2026 Hardware Guide

Quantum Hardware Guide

Which Platform Is Right for Your Problem?

There are now five distinct quantum hardware approaches available through cloud APIs, each with different physical principles, performance characteristics, and cost structures. Superconducting circuits, trapped ions, neutral atom arrays, photonic processors, and quantum annealers are not interchangeable: the hardware you choose directly affects what algorithms you can run, how deep your circuits can be, whether you pay per shot or per minute, and how much classical overhead you need for problem reformulation. This guide covers all five approaches with 2026 performance data so you can pick the right platform from the start.

Hardware at a Glance

Approach Qubit count (2026) 2-qubit fidelity Coherence time Connectivity Cloud access
Superconducting 127–1000+ 99–99.9% (IBM Heron r2 best pairs) ~100 microseconds Heavy-hex / grid IBM Quantum (free), Google (research), AWS Braket
Trapped Ion 25–56 (algorithmic) 99.5–99.9% Seconds to minutes All-to-all IonQ via Braket / Azure / Google Cloud, Quantinuum via Azure Quantum
Neutral Atom 256–1000+ 99–99.5% ~1 second Reconfigurable QuEra Aquila via AWS Braket (analog Hamiltonian simulation)
Photonic N/A (optical modes) N/A (probabilistic gates) ~nanoseconds (photon loss) Integrated optical circuits Xanadu Borealis via Xanadu Cloud
Quantum Annealing (D-Wave) 5000+ N/A (not gate-based) Microseconds (annealing schedule) Pegasus graph D-Wave LEAP (free tier: 1 minute QPU/month)
Topological Early stage (8 logical qubits demonstrated, Feb 2025) Not yet publicly benchmarked In principle much longer (hardware protected) Linear chain Azure Quantum (preview access via Azure Quantum Elements)

Superconducting Qubits

The industry workhorse with the widest ecosystem

  • IBM
  • Google
  • Rigetti
  • IQM
Qubit count (2026)
127–1000+
2-qubit gate fidelity
99–99.9% (IBM Heron r2 best pairs)
Coherence time
~100 microseconds
Connectivity
Heavy-hex / grid (limited)

How it works

Superconducting qubits are tiny nonlinear circuits cooled to ~15 millikelvin inside a dilution refrigerator. A Josephson junction creates an anharmonic energy spectrum so the ground state |0⟩ and first excited state |1⟩ can be selectively driven with microwave pulses. Two-qubit gates (like CNOT or cross-resonance) couple neighbouring qubits capacitively or inductively. IBM's heavy-hex lattice reduces unwanted qubit-qubit crosstalk. Google Willow (2024) demonstrated error rates below the fault-tolerance threshold for the first time.

Key systems (2026)

IBM Eagle (127 qubits), Heron r2 (156 qubits, ~99.9% two-qubit fidelity on best pairs), Condor (1121 qubits). Google Willow (105 qubits, ~99.5% two-qubit fidelity, demonstrated below-threshold error correction).

Strengths

  • Mature ecosystem with Qiskit, Cirq, and tket
  • Free cloud access via IBM Quantum open plan
  • Fastest gate times (10-100 ns) in the industry
  • Largest absolute qubit counts available today
  • Extensive documentation, courses, and community

Limitations

  • Short coherence (~100 microseconds) limits circuit depth
  • Limited qubit connectivity requires SWAP overhead
  • Requires dilution refrigerator at 15 millikelvin
  • Fabrication variability between chips

When to use this platform

Learning, prototyping, and running most textbook algorithms (Grover, Shor, VQE, QAOA). The free IBM Quantum tier makes this the default starting point for anyone new to quantum computing.

Trapped Ion Qubits

Highest gate fidelity and all-to-all connectivity

  • IonQ
  • Quantinuum
  • Oxford Ionics
  • AQT
Qubit count (2026)
25–56 (algorithmic)
2-qubit gate fidelity
99.5–99.9%
Coherence time
Seconds to minutes
Connectivity
All-to-all (any qubit pair)

How it works

Individual atoms (typically Ytterbium or Barium) are ionised and suspended in a Paul trap using oscillating electric fields. Laser pulses or microwave radiation drive transitions between atomic hyperfine levels, acting as |0⟩ and |1⟩. Two-qubit gates use the shared motional modes of the ion chain: exciting the motion of one ion vibrates the whole chain, coupling it to any other ion. This gives full all-to-all connectivity with no SWAP overhead, which is a significant circuit depth advantage over nearest-neighbour architectures.

Key systems (2026)

IonQ Aria (25 qubits, algorithmic qubits #AQ 25), IonQ Forte (#AQ 35+). Quantinuum H2 (56 qubits, ~99.9% two-qubit gate fidelity, highest published two-qubit fidelity of any commercial system).

