What Does It Cost To Access A Quantum Computer Today?

When IBM began shipping its first commercial scientific computer, the 701, in 1953, customers rented it by the month, paying around US$15,000 for a single shift of use. A 1956 antitrust consent decree later obliged IBM to offer its machines for sale on comparable terms, and by the late 1960s time-sharing firms such as Tymshare were selling remote computing access to organisations without a machine of their own, a precursor to modern cloud computing.

Quantum computing is working through the same choice between owning, leasing, and buying time, with public disclosures beginning to paint a picture of what each option costs. Universities can buy a small research machine outright for US$1.1 million; governments sign multi-year deals worth tens of millions that bundle hardware with research, training, and upgrades; cloud platforms sell time by the task or the minute. No standard price has emerged yet, but a menu of access models is taking shape.

On-Premises Access Runs From Research Machines To Bundled Rights

The most direct route is buying a small quantum system outright. IQM launched its five-qubit Spark in 2023 at under €1 million (US$1.1 million), with training and a year's free maintenance. Meanwhile, Rigetti disclosed two orders for nine-qubit Novera systems in 2025 for an aggregate price of c.US$5.7 million, including a dilution refrigerator and control electronics. Ownership at this level buys hands-on control of the hardware, a particular benefit for practical training and experimentation.

Meanwhile, EuroHPC, the EU’s joint supercomputing initiative, has co-funded quantum computer purchases for national supercomputing centres, with higher disclosed costs:

• €5 million (US$5.7 million) for IQM’s 24-qubit VLQ in Czechia

• €8.5 million (US$9.7 million) for Quandela’s photonic Lucy in France

• €12.28 million (US$14 million) for AQT’s trapped-ion PIAST-Q in Poland

• €13 million (US$14.8 million) for Pasqal's neutral-atom system in Italy

EuroHPC owns each system, co-funding up to half their acquisition, integration, and operating costs, while host centres run them for European users. Public procurement is typically built around a tender for a defined asset funded upfront, which is why outright ownership, the capital expenditure (capex) model, is prominent in disclosed deals. However, private buyers may prefer alternative purchase models.

Outside EuroHPC, buyers are increasingly separating the right to use a machine from outright ownership, and recent contracts show how far that separation can stretch. D-Wave's US$20 million Advantage2 sale to Florida Atlantic University is accompanied by a memorandum that could create an on-campus quantum applications academy. Quantinuum's filings describe a customer that chose to lease instead, securing exclusive on-premises access to a processor over 45 months, with the deal’s net investment recorded at US$21.6 million. Swiss Quantum Technology, a Swiss company focused on quantum hybrid applications and research, signed a €10 million (US$11.6 million) agreement with D-Wave to support Italy’s Q-Alliance in Lombardy. The deal moves further away from ownership again. Rather than buying a machine outright, the company is buying half of an Advantage2 system’s capacity for five years, with an option to purchase the machine itself. In this case, the buyer holds usage rights rather than a half-share of the hardware. From an enterprise perspective, In Japan, Toyota, Hitachi, Mizuho, MUFG, and Mitsubishi Chemical share a dedicated IBM machine run by the University of Tokyo for its industry consortium, showing that enterprises can secure dedicated capacity without owning it.

Each of these headline numbers reflects a different mix of ownership, access, and support. Read together, these deals suggest buyers are calibrating how much of the asset to hold against their confidence in its long-term value, and an option to buy later keeps that decision open alongside access today. Agreements that remain unannounced may well combine these elements in different ways.

