Industries · Energy & utilities
Energy & utilities
Written for
Transmission and distribution planning, system operations and asset management at a TSO, DSO or utility.
The decision
Which reinforcement to commit capital to, which units to commit, how the network behaves when a line trips.
How it is done today
MILP and heuristic planning tools, with contingency analysis run case by case.
When it goes wrong A plan that is not shown optimal is hard to defend to a regulator; a missed contingency is an outage.
Applications
| Application | You give | You get | Status |
|---|---|---|---|
| Transmission expansion planning Proof Optimum proven over 1,048,576 plans; a reduced-order encoding came out 9% worse on the same photonic QPU · Read the case | A MATPOWER-style network and the candidate lines | The plan, optimum proven, with the full-model vs reduced-model gap | Live on quantum hardware |
| Contingency and cascading-failure analysis Proof Network response simulated to 8 time steps on 156 qubits; the classical simulator we ran stopped at 6 · Read the case | The network | Which assets fail together and where the response concentrates | Live on quantum hardware |
| Unit commitment and economic dispatch Proof Unit commitment at 25,755 variables and 48,939 constraints ran feasibly on a photonic QPU · Read the case | Generators, periods and constraints | A feasible dispatch, benchmarked against your MILP | Deployed through the platform network |
| Distribution-grid reconfiguration Proof 69-bus feeder reconfiguration through the platform network · Read the case | The feeder network | A switching plan | Deployed through the platform network |
| Renewable integration and storage siting | The network and the candidate sites | Sites ranked under contingency | Mapped |
| Battery, catalyst and capture chemistry | Candidate materials | A ranking with the smallest difference it resolves | Ready · computed, hardware next |
What changes for you
- The plan is checked against every alternative, not sampled.
- The cost of the simplified model your tool uses is measured, not assumed.
- The network’s real response is simulated past where the classical reference stops.
- These findings do not establish a commercial quantum benefit.
First engagement
You send One expansion case, as a network file with in-service status honoured.
You get back The proven plan and the gap. Two weeks.
Proof · Transmission expansion planning
Research result
1,048,576 plans · optimum proven
the chosen expansion plan proven optimal over every one of 2^20 candidate build plans
Baseline Exhaustive classical search over all plans
Platform CPU search; network response on a 156-qubit processor and a photonic QPU
Decision
Which network reinforcement plan to commit capital to
What is modelled
20 candidate reinforcements, so 2^20 build plans, each with the network loading it produces.
Operational outcome
- A proven-optimal plan, not a sampled one
- The cost of a reduced-order model quantified: 9% worse on the same QPU
- Network response simulated to 8 time steps on 156 qubits
Problem size 20 binary investment decisions · 1,048,576 plans · 156-qubit network response · Scope These findings do not establish a commercial quantum benefit.
Benchmark
Published reference Unit commitment at 25,755 variables and 48,939 constraints ran feasibly on a photonic QPU, with a commercial MILP solver ahead at short runtimes; generation scheduling at 26 generators × 24 periods on 36 trapped-ion qubits, no optimality proof. arXiv:2505.00145; platform-network results, 2025–26
Merlin Optimum proven over 1,048,576 plans · reduced-order model 9% worse on the same QPU
Baseline Exhaustive classical search; the full model on the same machine
Further results in this industry
- On one photonic QPU in one session, a reduced-order (degree-two) encoding of a transmission-expansion case came out 9% worse than the full model.
- A 156-qubit network-response simulation ran to 8 time steps on the processor; the classical simulator we ran stopped at 6.
Case studies
Energy · networksSimulation · Quantum hardware
A network-expansion plan proven optimal over every one of more than a million alternatives, plus the network’s real-time response simulated on quantum hardware.
Energy · utilitiesQuantum hardware
The response of a 156-node network simulated to eight time steps on a 156-qubit processor; the classical simulator we ran stopped at six. For a planner, that is the contingency case read further than the reference tool reaches.
Energy · utilitiesPlatform network · Quantum hardware
A unit-commitment problem of 25,755 variables and 48,939 constraints solved feasibly with a photonic quantum processor inside a hybrid MILP decomposition, on a national-laboratory testbed. The commercial MILP solver was better at short runtimes; the hybrid was competitive at longer ones. These findings do not establish a commercial quantum benefit.
Energy · utilitiesPlatform network · Quantum hardware
Switch reconfiguration on a 69-bus distribution network run on a superconducting quantum processor through our platform network. The optimum was found more than seven times as often as with the unoptimised quantum workflow; the baseline is a quantum workflow, not a classical solver. The scale at which this would beat classical methods, about 230 buses, is a projection and has not been shown.
Other industries
Semiconductors & advanced materials · Chemicals & process manufacturing · Mobility & automotive · Telecom · Pharma & life sciences · Finance · Security & critical infrastructure · Aerospace & sovereign programmes