How Many Quantum Computers Exist in 2026? Verified Count

7 Jun, 2026

Introduction

Infographic displaying verified global quantum computer count for 2026 with definitions and estimate variance charts

Engineers evaluating quantum computing for production workloads face a fundamental inventory problem: nobody agrees on how many quantum computers actually exist. Estimates for 2026 range from under 50 to over 200 systems, depending on whether you count laboratory prototypes, cloud-accessible processors, or only fault-tolerant machines with logical qubits. This uncertainty isn't academic—it directly impacts procurement decisions, security postures, and R&D investment timelines. A team planning to integrate quantum random number generation or variational quantum eigensolvers into their pipeline needs to know which systems are production-accessible, which are research-only, and which are merely announced roadmaps.

This article delivers a verified global count of quantum computers operational in 2026, establishes rigorous taxonomy and inclusion criteria, and explains why published estimates diverge by an order of magnitude. We examine deployment status by modality, geographic distribution, and cloud accessibility, with concrete criteria for distinguishing quantum computers from quantum processors, simulators, and aspirational announcements.

Executive Summary

TL;DR: As of mid-2026, approximately 35–45 quantum computers meet strict operational criteria (programmable, quantum-mechanical processing, cloud or on-premise access, ≥20 physical qubits), while 120–150 systems exist under broader definitions that include sub-20-qubit prototypes and research-only devices.

Strict count (35–45): Systems with ≥20 qubits, programmable gates, cloud or contractual access, and demonstrated quantum advantage or utility beyond classical simulation.
Broad count (120–150): Includes sub-20-qubit research devices, non-programmable annealers, and systems with limited access or unpublished calibration data.
Superconducting systems dominate: ~60% of strict-count systems; trapped-ion, photonic, neutral-atom, and semiconductor modalities comprise the remainder.
Geographic concentration: ~55% deployed in North America, ~30% in Europe/UK, ~12% in Asia-Pacific, ~3% elsewhere.
Cloud accessibility gap: Only ~25 systems offer production-grade API access with SLAs; the remainder require academic partnerships or direct contracts.
Logical qubit systems: 3–5 systems have demonstrated error-corrected logical qubits at scale; these are not yet general-purpose computers.

Quick Q&A for Direct Answers:

Q: How many quantum computers exist worldwide in 2026? A: 35–45 by strict engineering criteria; 120–150 inclusive of research prototypes.
Q: What counts as a 'real' quantum computer versus a quantum processor? A: A quantum computer requires programmable universal gate sets, classical control interfaces, and demonstrated execution of non-trivial algorithms; processors lacking these are components, not systems.
Q: Why do estimates vary from 30 to 200+ systems? A: Variance stems from inclusion criteria (annealers, simulators, unpublished prototypes), access definitions (cloud vs. physical), and whether error correction is required.

What Counts as a Quantum Computer: Taxonomy and Inclusion Criteria

The divergence in global counts begins with definitional ambiguity. We propose a four-tier taxonomy derived from engineering practice and procurement realities:

Tier 1: Production Quantum Computers (Strict Count)

These systems satisfy all of the following:

Programmable universal gate set: Supports arbitrary single-qubit rotations and at least one entangling two-qubit gate (e.g., CNOT, iSWAP, Mølmer-Sørensen).
≥20 physical qubits with published connectivity graph: Below 20 qubits, most systems are classically simulable via tensor network methods, making quantum advantage claims unverifiable.
Classical control and readout infrastructure: Integrated microwave or laser control, ADC-based readout, and compiler stack (Qiskit, Cirq, PennyLane, or proprietary equivalent).
Documented access pathway: Public cloud API, contractual on-premise deployment, or documented academic partnership with SLAs or queue transparency.
Published calibration data: Gate fidelities, coherence times (T₁, T₂), and readout error rates in peer-reviewed literature or technical documentation.
Demonstrated algorithm execution: Published execution of non-trivial circuits (depth >10, width >10) beyond random gate validation.