Bespoke Deals Buy Far More Than Hardware

A different access model emerges at the top end of the market. Some of the largest public deals follow a common logic, in which the quantum machine is bought as the centrepiece of a wider national programme instead of as a standalone asset. QuEra's ¥6.5 billion (US$41 million) 2024 contract with Japan's National Institute of Advanced Industrial Science and Technology (AIST) covered a neutral-atom system integrated with AIST's ABCI-Q supercomputer. The Basque Government has taken that narrative a step further, as its €80 million (US$91 million) contract for Europe's first IBM Quantum System Two in San Sebastián covers the quantum computer plus research and training within a regional programme funded to 2028. Denmark is following the same template. QuNorth, owned equally by the state’s Export and Investment Fund (EIFO) and the Novo Nordisk Foundation, is spending €80 million (US$91 million) to acquire and operate Magne, a neutral-atom machine built by Atom Computing with Microsoft software and designed for 50 error-corrected logical qubits, with priority access for Nordic researchers and companies from around the turn of 2027. In all three cases, the headline figure prices a long-term national capability, of which the hardware is only the most visible part.

IBM's role in the Basque deal shows how a listed product becomes a bespoke one. IBM offers an on-premises plan alongside its cloud tiers, giving a customer dedicated access to a system on the customer’s own site that IBM services and maintains, with the price set only by quote. IonQ's QuantumBasel partnership, worth more than US$60 million across successive arrangements to 2029, runs the other way. The arrangement transfers ownership of an installed Forte Enterprise and a future Tempo system to QuantumBasel and is expected to span four hardware generations. Whether the vendor keeps the machine under its own management, as IBM does, or hands over title, as IonQ is doing, the buyer is committing to one company's roadmap for years, which is a significant technological and commercial choice.

Contracts Increasingly Buy The Next Machine

The case for renting also rests on how quickly the hardware is improving, since a quantum system bought today can be overtaken by its successor within a year, a pattern familiar from AI. NVIDIA's H100, launched in 2022, was followed in late 2023 by the H200, which NVIDIA said nearly doubled inference speed on a large language model, and chief executive Jensen Huang told investors in May 2024 that NVIDIA was on a 'one-year rhythm'. This left data centre buyers weighing every purchase against the next generation.

Quantum buyers and vendors are already writing upgrade paths into contracts to manage that risk. Germany's €25 million (US$28.4 million) Euro-Q-Exa programme covers a 54-qubit IQM machine inaugurated in February 2026 and an additional 150-qubit system due by year end. Pasqal's Italian system is due to move in 2027 from analogue operation (steering qubits continuously to mimic a physical system) to a hybrid mode, and Rigetti described both their 2024 Novera orders as upgradeable. Buyers are increasingly paying for what a system will become, making delivery dates, acceptance tests, and upgrade costs as important as the headline price.

The Cloud Offers Access Without The Capital Outlay

For buyers who do not want or require ownership or access to a dedicated quantum system, the cloud offers another way to gain access. Amazon Braket offers access to hardware from AQT, IonQ, IQM, QuEra, and Rigetti, while Microsoft's Azure Quantum offers access to IonQ, Pasqal, Quantinuum, and Rigetti. IBM runs its own fleet of processors with more than 100 qubits. Prices run from IBM's free 10 minutes per rolling 28-day window and US$96 a minute on pay-as-you-go to US$125,000 a month for Quantinuum’s Standard subscription on Azure.

Amazon's tariffs show how far the cloud lowers the barrier to entry for research teams. Braket charges US$0.30 per task (a submitted job running one circuit repeatedly) plus a device-specific fee per shot (one execution). As of September 2026, a 1,000-shot task costs US$0.725 on Rigetti's 108-qubit Cepheus and US$80.30 on IonQ's Forte without error-mitigation. Teams can therefore work on processors far larger than a research machine without committing millions in capex,though users share queues and give up hardware control.

Industry buyers have largely taken this route or pooled their access, and corporate purchases of on-premises machines remain rare among disclosed deals. HSBC’s 2025 bond-trading trial, for instance, ran on IBM’s Heron processor.

Different Buyers Are Finding Different Ways In

Like mainframe customers in the 1950s and 1960s, quantum buyers are choosing between owning, leasing, and buying time before a definitive answer to what a unit of useful computation is worth is cemented. Purchase models differ in control, physical access, maintenance, and upgrade terms, and how far today’s contracts can future-proof buyers against fast-evolving technology. The question is not simply price, but which arrangement secures improving capability at a price useful results can justify.