Develop circuits in a notebook, test them on the emulator and run them on real quantum processors — with the same code, billed by the second, without a licence.
A quantum processor is not a faster computer for everything. It computes using superposition and entanglement, giving it an advantage in a few well-defined problem classes — but a disadvantage in almost everything else.
01 · Classical memory per additional qubit
doubled
Molecules, catalysts and battery materials themselves obey quantum mechanics. A classical computer approximates their behaviour, and the effort grows exponentially with every particle involved. A quantum processor, on the other hand, maps the system directly.
02 · Variables on the atom processor
100
Route planning, allocation plans, selection from many combinations: the number of possibilities grows with every variable. The neutral atom processor arranges atoms as a graph for this purpose — one atom per variable, the distance between two atoms represents their relationship.
03 · Qubits, gate-based
12
Photonic circuits deliver measurement results from probability distributions that can only be reproduced classically with great effort. This is used for Monte Carlo methods, for generative models and as a benchmark for new algorithms.
Three steps in the same working environment and with the same program code. Only the last one costs processor time.
The workspace starts with Jupyter and sechs pre-installed development environments for quantum programs. You write the circuit in Python like any other program and store results in the project's Object Storage.
It is the same managed workspace as on the AI platform, with the same hourly rate. When idle, it stops automatically after 60 minutes.
The emulator calculates the circuit exactly on classical hardware — without noise or with a noise model that simulates the processor. This way you know what result to expect before the first paid processor run.
Up to 32 qubits, the complete state fits into memory. On the NG series graphics cards, the same circuits run many times faster than on the instance's processor.
Once the circuit is tested, you send the same code to one of the two processors. The job is queued and executed, the measurement results return to your workspace.
You are billed for the time the processor executes your job, to the exact second. The waiting time in the queue costs nothing.
Both processors on offer are essentially optical machines: laser light traps individual atoms or guides individual light particles through a glass chip. What arrives in the cloud is an interface — you send a circuit and receive measurement results back.
An emulator holds the complete state of the circuit in memory: 2 to the power of n complex numbers for n qubits, 16 bytes each in double precision. Every additional qubit doubles the requirement — 17.2 GB for 30 qubits, 68.7 GB for 32. The table therefore lists the reference value for each configuration up to which the state fits into half of the memory.
| Configuration | Computes on | Memory | Qubits | Environment / h | Cap / month |
|---|---|---|---|---|---|
| NP3Learning, courses and first circuits with a few qubits | 4 vCPU, memory | 8 GB | 27 | €0.0567$0.0658 | €41.39$48.01 |
| NM4Develop algorithms, search parameters, cross-check results | 8 vCPU, memory | 64 GB | 30 | €0.3045$0.3532 | €222.29$257.86 |
| NM6The largest state vectors without a graphics card, long overnight runs | 32 vCPU, memory | 256 GB | 32 | €1.2179$1.4128 | €889.07$1,031.32 |
| NG-L4Many consecutive circuits, variational methods, noise models | 1 × L4, GPU memory | 24 GB | 29 | €0.8507$0.9868 | €621.01$720.37 |
| NG-H100Large circuits in double precision, with fast FP64 calculation | 1 × H100 SXM, GPU memory | 80 GB | 31 | €3.0561$3.5451 | €2,230.95$2,587.90 |
Hourly rate = instance price plus 15% platform share, as with the AI platform workspaces; the monthly cap corresponds to 730 hours. All amounts net, plus 19% VAT.
Calculation: n = largest integer with 2n × 16 bytes ≤ 50% of the memory. Example NM6: Half of 256 GB holds 232 amplitudes (68.7 GB), for 233 (137.4 GB) it is no longer sufficient. In single precision, one more qubit fits.
This is what a program looks like in the workspace: a circuit of two qubits, a thousand repetitions on the emulator, the result as frequencies. About half of the measurements result in 00, the other half in 11 — never 01 or 10, because the two qubits are entangled. On real hardware, a few such hits are added due to noise.
from qiskit import QuantumCircuit
from qiskit_aer import AerSimulator
# Two entangled qubits (Bell state)
qc = QuantumCircuit(2)
qc.h(0)
qc.cx(0, 1)
qc.measure_all()
ergebnis = AerSimulator().run(qc, shots=1000).result()
print(ergebnis.get_counts()) # approx. {'00': 507, '11': 493}
Each workspace starts with the same 6 open development environments. Two of them are also the access to the two processors — a program that runs on the emulator also runs there without modification.
