Is quantum computing useful for energy and power grids today?

TL;DR — status as of September 13, 2026. No. A power grid is scheduled by optimization software, and on every grid problem with a public, dated measurement, the classical side still wins — or the quantum side has not published a comparison. The only independent benchmark of hybrid quantum pipelines on grid problems — AC optimal power flow and unit commitment at 5–13 generators — found they did not beat uniform random sampling (Gaidai & Mukherjee, arXiv:2607.15543, July 2026). Meanwhile, the software that runs actual grids has been classical mixed-integer programming since PJM switched its market in 2005, and the standard academic test fleet exceeds 900 generators (Knueven, Ostrowski & Watson, Sandia presentation). What the energy sector has in 2026 are funded explorations — $7M from AFRL (Eaton, August 6, 2026), $6.2M from ARPA-E (Infleqtion, February 9, 2026) — and feasibility demos without a published classical baseline (EDF–Pasqal–GENCI, January 21, 2025). Rosetta Quantum tracks the verdict in one unit: generators. The largest quantum grid benchmark ran 13; classical test suites run 900+.

Why is a power grid an optimization problem?

Because every few minutes, someone has to decide which generators run, at what output, through which lines — and each of those decisions is a constrained optimization problem with binary variables. Unit commitment (UC) decides which plants switch on across the next day. Optimal power flow (AC-OPF) sets outputs and voltages subject to physics. Contingency analysis asks which combinations of failures would break the system. These are the same class of problem as routing and logistics — combinatorial optimization — which is why every honest verdict about quantum-for-grids inherits the verdict of quantum-for-optimization: promising theory, no measured advantage at scale. The renewables transition makes the instances harder (more variability, more scenarios, more devices), which is the real reason utilities are paying attention — and the real reason the classical baseline keeps improving too.

What does the classical side already do in production?

It runs the actual grid, daily, and has for two decades. U.S. market operators moved unit commitment from Lagrangian relaxation to mixed-integer programming starting with PJM in 2005, and operators want a UC solution within 10–15 minutes of wall-clock time (Knueven, Ostrowski & Watson, Sandia/OSTI). The reference study of MIP formulations tests instances of 900+ generators over 48-hour horizons at a 0.01% optimality-gap tolerance (INFORMS Journal on Computing, 2020). That is the bar. Any quantum claim on grid optimization that does not benchmark against a well-tuned MIP solver on the same instances is a weak-baseline claim by construction.

What has quantum actually measured on grid problems?

Three public, dated results — and none of them favors quantum yet.

First, the independent benchmark Rosetta already covered in our hybrid audit: Gaidai & Mukherjee benchmarked hybrid quantum-classical pipelines (QAOA-family) on AC-OPF and unit commitment instances of 5–13 generators and found they did not outperform uniform random sampling of the solution space (arXiv:2607.15543, July 2026).

Second, a study at NTNU ran stochastic unit commitment — 3 generators, 3 time periods, with up to 15,000 demand scenarios — through D-Wave's hybrid solver against Gurobi. The authors report Gurobi remained superior on the instances tested, with the hybrid approach producing feasible solutions close to classical ones only under strict runtime limits, and performance sensitive to penalty-parameter tuning (arXiv:2512.03925, December 2025). Annealing-based grid claims should be read with the annealing-vs-gates map at hand.

Third, the survey record: a January 2026 survey of quantum approaches to unit commitment catalogs four paradigms (annealing, variational hybrids, quantum ML, quantum-inspired) and documents no demonstrated advantage over production MIP solvers (arXiv:2601.01777, January 5, 2026). A May 2026 review in Nature Reviews Electrical Engineering maps nine grid task families, concludes the credible near-term role is inside otherwise classical workflows, and insists — as Rosetta does — that comparisons be made against well-optimized classical baselines, not simplified ones (Lin et al., Nat. Rev. Electr. Eng. 3, 406–420, May 21, 2026).

What are utilities actually piloting?

Exploration, stated in future tense — which is the correct tense. The public record as of September 13, 2026:

Two of these rows will produce checkable numbers: the AFRL program will run for 24 months and has committed to a proof-of-concept demonstration, and the ARPA-E ENCODE program will publish against a stated motivation that classical solvers “are reaching their computational limits” — a claim that will need same-instance evidence when the results land. Rosetta will read both rows against the pilot-evaluation gates when they report.

Where could quantum plausibly help — and what would count as evidence?

The honest candidate is stochastic unit commitment. Renewables force operators to schedule against uncertainty, and the scenario tree blows up combinatorially — that is the stated motivation of the NTNU study (arXiv:2512.03925) and of both federal programs. If a quantum advantage on grids ever materializes, it will have to appear as a measured crossover on exactly that kind of instance. The evidence bar does not move: same instances, a strong tuned MIP baseline (not uniform sampling, not an untuned heuristic), full wall-clock accounting of the hybrid split, and artifacts a third party can re-run. Until a result clears that bar, grid quantum computing will remain what the record shows today: 13 generators on the quantum bench, 900+ on the classical one.

What we don't know

• Whether private pilots hold unpublished comparisons. EDF, EPB or Duke may have internal classical baselines they chose not to publish. Absence of a published comparison is what we can verify; absence of any comparison is not.
• Whether production-scale operational numbers match the public record. Market operators do not fully disclose production solve times or gaps; our 900+/0.01% figures come from the reference academic test suite, not from PJM's control room.
• Where the stochastic-UC crossover sits, if anywhere. No measured crossover point exists for any grid problem. Treat any date you hear as a forecast, not a measurement.
• Our own bench limit. Rosetta has not yet run unit-commitment instances in-house; our measured evidence on quantum optimization comes from adjacent problem classes. A grid instance on our own bench will be logged in this ledger when it happens.

Sources

• Gaidai & Mukherjee — Benchmarking Hybrid Quantum-Classical Algorithms for Power Grid Optimization Problems, arXiv:2607.15543 (Jul 2026)
• Lin et al. — Quantum computing for smart grid, Nature Reviews Electrical Engineering 3, 406–420 (May 21, 2026)
• Hasanzadeh, Rajabi & Kargarian — A Survey on Applications of Quantum Computing for Unit Commitment, arXiv:2601.01777 (Jan 5, 2026)
• Ribes & González Grandón (NTNU) — quantum-annealing stochastic unit commitment study, arXiv:2512.03925 (Dec 2025)
• Knueven, Ostrowski & Watson — Mixed Integer Programming Formulations for the Unit Commitment Problem, Sandia/OSTI presentation · INFORMS Journal on Computing (2020)
• Eaton — $7M AFRL contract press release (Aug 6, 2026)
• Infleqtion — ARPA-E ENCODE program announcement (Feb 9, 2026)
• The Quantum Insider — Pasqal, GENCI and EDF EV smart-charging demo (Jan 21, 2025)
• Utility Dive — Quantum computing is coming fast. Are utilities ready? (Aug 14, 2026)
• Rosetta Quantum — When does “hybrid quantum-classical” hide the classical doing the work? · Is quantum useful for routing and logistics? · Why a weak classical baseline ruins a quantum benchmark · How to evaluate a quantum pilot without burning budget · Quantum annealing vs gate-based

Rosetta Quantum publishes verdicts with raw, reproducible data. This is educational content, not a product claim.