Applying these criteria yields 35–45 systems globally as of June 2026. IBM Quantum leads with 8–10 systems (Falcon, Heron, and Flamingo processors), Google Quantum AI operates 2–3 Sycamore-derived systems, IonQ maintains 4–5 trapped-ion systems, Quantinuum (H0-H2 series) deploys 3–4, and Rigetti, D-Wave (Advantage systems, controversially universal), Xanadu, QuEra, Pasqal, and others account for the remainder.

Tier 2: Research Quantum Systems (Broad Count Additions)

Systems meeting most Tier 1 criteria but lacking one or more:

Sub-20 qubits but with novel modality or architecture (e.g., silicon spin qubits, topological qubits).
Non-programmable or restricted gate sets (D-Wave Advantage annealers, which support optimization problems but not arbitrary circuits).
Research-only access with no published queue or SLA.
Unpublished calibration data or limited algorithm demonstrations.

Tier 2 adds approximately 80–100 systems, bringing the broad count to 120–150. These include university laboratory systems (MIT, TU Delft, ETH Zurich, Tsinghua, NUS), national laboratory deployments (NIST, Oak Ridge, RIKEN), and startup prototypes not yet cloud-deployed.

Tier 3: Quantum Processors and Components

Not countable as computers. These are bare dies or cryogenic packages requiring external control electronics, often characterized by the manufacturer but not operable as standalone systems. Examples include Intel's Tunnel Falls (12 qubits, silicon spin) and various foundry test chips. Industry analysts frequently conflate processors with computers, inflating counts. Evidence-based analysis of quantum processor existence clarifies why bare dies and test chips don't constitute operational quantum computers.

Tier 4: Quantum Simulators and Emulators

Classical systems simulating quantum mechanics (tensor network emulators, analog quantum simulators using ultracold atoms without digital gate sets). These serve important research functions but are not quantum computers by any standard definition. Our evidence-based analysis of what constitutes a genuine quantum computer distinguishes these categories rigorously.

Verified Quantum Computer Count by Modality

Modalities exhibit starkly different maturity and deployment patterns. The following inventory reflects June 2026 status with conservative verification standards:

Superconducting Transmon Systems (~22–28 Tier 1; ~40–50 Tier 2)

Dominant modality by system count and qubit count. IBM Quantum's roadmap delivered Heron (133 qubits) and Flamingo (156 qubits) processors; Google Sycamore-derivative systems operate at 70–105 qubits with improved two-qubit gate fidelities (~99.5%). Rigetti's Ankaa-3 (84 qubits) and Aspen-M series contribute 3–4 systems. Alibaba, Baidu, and startups (Alice & Bob, IQM, Oxford Quantum Circuits) add European and Asian deployments.

Verification challenge: IBM and Google publish detailed system parameters; smaller providers often lack independent benchmarking. We count only systems with third-party circuit execution verification or published randomized benchmarking data.

Trapped-Ion Systems (~8–10 Tier 1; ~15–20 Tier 2)

IonQ's Forte (36 algorithmic qubits) and Aria series, Quantinuum's H2 (32 fully connected qubits), and Honeywell-derived systems offer all-to-all connectivity advantages. Our modality-by-modality guide to quantum computing companies details how trapped-ion architectures trade slower gate speeds for superior connectivity and coherence.

European trapped-ion systems (AQT/Innsbruck, IonQ's European deployment, TIQI at ETH Zurich) add 4–5 Tier 2 systems. Verification is strong: IonQ and Quantinuum publish extensive application benchmarks and have demonstrated quantum error correction with logical qubits.

Photonic Systems (~2–4 Tier 1; ~8–12 Tier 2)

Xanadu's Borealis (216 squeezed-state qubits, Gaussian boson sampling) and Jiuzhang 3.0 (USTC, 255 detected photons) demonstrated quantum advantage but with restricted, non-universal gate sets. PsiQuantum's million-qubit roadmap remains pre-deployment. Tier 1 status is marginal for photonic systems due to programmability limitations; we count Borealis and Jiuzhang as Tier 1 with caveats, plus 1–2 emerging programmable photonic processors.