The photonic processor computes gate-based, like the emulator: you place gates on qubits and measure. The neutral atom processor computes analogue: you do not describe gates, but how an arrangement of atoms should evolve over time — often the more natural form for optimisation and simulation.
12Qubits
Per second, net€0.3333$0.3867
Equals per hour€1,200.00$1,392.00
100Qubits
Per second, net€1.0000$1.1600
Equals per hour€3,600.00$4,176.00
This is how the camera of a neutral atom processor sees its qubits: 100 atoms in a flawless register. During measurement, only the atoms in the ground state remain — here the chequerboard of a magnetic order. Afterwards, it is reloaded — each trap only catches an atom with a probability of about 55% —, and optical tweezers push the atoms row by row into the block.
Illustration, not a photograph · calculated according to the process in the processor
Which technology will prevail remains to be seen. Each has a different strength and a different limit — and only two are currently available in this offering. If you develop on the emulator, you are not tied to any of them.
| Technology | Information carrier | Operation | Strength | Limit | Available here |
|---|---|---|---|---|---|
| Photons | Single light particles in waveguides of a chip | Chip at room temperature, light source and detectors cryogenically cooled | Hardly any interference from the environment, connection to glass fibre | Entangling gates only succeed with a probability | Available12 qubits |
| Neutral atoms | Single atoms, held in a vacuum by laser beams | Vacuum cell, atoms cooled by laser to millionths of a Kelvin | Many qubits, freely selectable arrangement | Few repetitions per second | Available100 qubits |
| Superconducting circuits | Circuits made of superconducting metal on a chip | Cooling to around ten thousandths of a Kelvin | Very fast gates, proven manufacturing | Short coherence time, complex cooling | Not on offer |
| Ion traps | Single charged atoms in an electric field | Ultra-high vacuum, control via laser | Very precise gates, every qubit can be coupled with every other | Slow gates, difficult to scale to many qubits | Not on offer |
| Spin qubits in silicon | Spin of single electrons in semiconductor structures | Cooling mostly below one Kelvin | Manufacturing related to today's chips, very small qubits | Early stage, few qubits so far | Not on offer |
Three items, each at its own rate: the working environment per hour, the processor per second, the workspace per gigabyte. The quantities below are an example for a small group developing an algorithm for a month and confirming it twice on real hardware.
| Item | Quantity | Rate | Amount |
|---|---|---|---|
| Workspace NP3Develop and emulate small circuits | 80 h | €0.0567$0.0658 / h | €4.54$5.27 |
| Emulator NG-L4Test runs up to 29 qubits, with noise model | 20 h | €0.8507$0.9868 / h | €17.01$19.73 |
| Photonic processorConfirmation runs of the gate-based circuit | 10 min | €0.3333$0.3867 / s | €200.00$232.00 |
| Neutral atom processorAn optimisation problem as an arrangement of atoms | 3 min | €1.0000$1.1600 / s | €180.00$208.80 |
| Workspace in Object StorageNotebooks, results and measurement data | 50 GB | €0.0070$0.0081 / GB per month | €0.35$0.41 |
| Total for the month | €401.90$466.20 |
Example quantities, no minimum purchase. All amounts net, plus 19% VAT.
Eight industries where research groups are testing quantum methods on real-world problems. Each row names the underlying problem class — and thus which part of the offering fits.
Electronic structure of battery and solar materials, catalysts for hydrogen production.
Suitable for: EmulatorPhotons
Binding energies of small molecules, folding questions as optimisation, selection of drug candidates.
Suitable for: EmulatorPhotons
Portfolio selection as optimisation, sampling for risk scenarios, detection of unusual patterns.
Suitable for: Neutral atomsPhotons
Deployment and resource planning, plus preparation for quantum-safe encryption.
