Is quantum computing useful for routing and logistics today?

Status as of: August 2026. Routing is the poster child of quantum optimization marketing and the hardest place to actually win, because the classical incumbents are brutal. This is the measured picture, not the brochure.

What is claimed?

Logistics is the pitch every quantum vendor reaches for: fleets, delivery windows, warehouses — combinatorial, expensive, everywhere. The claim, stated loosely, is that quantum optimization (QAOA on gate machines, or annealing on D-Wave) will crack routing problems that choke classical computers. The claim is almost never stated with the one thing that would make it checkable: against which classical solver, on which instance, at which size.

What is actually proven?

The classical side is not a straw man. It is a wall.

On the classical benchmark CVRPLIB-X (100 to 1,000 customers), a July 26, 2026 head-to-head has Hexaly 15.0 landing within 0.1%–0.9% of best-known across every size inside one minute, and Google OR-Tools within roughly 3.6%–7.8% in the same minute. These are 1,000-city routes, solved to near-optimal, on a laptop-class run. That is the baseline any quantum method has to beat.

Now the quantum side, on real hardware. The largest vehicle-routing problem actually executed on a gate-based quantum computer in the peer-reviewed record is 3 nodes, 2 vehicles. Scaling to 4 nodes produced a 12-qubit circuit with roughly 1,685 RZ + 883 RX + 425 two-qubit gates; it ran for over 4 hours and returned an infeasible solution (Azfar et al., 2025). The authors make no advantage claim; they name circuit depth and noise as the wall.

The annealing "case studies" are real work — and they are hybrid. In the most cited real-world one, a Spanish package-delivery study on D-Wave (Osaba et al., Scientific Reports, Oct 2024) ran 14–29 deliveries through a hybrid solver (LeapCQMHybrid) and matched OR-Tools' optimal routes where constraints allowed, drifting ≤6.3% only when forced to honor extra priority rules OR-Tools ignores. A December 2025 warehouse-AGV study scaled to 1,000 vehicles — but only by clustering the problem classically down to under 10,000 variables first, and its authors state plainly that annealers "lag behind these mature solvers" on general mixed-integer problems. In a hybrid pipeline, the classical layer does the routing; the quantum layer decorates it.

Where would the crossover be?

Not close. A 2025 resource study from Volkswagen and the Jülich Supercomputing Centre (Onah & Michielsen) asked what it would take to encode the smallest CVRPLIB instance on a gate machine at all.

Direct QUBO encoding of that smallest instance demands a minimum of 5,305 logical qubits; a space-efficient HOBO encoding still needs 7,685 for the Golden₅ case (down from 200,000+, but nowhere near hardware). Projected next-generation devices land at 400–1,200 qubits. Their verdict: early quantum advantage in capacitated routing is "unlikely on NISQ hardware even in the best-case scenario," with a few benchmark instances "one to two hardware generations away." That is an estimate on hardware that does not exist yet — the crossover point (see What is the crossover point?) for routing is a projection, not a measurement.

Why is routing the weak-baseline trap?

Because it is trivially easy to look good. Route quality against a naive nearest-neighbor or a random baseline improves with almost anything, quantum included — and that delta gets reported as "quantum optimization works." It means nothing. A routing result only counts if it beats the best known classical solver — OR-Tools, LKH, Hexaly, Gurobi — on the same instance at the same size and time budget. The strength of the classical incumbent is exactly what a weak-baseline benchmark hides (see Why a weak classical baseline ruins a quantum benchmark). In routing, the incumbent is world-class, so the bar is high and honest reports clear it by staying quiet about advantage.

What does Rosetta's own bench say?

We have not yet sealed a routing-specific verdict, and we will not imply one we did not run. Our nearest measured neighbor is portfolio optimization — the same shape of fight: QAOA versus a strong constraint solver (CP-SAT) on identical instances. In our sealed run (V-0012, 20 trials at n=12/16/20), CP-SAT hit the proven optimum 20/20 while QAOA stayed 25%–48% away, with the quantum side simulated noise-free — the most generous possible setting — and it still lost. Routing sits in the same regime as portfolios against an even stronger classical field. A dedicated Rosetta routing bench is on the backlog; until it is run, this verdict rests on the external record above, cited by date.

What we don't know

We do not know the true crossover size for capacitated routing — no one has measured it, because no machine can yet encode a benchmark instance, so every number here on the quantum side of the ledger is either a small hardware demo or a resource estimate on hardware that does not exist. We do not know whether a future non-QAOA, non-annealing quantum routing algorithm changes the picture. We have not sealed our own routing run, so we lean on external results and label them as such. And the hybrid annealing case studies are genuine engineering — we are not calling them worthless, only noting that the routing in them is classical.

Rosetta Q publishes verdicts with reproducible raw data. This is educational content, not a product claim. Every dated figure links to its source; negatives are reported, not hidden.