Neutral-Atom Systems (~3–5 Tier 1; ~6–10 Tier 2)

Rapid 2024–2025 growth. QuEra's Aquila (256 neutral atoms, Rydberg blockade gates) and Pasqal's Fresnel series offer analog and digital modes. Harvard/MIT QuEra collaboration demonstrated 48 logical qubits with error correction. Atom Computing's 1000+ atom system (2025 announcement) awaits independent verification for Tier 1 inclusion. Benchmark comparisons across modalities reveal neutral-atom strengths in analog quantum simulation and emerging digital circuit capabilities.

Semiconductor/Silicon Spin (~1–2 Tier 1; ~8–10 Tier 2)

Intel's Tunnel Falls and similar academic systems (TU Delft, UNSW, RIKEN) remain predominantly sub-20 qubits and research-only. Delft's 2025 6-qubit silicon MOS system with ~99% single-qubit fidelities is promising but not yet Tier 1. Modality faces manufacturing scalability challenges despite CMOS compatibility advantages.

Topological and Exotic (~0 Tier 1; ~3–5 Tier 2)

Microsoft's Majorana-based approach (2023 retraction, 2025 revised claims) and various anyonic systems have zero verified operational computers. These remain speculative with no published algorithm execution.

Why Estimates Vary: Sources of Count Divergence

The 35–45 versus 120–150 discrepancy, and more extreme estimates (some analysts claim 200+), stems from five specific methodological differences:

1. Annealer Inclusion (D-Wave Controversy)

D-Wave Advantage systems (≥5,000 qubits) implement quantum annealing, not gate-based computing. They solve specific optimization problems via quantum tunneling in transverse-field Ising models but cannot execute Shor's algorithm, Grover's search, or arbitrary quantum circuits. Strict definitions exclude them; broad definitions include 5–7 Advantage systems. This single decision accounts for ~15% of count variance.

2. Cloud vs. Physical System Counting

IBM Quantum offers 8–10 distinct physical systems via cloud but consolidates access through unified APIs. Some counts tally physical cryostats (higher); others count cloud-accessible "systems" as logical endpoints (lower). Amazon Braket, Azure Quantum, and Google Cloud abstract underlying hardware, creating double-counting risks when analysts aggregate provider and platform inventories.

3. Prototype and Announcement Inclusion

Startups frequently announce qubit counts for systems not yet operational or calibrated. We exclude pre-operational announcements (e.g., certain 2025–2026 roadmap claims) pending independent verification. Industry reports from consulting firms often include these, inflating counts by 20–40 systems.

4. Simulator and Emulator Conflation

Classical simulators (e.g., NVIDIA cuQuantum, Atos QLM) and analog quantum simulators (ultracold atoms without digital gates) are sometimes marketed or catalogued as quantum computers. We strictly exclude both; some market analyses include them, adding 30–50 spurious systems.

5. National and Classified Systems

China's quantum computing program includes systems at USTC, CAS, and potentially military institutions with limited publication. We count 4–6 verified Chinese systems (Jiuzhang, Zu Chongzhi series) but acknowledge potential undercounting of classified or unreported systems. Similarly, US national laboratory systems (Sandia, Lincoln Laboratory) may not be fully disclosed. Our worldwide inventory analysis addresses geographic blind spots and verification methodologies for restricted-access systems.

Quantum Computer Deployment Status: Production vs. Research

Access modality determines practical utility for engineering teams:

Production Cloud Access (~25 systems)

Systems with documented APIs, pricing, queue management, and SLAs:

IBM Quantum: 8–10 systems via Qiskit Runtime (premium tiers for Heron/Flamingo).
Amazon Braket: 5–6 systems (IonQ, Rigetti, OQC, QuEra, D-Wave) with pay-per-shot pricing.
Azure Quantum: 4–5 systems (IonQ, Quantinuum, Rigetti, Pasqal).
Google Quantum AI: 1–2 systems (restricted access, research partnerships).
Xanadu Cloud: 1 system (Borealis, Gaussian boson sampling).
Direct provider access: IonQ, Quantinuum, Pasqal offer direct contracts bypassing hyperscalers.

Research and Partnership Access (~10–15 Tier 1 systems)

Systems with documented existence and calibration but restricted access:

National laboratory systems (US: NIST, Sandia, Oak Ridge; EU: Forschungszentrum Jülich, CEA; Asia: RIKEN, NII).
University-hosted systems with application-based access (MIT, TU Delft, ETH Zurich, USTC, Tsinghua).
Corporate research systems (Samsung, IBM internal development, Google internal).

On-Premise Commercial Deployment (~5–10 systems)

Direct physical installation at customer sites, rare due to cryogenic infrastructure requirements:

IBM Quantum System One and System Two: ~5 installations (Cleveland Clinic, RIKEN, University of Tokyo, Fraunhofer, École de Technologie Supérieure).
IQM on-premise systems: 2–3 European installations (VTT Finland, Leibniz Supercomputing Centre).
Rigetti quantum processing units integrated with classical HPC: 1–2 sites (UK National Quantum Computing Centre, DOE facilities).

Logical Qubit Systems: The Error-Correction Threshold

A critical distinction for 2026 counts: no quantum computer with sufficient logical qubits for fault-tolerant universal computation yet exists. However, 3–5 systems have demonstrated logical qubit primitives:

Google Sycamore (2024): 49 physical qubits → 1 logical qubit with 2.9% error vs. 3.0% physical error (below threshold). Extended to 105 qubits with improved codes.
Quantinuum H2 (2024): 32 physical qubits → 4 logical qubits with transversal CNOT gates using [[8,2,2]] color code.
QuEra/Harvard (2024): 48 logical qubits in reconfigurable neutral-atom array, demonstrated logical entanglement and error detection.
IBM Heron (2025): Heavy-hexagon lattice with error suppression; logical qubit demonstrations ongoing, not yet published.

These are proof-of-concept demonstrations, not production logical quantum computers. Our reliability metrics analysis details logical qubit overhead requirements: ~1,000 physical qubits per logical qubit for surface code with 10⁻¹² logical error rates, implying million-qubit systems for meaningful fault-tolerant applications. Current "logical qubit" demonstrations use 10–100× fewer physical qubits with higher error targets.

Failure Modes in Quantum Computer Counting

Engineers and analysts encounter specific failure modes when evaluating quantum computer inventories:

Failure Mode 1: Qubit Count Inflation

Symptom: Providers report maximum qubit counts without connectivity, fidelity, or calibration context.

Diagnostic: A 1,000-qubit system with 99% single-qubit and 90% two-qubit gate fidelity has less effective computational capacity than a 100-qubit system with 99.9% and 99.5% respectively. Request quantum volume, CLOPS (circuit layer operations per second), or application-specific benchmarks.

Mitigation: Normalize to algorithmic qubits (IonQ's metric) or effective qubits adjusted for fidelity and connectivity constraints.

Failure Mode 2: Access Pathway Opacity

Symptom: System exists physically but cannot be accessed for meaningful workload execution.

Diagnostic: Check cloud provider integration, queue depth, and pricing transparency. Research-only systems may have 6–12 month application queues.

Mitigation: Prioritize systems with documented API response times and published availability metrics.

Failure Mode 3: Modality Mismatch for Workload

Symptom: Selecting system based on qubit count alone, ignoring connectivity and gate set suitability.

Diagnostic: Optimization problems suit annealers or analog neutral-atom modes; quantum chemistry requires all-to-all connectivity (trapped-ion advantage); quantum machine learning benefits from high circuit depth (superconducting with low error rates).

Mitigation: Map algorithm requirements to modality strengths before system selection. Supply chain constraints may further limit modality availability by region.