Strengths

  • Highest gate fidelities (~99.9% two-qubit) of any commercial platform
  • All-to-all connectivity eliminates SWAP overhead
  • Long coherence times (seconds) enables deep circuits
  • Identical qubits by nature (same ion species)

Limitations

  • Slow gate times (1-100 microseconds vs nanoseconds for superconducting)
  • Lower absolute qubit counts than superconducting
  • Sensitive to vibration and laser noise
  • Scaling to hundreds of qubits requires multi-trap architectures

When to use this platform

Algorithms requiring deep circuits or high fidelity: quantum error correction demonstrations, variational algorithms with many layers, and chemistry simulations where accuracy matters more than speed.

Neutral Atom Qubits

Reconfigurable arrays of thousands of atoms

  • QuEra
  • Pasqal
  • Atom Computing
  • Infleqtion
Qubit count (2026)
256–1000+
2-qubit gate fidelity
99–99.5%
Coherence time
~1 second
Connectivity
Reconfigurable (atoms can be moved)

How it works

Neutral atoms are laser-cooled to microkelvin temperatures and trapped in arrays of optical tweezers: tightly focused laser beams that each hold a single atom. The atomic ground and hyperfine states encode |0⟩ and |1⟩. Unlike trapped ions, the atoms carry no charge, making them immune to electric field noise. Two-qubit gates use Rydberg excitation: laser pulses promote atoms to high-energy Rydberg states where enormous electric dipole moments create strong nearest-neighbour interactions (Rydberg blockade). Uniquely, the tweezer arrays can be reconfigured mid-circuit, effectively rewiring the qubit connectivity on the fly.

Key systems (2026)

QuEra Aquila (256 qubits, analog Hamiltonian simulation mode, available on AWS Braket). Pasqal Fresnel (100 qubits). Atom Computing Phoenix (1180 qubits demonstrated in 2023, digital gate mode).

Strengths

  • Reconfigurable connectivity: atoms can be physically moved between operations
  • Large qubit counts (256-1000+) already demonstrated
  • Long coherence times (~1 second)
  • Supports both analog and digital operating modes

Limitations

  • Atom loss means qubit loss (no mid-circuit atom replacement yet at scale)
  • Gate fidelities still improving (currently ~99%)
  • Slower gates than superconducting
  • Cloud access currently limited to analog mode (QuEra Aquila on Braket)

When to use this platform

Quantum simulation of spin models and optimization problems, especially when dynamic connectivity is needed. The analog Hamiltonian simulation mode on QuEra Aquila is ideal for studying quantum phase transitions and Ising-type optimization.

Photonic / Continuous-Variable

Quantum computing with light at room temperature

  • Xanadu
  • PsiQuantum
  • QuiX Quantum
Qubit count (2026)
N/A (optical modes)
2-qubit gate fidelity
N/A (probabilistic gates)
Coherence time
~nanoseconds (photon loss)
Connectivity
Integrated optical circuits

How it works

Photonic quantum computers encode quantum information in the quantum states of light. The continuous-variable (CV) approach used by Xanadu works with squeezed light: states of light where the quantum noise in one quadrature is reduced below the vacuum level. Beamsplitters and phase shifters implement linear optical transformations. Measurement is done by homodyne detection. Crucially, the photonic chip itself operates at room temperature, though the best single-photon detectors (superconducting nanowire) require cooling. PsiQuantum targets fault-tolerant photonic computing via photon fusion at scale.

Key systems (2026)

Xanadu Borealis (216 time-bin modes, demonstrated Gaussian Boson Sampling advantage in 2022, available via cloud). PsiQuantum (in development, targeting millions of photons via silicon photonics foundry processes).

Strengths

  • Room temperature chip operation (no dilution fridge)
  • Photons travel at the speed of light: natural for quantum networking
  • Silicon photonics leverages mature semiconductor fabs
  • Photons do not decohere (they only get lost)

Limitations

  • Probabilistic gates: two-photon interactions are hard to engineer deterministically
  • Photon loss is the dominant error source
  • Universal quantum computation requires very large overhead
  • High-quality single-photon sources remain challenging

When to use this platform

Boson sampling experiments, Gaussian quantum chemistry simulations, and quantum networking or repeater research. Not yet competitive with gate-based platforms for general algorithms.

Quantum Annealing (D-Wave)

Massive qubit counts for combinatorial optimization

  • D-Wave
Qubit count (2026)
5000+
2-qubit gate fidelity
N/A (not gate-based)
Coherence time
Microseconds (annealing schedule)
Connectivity
Pegasus graph (~15 connections per qubit)

How it works

Quantum annealing is not gate-based computing. The processor encodes an optimization problem as an Ising Hamiltonian: a network of interacting spins. The system starts in a quantum superposition (transverse field) and slowly evolves (anneals) toward the ground state of the problem Hamiltonian. Quantum tunneling lets the system escape local minima that would trap a classical simulated annealer. D-Wave Advantage uses a Pegasus graph where each qubit connects to up to 15 others. Problems must be formulated as QUBO (Quadratic Unconstrained Binary Optimization) or Ising models. D-Wave also offers hybrid classical-quantum solvers for large problems that exceed the native QPU connectivity.

