The quantum company directory.
Search by name or technology. Open a profile for its sources and, where available, its QRank score.
Editorial research assessment, not a performance benchmark or stock rating.
Whole-directory coverage across 8 sectors. The results below follow your selected sector and filters.
These groups overlap: an assessed organization is also listed. Score edition .
Find a company. Follow the evidence.
Search quantum organizations by name, technology or application. Find official websites and, where reviewed, sources, access routes and limitations. Listed and assessed profiles are clearly distinguished.
Who is turning quantum hardware into usable systems?
15 assessed in computing hardware · Score edition 23 Sept 2026 · Method 1.0
Your question: Which organizations match computing hardware and my filters?
Develops quantum hardware intended to reduce the overhead of error correction.
Limitation: A design intended to encode a logical qubit is not a completed demonstration of universal fault-tolerant computation. Suppression of one error channel does not remove all logical errors.
rack-mounted trapped-ion systems; cloud access
turnkey superconducting quantum computers; cryogenic systems
ARQUE Systems develops silicon electron-spin quantum computers from research at RWTH Aachen and Forschungszentrum Jülich. Its scaling strategy combines semiconductor fabrication, electron transport between operating locations and eventual integration with cryogenic control electronics. The intended value is a compact processor architecture that can connect qubits without dedicating a separate control path to every interaction. Based in Aachen, Germany.
Limitation: The company’s large qubit-count roadmap and comparative performance multipliers are ambitions, not verified capacities of the announced small processor.
Traps ytterbium atoms and uses controlled gates, movement and measurement to run circuits and investigate logical encodings.
Limitation: Logical-memory and selected circuit demonstrations do not establish general application advantage. Magne’s announced commissioning is prospective; public self-service execution was not established by this review.
The assessment is primarily supported by organization-originated evidence. External evidence does not meet the published half-of-weight threshold.
This score reflects documented maturity using the computing hardware weighting.
Strength: Measured logical operations, atom-loss handling and a documented collaboration with Microsoft.
Limitation: Logical-memory and selected circuit demonstrations do not establish general application advantage. Magne’s announced commissioning is prospective; public self-service execution was not established by this review.
Primarily organization-sourced · Assessment reviewed
Score edition · Method 1.0. Editorial opinion within this sector, not a probability or performance benchmark.
carbon-nanotube qubits; semiconductor-compatible fabrication
Programs an energy landscape and samples candidate solutions with superconducting flux qubits. Leap also offers solvers that combine classical processing with quantum annealing.
Limitation: An annealer does not execute arbitrary gate-model circuits. Reported hybrid business outcomes do not by themselves establish quantum speedup; the separate gate-model program is not validated by annealing results.
External sources support at least half of the weighted assessment. They can include collaborators; this does not imply independent replication or technical superiority.
This score reflects documented maturity using the computing hardware weighting.
Strength: An operating cloud service, detailed hardware controls and external experimental use give researchers concrete ways to work with annealing.
Limitation: An annealer does not execute arbitrary gate-model circuits. Reported hybrid business outcomes do not by themselves establish quantum speedup; the separate gate-model program is not validated by annealing results.
External corroboration · Assessment reviewed
Score edition · Method 1.0. Editorial opinion within this sector, not a probability or performance benchmark.
Develops semiconductor quantum processors with a focus on manufacturing compatibility and spin control.
Limitation: This is not an eight-qubit entangled processor benchmark. Reliable exchange across the array and larger connected architectures remain unresolved in this experiment.
EeroQ develops quantum hardware using electrons held above superfluid helium. Its control-chip approach uses established semiconductor fabrication and mobile electrons, aiming to connect storage, gate and readout locations without a dedicated control wire for every electron.
Limitation: Addressable electron capacity, retained charge and a quantum gate have different meanings. The control architecture does not itself establish spin-preserving computation, a logical-qubit count or application advantage.
MAGIC trapped-ion architecture; scalable control
Develops compact quantum systems designed to operate alongside conventional data-centre equipment.
Limitation: The integration demonstration does not establish fault-tolerant computation or independent application performance. Announced integration targets should not be treated as delivered logical-qubit capability.
quantum systems; digital annealing; hybrid computing
Scope: Computing hardware. Reading a page never marks its records reviewed.
Keep a baseline and explicitly mark the evidence you have reviewed. This optional journal stays in this browser.
Directory inclusion is separate from assessment.
Gate-based circuits, analog simulation and annealing offer different controls and solve different classes of problems. A larger physical-qubit count alone does not establish better useful performance.
Use these questions when reading a company’s sources. They explain comparisons; they do not assign or change a QRank score. Every numerical example below is synthetic.
A selected result and a median answer different questions. Compare the same statistic, selection rule, measurement method and population before ordering two devices.
Synthetic sample: five coherence times are 40, 50, 60, 70 and 100 microseconds. The best is 100; the median is 60. Both describe this sample, but neither establishes the performance of every qubit or a useful circuit.
