Quantum annealing
Also known as: quantum annealer, adiabatic quantum computing, D-Wave, QUBO
Quantum annealing is a specialized type of quantum computing designed to find low-energy solutions to optimization problems, such as scheduling or routing, by letting a quantum system settle into its best configuration.
It's different from 'gate-based' quantum computing used by Google and IBM. While D-Wave's annealers have been commercially available for more than a decade, clear evidence of an advantage over the best classical methods on practical problems has remained hotly debated.
Quantum annealing, explained at three levels
At home
Imagine a ball rolling across a bumpy landscape of hills and valleys, trying to reach the lowest valley. A normal ball can get stuck in a small dip. Quantum annealing builds that landscape out of a problem, so the lowest point is the best answer, and lets a quantum system settle into it, sometimes tunnelling through hills instead of climbing over them. It's designed for finding the best arrangement among many options.
At work
Quantum annealing is a specialized form of quantum computing for optimization problems, such as scheduling, routing, or portfolio selection. It's sold mainly by D-Wave, founded in British Columbia, through cloud access. Unlike gate-based quantum computers, it can't run general quantum algorithms. Businesses care because it's one of the few quantum tools available at large scale today, though classical solvers often match its results.
At home
Using an annealer means rewriting your problem as a QUBO, short for quadratic unconstrained binary optimization. Every decision becomes a yes-or-no variable, and costs and rules become weights between pairs of variables, so the best solution has the lowest total energy. The machine's qubits represent those variables, and the couplers between them represent the weights. The annealer then settles toward a low-energy state, and reading the qubits gives a candidate solution. Users typically run it many times and keep the best answer.
At work
D-Wave's Advantage systems have thousands of qubits, but each connects to only some others, so a large problem must be embedded, with several physical qubits chained together to act as one variable. That shrinks the problem size the hardware can handle directly. Hybrid solvers address this by splitting problems, sending pieces to the quantum processor and doing the rest classically. Annealers aren't error-corrected and their qubits are noisier than gate-based ones, so result quality varies from run to run.
At home
Annealing begins with the qubits in a simple state with an easy energy landscape, then slowly transforms the landscape into the one encoding your problem. The adiabatic theorem says that if the change is slow enough, the system stays in its lowest-energy state throughout, ending at the best answer. The catch is that 'slow enough' depends on the narrowest energy gap along the way. In my opinion, that gap is the whole challenge, because for hard problems it can make the required time impractically long.
At work
Benchmarking annealers is contentious because classical competitors keep improving: simulated annealing, parallel tempering, tensor networks, and specialized heuristics. A fair evaluation compares time to a target solution quality against the best classical solver on the same problem, including time spent translating and embedding it. D-Wave's 2025 Science paper claimed beyond-classical results for simulating magnetic materials, though some groups challenged it with classical methods. Simulating physical systems may be the most promising use, since the hardware mimics them directly.
Real-world examples
At home
Jerome read that a Canadian company sold quantum computers to NASA and Lockheed Martin. He learned that D-Wave made quantum annealers, a specialized type of quantum computer aimed at optimization problems.
At work
Aisha leads operations research at a logistics company. Her team tested a quantum annealing service for delivery routing. The hybrid solver produced good routes, but not better than their existing classical optimization software, so they kept it as an experimental tool.
How it compares
| feature | Quantum annealing | Gate-based quantum computing |
|---|---|---|
| Purpose | