AI September 9, 2026

Quantum Computing Cloud Pricing in 2026: Costs, Plans, and Providers

Quantum computing cloud access in 2026 costs anywhere from $0 to hundreds of dollars for a single 10,000-shot experiment, depending entirely on the provider and hardware you pick. Amazon Braket charges $0.30 per task plus a per-shot fee that runs from $0.000425 on Rigetti Cepheus to $0.08 on IonQ Forte, so the same experiment costs about $4.55 on Rigetti and $800.30 on IonQ. Azure Quantum uses gate-shot and execution-time billing with minimum program charges, IBM Quantum offers 10 free minutes a month, and Google's Willow remains quote-based early access. The billing unit, not the qubit count, drives your bill.

Key takeaways

The billing unit matters more than the headline rate. Per-shot, gate-shot, per-minute, hourly, and monthly subscription models are not interchangeable, and comparing rates across them is meaningless without a fixed workload.
On Amazon Braket, a standardized 10,000-shot job costs roughly $4.55 (Rigetti Cepheus), $14.80 (IQM Garnet), $100.30 (QuEra Aquila), or $800.30 (IonQ Forte), before storage, notebooks, and classical compute.
Azure Quantum's IonQ minimum program charges start at $12.42 without error mitigation and jump to $97.50 with it on Aria 1, so short experiments carry a floor cost.
IBM Quantum's Open Plan gives up to 10 minutes of QPU time per month at no charge, the widest free door into real hardware.
Reserved QPU hours run $2,500 to $7,000 per hour on Braket, so reservations only pay off for sustained, high-volume work.
Published QPU prices never equal the final bill: notebooks, storage, networking, and classical compute are billed separately on AWS and Azure.

How does quantum computing cloud pricing work in 2026?

Quantum cloud pricing in 2026 works through several incompatible billing units: per-task fees, per-shot charges, gate-shots, execution time, reserved hours, and monthly subscriptions. Each provider picks its own model, and some let hardware partners set different ones on the same platform. You cannot compare a per-shot rate against a gate-shot rate or an hourly reservation without first fixing the workload.

A shot is one execution of a quantum circuit. Because quantum results are probabilistic, a practical experiment runs thousands of shots to build a statistical distribution. A gate-shot multiplies the number of quantum gates in your circuit by the number of shots, so deeper circuits cost more even at the same shot count.

Execution-time billing charges for the seconds your job occupies the processor. Reserved hours give you exclusive access to a device for a fixed hourly rate, regardless of how many shots you run inside that window.

Tip: Before comparing any two quotes, write down the billing unit. A $0.02 per-10-millisecond execution rate and a $0.08 per-shot rate answer different questions, and the "cheaper" one depends entirely on your circuit depth and shot count.

Simulators sit apart from real hardware. Amazon Braket's local simulator is free, its managed SV1 state-vector simulator costs $0.075 per minute, and Azure's Rigetti QVM simulator is free. Simulators are where most development happens; real QPU time is where the bill grows.

How much does a circuit cost on Amazon Braket?

A single 10,000-shot circuit on Amazon Braket costs $4.55 to $800.30 depending on hardware, using the transparent formula of $0.30 per task plus a device-specific per-shot fee. The per-shot fee is the swing factor, ranging from $0.000425 on Rigetti Cepheus to $0.08 on IonQ Forte, a difference of roughly 188 times.

Here is the arithmetic. The task fee is fixed at $0.30. The shot cost is 10,000 multiplied by the device rate. Add them for the QPU subtotal, according to Amazon Braket's 2026 pricing.

Device	Per-shot rate	10,000-shot QPU cost	Reservation (per hour)
Rigetti Cepheus-1-108Q	$0.000425	$4.55	$4,100
IQM Garnet	$0.00145	$14.80	$3,000
QuEra Aquila	$0.01	$100.30	$2,500
IonQ Forte	$0.08	$800.30	$7,000

These figures cover QPU access only. They exclude Braket notebooks, S3 storage for results, classical compute, and networking, all billed through your AWS account. A realistic monthly experiment budget adds those overheads on top.

Rigetti's Cepheus-1-108Q became generally available on Amazon Braket on April 7, 2026, according to the AWS Quantum Technologies blog. AWS also reported that Braket program sets can run certain multi-circuit workflows up to 24 times faster by cutting orchestration and submission overhead, which lowers effective cost for batch jobs.

How do Azure Quantum's gate-shot and minimum charges compare?

