Production quantum hardware paradigms in 2026.

1.Superconducting transmon

The dominant paradigm by qubit count. IBM Heron r2 (156Q) and Rigetti Cepheus (108Q) are the production references. Two-qubit gate fidelities reach 99.7% (Heron) and 99.1% (Cepheus). Coherence times are tens to hundreds of microseconds — short enough that deep circuits accumulate error faster than trapped-ion. Per-shot pricing is the cheapest in the category (Rigetti at $0.000425/shot via Braket) reflecting the fast gate times and operating economics. Best fit: variational algorithms with shallow ansatz, sampling-heavy workloads, hardware-aware compilation experiments.

2.Trapped-ion

IonQ Aria (25Q), IonQ Forte (36Q), and Quantinuum H2 (56Q, 12 logical) lead production deployments. Two-qubit fidelities are 99.5-99.8% in production, with R&D demonstrations reaching 99.99% via electronic qubit control (IonQ + Oxford Ionics merger). Coherence times reach microseconds to milliseconds — essentially unlimited for current circuit depths. Per-shot pricing is one to two orders of magnitude higher than superconducting ($0.03-$0.08 per shot on Braket; $97-168 per-program minimum on Azure) reflecting slower gate operations and operating cost. Best fit: deep-circuit quantum chemistry, algorithms where 2Q fidelity dominates the shot budget, logical-qubit demonstrations.

3.Neutral-atom (Rydberg / QCCD)

QuEra Aquila (256-atom analog) and Pasqal Fresnel (100-atom programmable) are the production references. Neutral-atom hardware operates in two modes: analog (Hamiltonian simulation via Rydberg blockade — what Aquila does) and digital (gate-based — Pasqal supports both). Pricing on Braket is per-shot ($0.01 for Aquila) for analog programs; Pasqal on Azure is per QPU-hour (EUR 3,000) for hardware access. Best fit: quantum simulation of spin systems, Maximum Independent Set problems, combinatorial optimisation via analog techniques.

4.Superconducting flux-qubit annealer

D-Wave Advantage2 (4,400+ qubits, Zephyr topology) is the only commercial annealer at scale. Annealers solve Quadratic Unconstrained Binary Optimisation (QUBO) problems via quantum annealing — fundamentally different from gate-model quantum computation. Per-shot or per-minute comparisons with gate-model providers are not directly meaningful. Free trial (1 min QPU + 20 min hybrid solver, expires after 1 month); production is quote-only annual subscription. Best fit: combinatorial optimisation (vehicle routing, portfolio optimisation, scheduling), hybrid quantum-classical pipelines.

5.Photonic

Xanadu (Borealis, X-series) is the leading commercial photonic player. Gaussian boson sampling and continuous-variable quantum computing are the primary applications. Not currently priced via the major cloud providers (Braket, Azure, IBM, Google) — accessed via Xanadu's own platform. Listed here for context; not in the per-provider analysis tables.

6.Why paradigm context matters for cost

Consider two hypothetical workloads. Workload A: 5-qubit shallow circuit, 1,000 shots, 2-layer ansatz. On Rigetti Cepheus the cost is $0.85 ($0.30 task + 1,000 × $0.000425). On IonQ Forte the cost is $80.30 (1,000 × $0.08 + $0.30). A 95× cost difference at the same shot budget — but the result quality is comparable because the shallow circuit doesn't accumulate enough 2Q error to differentiate.

Workload B: 20-qubit deep circuit (50 layers), VQE convergence target. On Rigetti Cepheus the algorithm needs 5-10× more shots than IonQ to converge because each 2-qubit gate is 0.9% less fidelity. The effective per-result cost flips: Rigetti at 10× shots is $4.55 vs IonQ at base shots $80.30 — Rigetti still cheaper, but the gap closes. At even deeper circuits (100+ layers) or tighter convergence targets, the per-shot economics reverse and trapped-ion is cheaper per converged result.

Paradigm choice is fundamentally an algorithm-fidelity-cost optimisation, not a per-shot price comparison.

See also: all providers index · methodology