Pillar A · State as of 2026-07-28

Is quantum computing useful for portfolio optimization today?

Not yet. As of July 2026, classical solvers like Gurobi and CP-SAT solve realistic portfolio-selection instances to a proven optimum in seconds, and no quantum method has shown a durable advantage over them at scale. A large 2025 benchmark found QAOA and quantum annealing failing to beat Gurobi — no better than random sampling on several instances — and our own 20 sealed runs put the classical solver at the proven optimum every time while QAOA sat 25–48% away. Portfolio optimization is a favorite quantum demo because it maps cleanly to a QUBO, but a clean mapping is not an advantage.
→ Leer en español
State as of: 2026-07-28

Status as of: July 2026. The verdict below is measured, not asserted: it rests on a public benchmark and on 20 sealed runs anyone can verify.

Short answer: not yet. Portfolio optimization — pick the subset of assets that maximizes return for a risk budget — is the single most common example used to sell quantum computing, because it maps cleanly onto the QUBO form that quantum optimizers eat. But mapping cleanly is not the same as winning. On realistic instances the strongest classical solvers return a proven optimum in seconds, and as of mid-2026 no quantum or hybrid method has beaten them in a way that survives a fair, tuned classical baseline. The clean mapping is exactly why the honest negative here is worth publishing: this is the problem quantum should be best at, and it still loses.

What the independent benchmark found

A large 2025 study, Quantum Portfolio Optimization: An Extensive Benchmark (arXiv:2509.17876), ran QAOA and quantum annealing against classical solvers across many portfolio instances. The finding was blunt: the quantum methods did not beat Gurobi, and on several instances did no better than random sampling — while Gurobi returned optimal or near-optimal solutions in seconds. This lines up with the emerging community standard, QOBLIB (Nature Computational Science, 2026), which states plainly that quantum advantage can only be claimed when the best known classical methods are outperformed. On portfolios, they aren't.

What our own sealed runs show

We ran the same fight ourselves and sealed every result. QAOA (depth-2, simulated) against OR-Tools CP-SAT on portfolio-selection instances, same instance, same budget, exact optimum as referee. The classical solver hit the proven optimum in all 20 runs. QAOA never did — and the gap did not shrink with size, it wandered:

How far QAOA sat from the proven optimum20 sealed runs · lower is better · classical = 0 (proven optimal)classical optimumn=1248% ±22n=1625% ±11n=2041% ±14Non-monotonic: 48 → 25 → 41. A bigger problem did not make quantumrelatively better. Variance is first-order; the "trend" does not survive.

The classical proof time did grow with size (about 0.05 s at 12 assets to ~29 s at 20, roughly 25× per four assets), so the classical side is not free — but "slower to prove optimality" is a very different thing from "beaten." CP-SAT still returned the optimum every time; QAOA returned an answer 25–48% worse and, at the larger sizes, ran out of its step budget before converging.

Why portfolios are the wrong place to look for an early win

The pitch is seductive because the math is tidy: mean-variance selection becomes a quadratic binary problem, and quadratic binary problems are what annealers and QAOA are built for. But tidiness cuts both ways — a problem that maps cleanly to quantum hardware usually also maps cleanly to a decades-tuned classical solver, and those solvers (branch-and-cut in Gurobi, CP-SAT's constraint propagation) are extraordinarily strong on exactly this structure. The honest place to hunt for early quantum advantage is a problem that is hard for classical methods specifically, not one that is easy for everyone.

What we know / What we don't know

What we know (measured, dated). An independent 2025 benchmark found QAOA and annealing failing to beat Gurobi on portfolios, sometimes matching random sampling. Our 20 sealed runs put CP-SAT at the proven optimum every time, with QAOA 25–48% away and no size trend. Classical proof time grows with size but still resolves in seconds to tens of seconds.

What we don't know. Whether a future hybrid method, better error mitigation, or a cleverer encoding changes this — progress is real and continuing (e.g. 2026 trapped-ion hybrid experiments report improvements in specific setups). What we can't yet call is the crossover: the problem size, if any, where a quantum method would overtake a tuned classical solver on a portfolio instance. Our runs are simulated (noiseless), so they flatter the quantum side; real hardware would face noise on top of the gap shown above.


Rosetta Q publishes verdicts with raw, reproducible data. This is educational content, not a product claim, and not investment advice. The verdict is "not yet," measured on the dates cited — it is a statement about today, not a prediction about the field.

Sources:
· Quantum Portfolio Optimization: An Extensive Benchmark (arXiv:2509.17876, Sep 2025)
· The Quantum Optimization Benchmarking Library — QOBLIB (Nature Computational Science, 2026)
· Gurobi Optimizer — commercial MILP/MIQP solver (reference baseline)
· Google OR-Tools CP-SAT — open-source constraint solver (reference baseline)
· Rosetta evidence ledger — RQ-0012, 20 sealed portfolio runs