Azure Quantum aggregates hardware partners under one interface but lets each set its own model, so there is no single Azure rate. Rigetti bills by execution time at $0.02 per 10 milliseconds ($2 per second), while IonQ bills by gate-shots with a minimum program charge that makes short experiments expensive.

IonQ Aria 1 on Azure costs $0.000220 per one-qubit gate-shot and $0.000975 per two-qubit gate-shot, per Microsoft's Azure Quantum pricing. The minimum program charge is $12.4166 without error mitigation, rising to $97.50 with it. IonQ Forte 1 charges $0.0001645 and $0.001121 per gate-shot, with a minimum of $25.7899 without mitigation or $168.195 with it.

The minimum charge is the hidden cost. Even a trivial test on Forte with error mitigation costs at least $168.195, because the floor applies regardless of how small the circuit is. Error mitigation on Aria 1 nearly multiplies the minimum by eight.

For sustained work, Azure Quantum offers an Aria-Forte subscription at $25,000 per month plus Azure infrastructure costs. That converts to roughly INR 21 lakh a month at current rates, so it targets institutions with continuous quantum workloads, not intermittent testers.

Microsoft's April 2026 Azure provider list included devices from 25 to 108 qubits, among them IonQ Aria 1, IonQ Forte, Quantinuum H2-2, Pasqal FRESNEL, and Rigetti Cepheus-1-108Q, according to Microsoft Learn's target list.

Which providers offer free or trial quantum access?

IBM Quantum offers the widest free door: its Open Plan provides up to 10 minutes of real QPU time per month at no charge, according to IBM Quantum documentation. Amazon Braket and Azure Quantum both offer free simulators, but their real hardware always carries a fee.

IBM structures access in tiers: the free Open Plan, a Standard pay-as-you-go option, Flex project allocations, and a Premium enterprise subscription. IBM does not publish a single universal dollar rate for Standard, Flex, or Premium in its public documentation, so those tiers are effectively quote-based and depend on allocation and contract terms.

Free simulators are the practical entry point on AWS and Azure. Braket's local simulator costs nothing, and Azure's Rigetti QVM simulator is free. You can develop and debug entire circuits on simulators before spending a rupee on QPU time.

Google Quantum AI positions Willow as an early-access program rather than a self-service plan. No public list price appears in Google's materials, so treat Willow as quote-based. Google reported its Quantum Echoes algorithm ran 13,000 times faster on Willow than the best classical comparison, but that is an algorithm-specific benchmark, not evidence of general cost savings.

Warning: IBM's free 10 minutes runs on real hardware, but it is capped monthly and shared. It is enough to learn the workflow and validate small circuits, not to run production research at scale.

Is renting cloud quantum access cheaper than buying hardware?

For nearly every organization in 2026, renting cloud quantum access is far cheaper than buying hardware. A quantum computer requires cryogenic or vacuum systems, specialized facilities, and expert staff, while on-demand cloud access starts at a few dollars per experiment. Ownership only makes sense at research-lab scale with continuous, heavy utilization.

The decision hinges on utilization. Task-and-shot pricing suits intermittent testing: you pay only for the shots you run. Reserved hours suit predictable, high-volume workloads, because an idle reservation still costs $2,500 to $7,000 per hour on Braket.

Work out the crossover. If a single 10,000-shot Rigetti Cepheus job costs $4.55, you could run over 900 such jobs for the price of one $4,100 reserved hour. Reservations only pay off when you saturate the device, running back-to-back jobs for the full hour.

The economics mirror general AI infrastructure, where rented capacity dominates until utilization is high enough to justify owning silicon. The same forces that drive CoreWeave's $100 billion AI cloud backlog push most quantum users toward rented, elastic access rather than owned machines.

How do hardware, fidelity, and error mitigation affect total cost?

Hardware choice, gate fidelity, shot count, and error mitigation shape the total bill more than the headline rate. Qubit count alone is not a price or value proxy, because platforms differ in connectivity, gate model, error rates, and how many shots a reliable result requires.

Consider error rates. IBM's Heron r3 processor has 156 qubits with a reported median two-qubit error rate of 1.17 × 10⁻³, according to IBM's newsroom. IBM has also demonstrated chips with up to 1,121 qubits, but those experimental systems differ from the processors routinely available through its cloud.

Lower fidelity forces higher shot counts to average out noise, and more shots mean a bigger bill. A noisy device with a cheap per-shot rate can cost more than a cleaner device if it needs ten times the shots to reach the same confidence.

