Quantum Data Centres: How QPUs Earn a Place in the Rack

Every major shift in computing has eventually had to earn its place in a rack, from the CPU multiplying across data centre halls to the GPU that followed it, moving from a specialist card to the dense clusters now training frontier AI models. Quantum processing units are the next candidate for that lineage, and the framing has long since shifted away from whether quantum machines can compute toward whether they can be operated at scale alongside the infrastructure the world already runs on. A quantum data centre is simply a facility built to meet the power, cooling and workflow demands of these machines, and while some are purpose-built, QPUs are more often co-located with classical systems inside existing high-performance computing centres. Modern facilities expect any new platform to conform to shared norms, from rack-mounted form factors to defined power and cooling envelopes, which means a QPU is now judged against that checklist rather than against the looser standards of a physics experiment.

Coordination is already happening at the national level. The Open Compute Project, the global collaborative community devoted to making data centre hardware more efficient, scalable and cost-effective, now treats quantum integration as a data centre design problem rather than a research question. In parallel, the UK's Quantum Data Centre of the Future programme, funded by Innovate UK, has produced a blueprint for a hybrid quantum-classical facility, and ORCA Computing has gone further by demonstrating a quantum computer operating inside a real data centre as part of a hybrid computing environment.

When Temperature Becomes a Facility Decision

The largest variable in quantum data centre design turns out to be temperature rather than software. Superconducting machines, which sit behind much of IBM's and Google's work, rely on dilution refrigerators to cool qubits to thousandths of a degree above absolute zero, and that demands heavy power, vibration damping and shielding. Trapped-ion and neutral-atom systems avoid millikelvin cooling but trade it for ultra-high vacuum chambers, calibrated lasers and magnetic field control, while photonic systems from the likes of PsiQuantum, Quandela and ORCA draw less power and can tap existing fibre-optic infrastructure. Because there is no standard way to build a quantum system, modality becomes a facilities decision as much as a physics one, which pushes operators toward a deliberately modality-agnostic stance.

The Scaling Constraints That Will Actually Bind

In practice the binding constraints tend to be the unglamorous ones. A 2026 paper from Oak Ridge National Laboratory, alongside related ORNL work, argues that scaling quantum will run into limits on cooling and helium-3, a material in limited supply and necessary for reaching the coldest temperatures, long before it meets architectural ones. Cooling, which today accounts for only 10 to 30% of the electronic load in a classical facility, is expected to flip that ratio and come to dominate. Bluefors, the Finnish manufacturer of cryogenic measurement systems, is addressing this with a modular cryogenic platform designed to expand cooling as qubit counts grow rather than requiring a rebuild each time.

Hardware is only half the equation, since sharing workloads across quantum and classical machines is a difficult software problem with a structural mismatch at its heart. As Jamie Friel, compiler team manager at Oxford Quantum Circuits, a Firgun Ventures portfolio company, has observed, high-performance computing workload managers are designed for jobs lasting days, whereas a quantum runtime may last barely a second. The industry habitually asks what counts as normal for quantum, when the more useful question is what counts as normal for a data centre.

Two Emerging Models for Hosting Quantum Systems

The first large-scale quantum facilities will be hybrid, handling classical work alongside quantum much as data centres already juggle CPUs and GPUs, because quantum will not render classical infrastructure obsolete. Two commercial models are emerging to serve those sites. The first is the vertically integrated hyperscaler campus, in which a single operator owns the entire stack: the hardware, the facility housing it, the cloud layer and the customer relationship. IBM exemplifies this, running the largest concentration of utility-scale quantum systems at its Poughkeepsie site in New York and at its first quantum data centre outside the United States in Ehningen, Germany, while others pursue variants of the same owner-occupier model at dedicated facilities in Pasadena, Washington and Boston.

Where the first model owns its building, the second rents it, with the quantum company keeping its hardware and its customers but installing the machines as a tenant inside a data centre that a specialist operator owns and runs, abstracting cryogenics and control electronics away from the buyer. OQC pioneered this colocation approach, moving from Equinix in Tokyo to London and then, in September 2025, into Digital Realty's JFK10 facility in New York, launched with Nvidia's GH200 Grace Hopper Superchips as the city's first quantum-AI data centre. That arrangement captures the quantum-AI co-evolution neatly, since AI and GPUs are for now doing the heavy lifting to bring quantum systems online, from calibration to error-correction decoding, laying the groundwork for quantum to return the favour as the technology matures. The same logic has placed a Quandela photonic processor inside an OVHcloud facility in France and an IBM-RIKEN system beside Japan's Fugaku supercomputer. In Europe the pattern runs through both policy and procurement, with the 2025 EU Quantum Europe Strategy placing HPC-quantum integration at its core and the EuroHPC Joint Undertaking coupling quantum processors directly to national machines such as Karolina in Czechia. What emerges is quantum as a complementary class of infrastructure, quantum pods with their own power and thermal needs sitting alongside HPC.

The Networked Quantum Data Centre

The phrase quantum data centre may conjure a single vast campus, but the more plausible path is modular and distributed, with several QPUs sitting inside one facility first and interconnected facilities following over time. The Royal Society's February 2026 paper frames precisely this, presenting multi-QPU sites as the most viable near-term architecture for distributed quantum computing and the basis for larger entanglement-based networks. The fibre industry has taken notice. At Fiber Connect 2026, the world's largest fibre broadband event, quantum was treated as an important infrastructure layer and emerged as a leading topic, prompting a complementary whitepaper. Nu Quantum is turning that into hardware, with the Cambridge-based quantum networking firm launching what it calls the world's first rack-mounted, data centre-compatible Quantum Networking Unit, which brokers real-time entanglement between separate processors so they behave as one larger machine. Packaging entanglement distribution into a standard 19-inch rack positions quantum networking as equipment a data centre already knows how to integrate and host.

Where the Advantage Will Accrue

Microsoft's quantum vice-president, Zulfi Alam, expects commercially valuable machines to sit in data centres by the end of the decade, and whether that arrives on schedule or sooner, three overlapping markets are forming around it: hosting quantum compute through cloud or managed facilities; quantum-plus-HPC and quantum-plus-AI infrastructure, where processors sit beside supercomputers and GPU clusters; and the networked data centre, where linked QPUs act as one machine. For an investor, the strategic point is that decisive advantages may accrue as much to whoever solves the unglamorous problems of cooling, power, orchestration and networking at scale as to whoever builds the most elegant qubit.

The competitive advantages within this domain favour European positions more than the hardware league tables suggest. Oxford Quantum Circuits has proved the colocation model across three continents, Nu Quantum is building the networking layer from Cambridge, Quandela is installed inside a French cloud facility, and Bluefors supplies the cryogenics underneath much of the field. The colocation model that OQC and others are proving suggests the winners will be those who make quantum behave like infrastructure rather than an isolated experiment. It’s important to remember that industrial infrastructure has historically been built quietly, in racks and cooling loops, long before anyone thinks to call it a revolution.