint. That is a deployment constraint, not an aspiration — it is enforced at the network layer.

The error-correction roadmap

One hundred physical qubits is a NISQ-class machine. The longer arc is surface-code error correction, which compresses thousands of physical qubits into a smaller number of logical qubits. The residency argument applies just as strongly there — perhaps more strongly, because logical qubits are where commercially meaningful chemistry and optimisation problems will run. Building the sovereign operational pattern now, on a NISQ device, is how you avoid having to retrofit it onto a fault-tolerant machine later.

EU law, GDPR, and the awkward bits

GDPR is the obvious framework, but it is an imperfect fit for quantum workloads because most of what gets submitted is not personal data. A QAOA instance for a logistics network doesn't contain anyone's name. The relevant frameworks are wider: the Data Governance Act, the Data Act, the forthcoming sovereignty requirements baked into EU procurement, and the sectoral rules that apply to specific customer cohorts — financial services, healthcare, critical infrastructure.

The practical consequence is that "is this workload allowed to run on a US cloud" is a question with different answers depending on which customer is asking. A pharma company running a protein-folding circuit against a candidate molecule has IP and possibly trial-relevant data considerations. A grid operator running an optimisation has critical-infrastructure considerations. A bank running a Monte Carlo replacement has DORA considerations. EU sovereign quantum capacity is the thing that makes all of those answers default to yes without a six-month legal review.

There is also the simpler operational point: an Irish customer running on Irish quantum infrastructure under Irish law has one legal regime to think about, not three. That removes a category of friction that currently keeps most regulated industries off quantum hardware entirely.

Why climate workloads come first

The first cohort on Ireland Quantum 100 is climate-science workloads — carbon-capture chemistry, photovoltaic materials, battery electrolytes, climate-relevant protein folding, grid optimisation, and climate finance. Two reasons.

First, these are problems where the quantum advantage on near-term hardware is most defensible. Variational chemistry on small active spaces, QAOA on bounded combinatorial problems, quantum-enhanced sampling for materials screening — these are workloads that map well onto a 100-qubit transmon device with realistic gate fidelities. They are not toy problems but they are also not waiting on fault tolerance.

Second, the workloads feed back into work we are already doing. IMPT's offset stack relies on identifying credible carbon-removal suppliers, and the chemistry that sits underneath direct-air-capture, mineralisation, and enhanced weathering is exactly the kind of small-active-space simulation a NISQ machine can contribute to. Sovereign infrastructure means a researcher can run that chemistry without exporting their candidate molecule to a foreign cloud first. The fuller technical roadmap for that cohort is on the climate workloads page.

What residency does not buy you

It is worth being honest about the limits. Sovereign quantum does not make a NISQ device suddenly fault-tolerant. It does not eliminate noise. It does not exempt you from doing the hard work of mapping your problem to a circuit that fits within current coherence times. It does not give you logical qubits before the surface code is ready. And it does not, by itself, make your results better than what you would get from a foreign cloud running the same circuit on similar hardware.

What it gives you is the freedom to run the workload at all, without the legal and commercial overhead that currently keeps regulated EU customers off quantum infrastructure. That is a smaller claim than "sovereign quantum is faster" — which would be nonsense — but it is the claim that actually matters to a procurement officer at a bank or a grid operator.

Where to start this week

If you are at an EU organisation thinking about quantum and the residency question is unresolved, the useful thing to do this week is small: pick one circuit you would actually want to run — a real one, not a Bell-state demo — and write down the data-classification questions your legal team would ask before you submitted it to a foreign cloud. That document is the input we need to size the early-access cohort. The Ireland Quantum 100 overview has the contact route. We will be more useful to you with a concrete workload in hand than with a generic enquiry.