Failure Mode 4: Classical Simulation Boundary

Symptom: Deploying to quantum systems for classically simulable problem sizes, incurring queue latency and cost without computational benefit.

Diagnostic: Tensor network methods (MPS, TTN) simulate ~50 qubits with limited entanglement; exact diagonalization handles ~20 qubits. Verify quantum advantage claims for your specific problem structure and instance size.

Mitigation: Establish classical baselines using simulators (Qiskit Aer, Cirq, PennyLane) before quantum deployment.

Performance and Scaling Trajectories

Quantum computer counts must be evaluated against growth trajectories and quality metrics:

Qubit Count Growth (Physical)

Superconducting systems: ~2× qubit count every 18–24 months (IBM: 27 qubits 2019 → 133 Heron 2024 → 156 Flamingo 2025; Google: 53 Sycamore 2019 → 70–105 2024). Trapped-ion: slower growth due to laser control complexity (11 2019 → 36 Forte 2024). Neutral-atom: rapid catch-up (256 Aquila 2023 → 1000+ announced 2025).

Gate Fidelity Improvements

Two-qubit gate fidelities: superconducting ~99.5% (Google, IBM), trapped-ion ~99.8% (Quantinuum), neutral-atom ~99.5% (QuEra, 2024). Error correction requires ~99.9% for practical overhead; ~99.99% for early fault-tolerant systems. Each 0.1% improvement reduces logical qubit overhead by ~2–3×.

Cloud Utilization Metrics

IBM Quantum reports >3 trillion circuits executed via cloud (cumulative through 2025). Typical queue depths: 1–4 hours for premium systems, 12–48 hours for free tier. Commercial workload growth (optimization, chemistry, ML) increasing 2–3× annually; research workloads stable.

Logical Qubit Roadmap

Industry consensus targets: 100 logical qubits by 2028–2029 (error-corrected, universal), 1,000 by 2032–2035. Physical qubit requirements: 100,000–1,000,000 depending on code efficiency and gate fidelity. Current demonstrations (3–48 logical qubits) are 2–3 orders of magnitude below this threshold.

Production Best Practices for Quantum Computer Evaluation

For engineering teams assessing quantum computing integration:

Decision Checklist: Counting and Selecting Systems

Define inclusion criteria explicitly: Universal gates? Minimum qubits? Cloud access? Published benchmarks? Document your criteria before comparing vendor claims.
Verify independent benchmarks: Prefer systems with third-party randomized benchmarking, quantum volume certification, or application benchmarks (e.g., QED-C, MLPerf for Quantum).
Evaluate total cost of ownership: Cloud per-shot pricing ($0.01–$1.00 per shot depending on modality and system) versus on-premise cryogenic infrastructure ($5M–$20M capital, $500K–$2M annual operating).
Assess classical integration maturity: Quantum-classical hybrid algorithms (VQE, QAOA, quantum machine learning) require tight loop integration. Evaluate Qiskit Runtime, Cirq, PennyLane, or proprietary stack maturity.
Plan for modality obsolescence: Current superconducting dominance may shift if neutral-atom or photonic systems achieve superior logical qubit scaling. Avoid vendor lock-in through abstraction layers.
Security posture for quantum threats: If evaluating quantum for cryptography (QKD, QRNG) or threat assessment (Shor's algorithm risk), inventory when cryptographically relevant quantum computers may emerge. Current estimates: 2030–2040 for RSA-2048 factorization, with high uncertainty.

Monitoring and Verification Protocol

Establish continuous verification for cloud-accessed systems:

Track published calibration data drift (gate fidelities, T₁, T₂) via provider APIs.
Benchmark standard circuits (GHZ state preparation, Bernstein-Vazirani, QFT) on each access to detect system degradation.
Monitor queue latency and availability; escalate if p95 latency exceeds 2× published SLA.
Audit classical simulation baselines quarterly; re-evaluate quantum advantage claims as classical algorithms improve.