Key systems (2026)

D-Wave Advantage (5000+ qubits, Pegasus graph topology). Advantage2 prototype (1200+ qubits, Zephyr graph with 20 connections per qubit, higher energy scale). Hybrid solvers handle problems up to 1 million variables.

Strengths

  • Largest qubit counts of any commercial quantum system (5000+)
  • Proven results on real optimization problems (scheduling, logistics)
  • Free tier available via D-Wave LEAP
  • Hybrid solvers scale to million-variable problems

Limitations

  • Only solves QUBO / Ising formulations: not a universal computer
  • Limited connectivity (Pegasus graph) requires problem embedding
  • Not suitable for algorithms requiring arbitrary quantum gates
  • Quantum advantage over classical optimization remains contested for most practical problems

When to use this platform

Any combinatorial optimization problem that can be expressed as QUBO: vehicle routing, scheduling, portfolio optimization, protein folding formulations, and network design. The Ocean SDK makes QUBO formulation straightforward.

Topological Qubits

Hardware-protected qubits for long-term fault tolerance (early stage)

  • Microsoft
Qubit count (2026)
Early stage (8 logical qubits demonstrated, Feb 2025)
2-qubit gate fidelity
Not yet publicly benchmarked
Coherence time
In principle much longer (hardware protected)
Connectivity
Linear chain (current prototypes)

How it works

Topological qubits encode quantum information in non-local degrees of freedom of a physical system, making the qubit state intrinsically protected from local noise. Microsoft's approach uses Majorana zero modes: exotic quasiparticles that appear at the ends of a semiconductor nanowire coated with a superconductor (InAs/Al heterostructure). The qubit is defined by the parity of two Majorana modes at opposite ends of the wire. Local perturbations (thermal noise, charge fluctuations) affect only one end and cannot flip the qubit state without a correlated event at both ends simultaneously. Microsoft's Majorana 1 chip (February 2025) demonstrated the first topological qubit system. The claimed advantage is that the hardware suppresses errors at the physical level, potentially requiring fewer error correction rounds than superconducting or trapped-ion qubits to achieve the same logical error rate.

Key systems (2026)

Microsoft Majorana 1 (February 2025): 8 topological qubits demonstrated, targeting a million-qubit chip on a single device. Hardware is based on a superconductor-semiconductor heterostructure fabricated using TSMC-compatible processes. Current status: early research prototype.

Strengths

  • Hardware-level error suppression (not just software correction)
  • Target: million qubits on a single chip via semiconductor fab processes
  • If successful: dramatically lower overhead than surface codes on other hardware

Limitations

  • Still in early research stage - no publicly available cloud access at scale
  • Majorana qubits are difficult to create and measure reliably
  • No gate fidelity numbers published comparable to other platforms
  • Independent verification of Majorana mode claims is ongoing
  • Years to decades from practical large-scale quantum computation

When to use this platform

Currently not available for practical use. Follow Azure Quantum Elements for research preview access. Relevant for long-term planning if your workload requires millions of reliable logical qubits.

Which Platform for Which Problem?

Use these decision cards to match your problem type to the right hardware. For a deeper comparison of the software frameworks that run on each platform, see the framework comparison page or explore the qubit connectivity glossary entry.

Cloud Cost Comparison

Approximate pricing as of early 2026. Shot-based pricing means you pay per circuit execution; task fees are fixed per job submission. Always verify current pricing on the provider's own page before budgeting a project.

Provider / system Detail Approximate cost Notes
IBM Quantum Open (free) systems up to 127 qubits Free Paid plans for priority queue access
IonQ Harmony (via Braket) Per-task + per-shot pricing $0.00035 / task + $0.01 / shot
IonQ Aria (via Braket) Higher fidelity system $0.00035 / task + $0.03 / shot
Quantinuum H-series (via Azure) Credit-based system ~$0.00265 / shot equivalent Paid tiers for more credits
D-Wave LEAP QPU time quota Free tier: 1 min QPU / month Paid plans for more time
Rigetti (via Braket) Aspen systems $0.00035 / task + $0.0009 / shot
AWS SV1 Simulator State vector simulator (up to 34 qubits) $0.075 / task + $0.00075 / shot
QuEra Aquila (via Braket) Analog Hamiltonian simulation $0.01 / task + $0.01 / shot

Costs change frequently. For current pricing, see: AWS Braket pricing, Azure Quantum pricing, and D-Wave LEAP plans.

Continue Learning