Error mitigation is a direct cost multiplier on Azure IonQ. Turning it on raises the Aria 1 minimum program charge from $12.4166 to $97.50, and the Forte 1 minimum from $25.7899 to $168.195. Published gate-shot rates alone understate the practical floor.

A worked example: pricing one research workload

Suppose you need to run a variational circuit 20,000 shots deep on a superconducting device, with development done first on a free simulator.

Simulator development on Braket's local simulator: $0.
Two validation runs on Rigetti Cepheus at 10,000 shots each: 2 × ($0.30 + 10,000 × $0.000425) = $9.10.
One production run at 20,000 shots: $0.30 + 20,000 × $0.000425 = $8.80.
QPU subtotal: $17.90.
AWS overhead (notebooks, S3 storage, classical compute): typically a few dollars, so budget roughly $25 to $30 all in.

The same production run on IonQ Forte would cost $0.30 + 20,000 × $0.08 = $1,600.30 for the QPU alone. The 90-times gap comes entirely from hardware choice, which is why fixing the billing unit and device before quoting a budget is essential. For a broader view of the platforms shaping this space, see our guide to the top AI tools for 2026.

Fault-tolerant quantum computing remains a roadmap item, not a shipping cloud service. AWS framed its deepened collaboration with QuEra as a path toward fault tolerance rather than a currently available product, so 2026 pricing covers noisy, near-term hardware only.

Bottom line

In 2026, the cheapest way into real quantum hardware is IBM Quantum's free 10 monthly minutes for learning, and Amazon Braket's transparent $0.30-plus-shot pricing for intermittent experiments, where a 10,000-shot Rigetti Cepheus job costs $4.55. Azure Quantum's minimum program charges and error-mitigation premiums make short IonQ runs costly, so read the floor before you run. Reservations at $2,500 to $7,000 per hour and the $25,000-per-month Azure subscription only pay off under heavy, sustained use. Fix your billing unit and device first, budget for AWS or Azure overhead, and the same experiment that costs $4.55 on one chip can cost $800 on another.

Frequently asked questions

How much does quantum computing cloud access cost in 2026?

A single 10,000-shot experiment ranges from $0 on free simulators to $4.55 on Amazon Braket's Rigetti Cepheus and up to $800.30 on IonQ Forte. IBM Quantum offers 10 free minutes of QPU time per month. Azure Quantum's IonQ minimum program charges start at $12.42 without error mitigation. Real costs depend on the billing unit, device fidelity, shot count, and cloud overhead like storage and classical compute.

Which quantum cloud provider is cheapest for small experiments?

IBM Quantum's free Open Plan, with up to 10 minutes of QPU time per month, is the cheapest for learning and small circuits. For paid on-demand work, Amazon Braket's Rigetti Cepheus at $0.30 per task plus $0.000425 per shot is the lowest-cost transparent option, at roughly $4.55 per 10,000-shot job. Free simulators on Braket and Azure cost nothing for development.

What is the difference between per-shot and gate-shot pricing?

A per-shot fee charges for each full execution of a circuit, so a 10,000-shot job costs 10,000 times the shot rate. A gate-shot multiplies the number of quantum gates in your circuit by the shot count, so deeper circuits cost more even at identical shots. Amazon Braket uses per-shot pricing; Azure Quantum's IonQ uses gate-shots. The two are not directly comparable without knowing circuit depth.

Does error mitigation increase quantum cloud costs?

Yes, significantly on Azure Quantum's IonQ devices. Enabling error mitigation raises the Aria 1 minimum program charge from $12.4166 to $97.50, and the Forte 1 minimum from $25.7899 to $168.195. Because these are floor charges, even a tiny test circuit incurs the full minimum. Published gate-shot rates alone understate the real cost of a practical, mitigated program.

Is renting quantum computing cheaper than buying a quantum computer?

For almost all organizations, yes. On-demand cloud access starts at a few dollars per experiment, while owning hardware requires cryogenic or vacuum systems, specialized facilities, and expert staff. Reserved cloud hours cost $2,500 to $7,000 on Amazon Braket, and only make sense for sustained, high-volume work. Intermittent testing is far cheaper with task-and-shot pricing than any form of ownership.

Which quantum cloud providers offer free access?

IBM Quantum's Open Plan gives up to 10 minutes of real QPU time per month at no charge. Amazon Braket's local simulator and Azure Quantum's Rigetti QVM simulator are both free for development. Google's Willow is quote-based early access with no public free tier. Free simulators let you build and debug entire circuits before paying for any real hardware time.