Suitable for: Neutral atomsEmulator
Route planning, allocation of ramps and charging points, distribution of vehicles to orders.
Suitable for: Neutral atoms
Allocation of frequencies as graph colouring, design of fail-safe networks.
Suitable for: Neutral atoms
Machine learning with quantum circuits, design questions in chip manufacturing.
Suitable for: PhotonsEmulator
Models of magnetic materials and phase transitions, teaching and method development.
Suitable for: Neutral atomsEmulator
Examples, not customers: no names, no quotes. Instead, each project states the initial situation, the chosen configuration and the monthly amount — calculated using the rates on this page. Each can be used as a template for your own request.
Health and pharma
Initial situation
A research group determines the binding energy of a small molecule using a variational method. The model requires twelve qubits and thus fits on the photonic processor.
Decision
NM4 for development, 60 hours a month; for confirmation 15 minutes on the photonic processor.
Result
€318.27$369.19
per month, net — before starting credit
Transport and logistics
Initial situation
An innovation team in logistics checks whether route planning for 60 vehicles can be solved as an optimisation problem on a graph — without prior experience with quantum programs.
Decision
NP3 for 40 hours of development; ten minutes on the neutral atom processor, one atom per node of the graph.
Result
€602.27$698.63
per month, net — before starting credit
Basic research
Initial situation
A university chair offers a semester of quantum programming for 30 students. Each works around 20 hours a month in the notebook.
Decision
30 NP3 work environments with a total of 600 hours; to conclude, five minutes on the photonic processor as a demonstration.
Result
€134.02$155.46
per month, net — before starting credit
Request as a template — Semester course in quantum programming
Questions from the first conversation about a quantum project — not the ones that are easy to answer.
Not to get started. The development environments describe a circuit as a program in Python: create qubits, set gates, measure. Anyone familiar with linear algebra will understand what is being calculated.
The cheapest workspace costs €0.0567 per hour and is completely sufficient for exercises and initial algorithms.
The emulator calculates the circuit on classical hardware and delivers the exact result, optionally with simulated noise. The processor executes it on real qubits; its result is a distribution from many measurements and carries the noise of the hardware.
The usual order is therefore: develop and test on the emulator, confirm on the processor.
In the full state vector up to 32 qubits on NM6 with 256 GB memory. Each additional qubit doubles the memory requirement; the table under emulators gives the guideline for each configuration.
Tensor network methods calculate circuits with little entanglement far beyond this limit. How far depends on the circuit and not on the machine.
To the second for the time the processor executes your job. Queue waiting time, transmission and evaluation in the workspace do not count towards this. There is no base fee and no licence.
One minute costs €20.00 on the photonic processor and €60.00 on the neutral atom processor, both net.
The emulators run on the platform in Germany and Finland. The two processors are located in Helsinki, Finland.
The circuit and the number of repetitions go to the processor, the measurement results come back, both encrypted. Neither the job nor the result leaves Germany and Finland. Your remaining data stays in the workspace.
No. Both development environments are installed in the same workspace, and both processors are activated in the same project.
The choice follows the problem: gate-based circuits run on the photonic processor, optimisation and simulation tasks that can be expressed as an arrangement of atoms run on the neutral atom processor.
No. With 12 or 100 qubits without error correction, this is impossible. Research estimates assume around one million qubits working together with error correction over days for this.
Anyone encrypting data that must remain protected for many years should nevertheless plan the switch to quantum-safe methods now — stored data traffic can be decrypted later.
Yes: €100.00 net for new projects, once per organisation. We book it to your credit account upon activation; it is offset against the first invoices and can be seen in the console under credit.
This is enough for 5 minutes on the photonic processor, for example — enough to confirm a circuit tested on the emulator on real hardware.
Via the access request: state your project, requirements and scope, without an account. Within two working days, you will receive a setup proposal with configuration and costs; after that, we set up the workspace in the project and activate the two processors.
There is no minimum term: the workspace is billed per hour, the processor per second.
Describe your project in four short details — without an account. Within two working days you will receive a setup proposal with equipment and costs; after that we will set up the working environment in your project and unlock the two processors. There is no minimum term.