Ask for: Find the sample size, tested device set, excluded measurements, distribution and calibration date. If only the best value is disclosed, leave the median unknown.
NIST: defining mean, median and other measures of location
Link to this checkOperating a gate alone and operating it alongside other gates can expose different unwanted interactions. Simultaneous benchmarking explicitly tests an operating context that an isolated measurement does not establish.
Synthetic comparison: an error estimate rises from 0.2% in isolation to 0.8% with neighboring operations active. That is a 0.6 percentage-point change, or four times the original estimate. Without the method and uncertainty, those two numbers do not diagnose the cause.
Ask for: Find the active qubit set, simultaneous operations, benchmark protocol, uncertainty and calibration window. Do not compare an isolated best case with a simultaneous average as though they were the same test.
Gambetta et al.: individual and simultaneous randomized benchmarking
Link to this checkA physical-gate error, a logical-cycle error and an application failure rate use different units and protocols. Keep the operation, code and decoding conditions with the result. A smaller percentage with a different denominator does not settle a comparison.
Synthetic classical event model: if ten independent operations each have a 1% chance of a flagged failure, the chance of at least one flag is 1 − 0.99¹⁰, about 9.56%. This is not a conversion from quantum gate infidelity to circuit failure: coherent errors, correlations and cancellation can break that interpretation.
Ask for: Find the error definition, benchmark, number of repetitions and confidence interval. For logical results, also find code distance, cycle definition, decoding method and any discarded trials.
Magesan et al.: the assumptions of interleaved gate benchmarking
Link to this checkCircuit execution is one part of a complete experiment. Preparing the job, waiting for access, repeating measurements and processing the output can each take time. Device usage and elapsed time are also different accounting quantities.
Synthetic job: 1,000 repetitions of a 50-microsecond gate sequence contribute 0.05 seconds. Add 0.25 seconds for readout/reset, 2 seconds of preparation, 10 seconds waiting and 1 second processing: the complete toy job takes 13.3 seconds.
Ask for: Find included stages, repetitions, error-mitigation work, queue treatment and output-quality requirement. Compare equivalent completed tasks before interpreting a speedup.
IBM Quantum: workload usage and execution-time components
Change the assumptions in the experiment-time lesson →
Link to this checkBackground sources checked 2026-10-16. The examples are QubitWire teaching calculations, not results from the cited experiments.
A maturity index across hardware approaches. Scores do not compare raw qubits or prove one machine is faster.
Demonstrated system behavior, workload results and stated error conditions.
Access to an actual system, operational documentation and support.
Named external experiments or deployments with reported outcomes.
Usable tooling, developer documentation and maintained integrations.
Company guides trace the path from an announcement to a measured component, accepted system or exact product certification.
It identifies the traded instrument and its relationship to the underlying shares. Readers should use that instrument when checking issuer filings or a licensed quote provider; a warrant or Helsinki share line needs its own identity and market metadata.
Full company guide, sources and limits →It establishes the agreement described by the parties. Production readiness needs additional evidence about fabricated devices, yield, repeatability, packaging and accepted system performance. The record deliberately treats those as separate steps.
Full company guide, sources and limits →Yes, as transaction history with its original date and status. Use the later completed-transaction and reporting evidence to distinguish the completed state from the earlier proposal.
Full company guide, sources and limits →No. The complete encoded operation still has to account for the error processes that remain, including those associated with control, measurement and correction. The value of a biased-noise architecture depends on how those parts work together.
Full company guide, sources and limits →It can show that a particular process supports functioning quantum devices under stated conditions. The next questions concern reproducibility and integration; a small result need not answer those questions to be technically useful.
Full company guide, sources and limits →It can clarify what was delivered, accepted and made available within the testbed. The supplier and customer may describe different boundaries, so retaining both documents helps readers understand the actual service or research capability.
Full company guide, sources and limits →No. It changes the organizational context, not the experiment that was performed. The result should retain its original provenance and be linked to the current ownership timeline.
Full company guide, sources and limits →The channels must support the required amplitude, noise, bandwidth, synchronization and readout relationships. A larger count cannot answer those compatibility questions by itself.
Full company guide, sources and limits →It can alter waveform construction, fitting, scheduling, parameter selection or result handling. A version record helps separate those changes from physical-device changes and makes a later comparison more meaningful.
Full company guide, sources and limits →The temperature requirements belong to particular components and detectors, not to the word photonic as a whole. A system-level description should identify those components and their operating conditions.
Full company guide, sources and limits →No. These security products are intended to address cryptographic threats, including future quantum threats. The business and technical evidence therefore centre on deployment, assurance and integration rather than processor benchmarks.
Full company guide, sources and limits →No. It states planned milestones. Completion requires the relevant authoritative record and the exact product configuration to which it applies.
Full company guide, sources and limits →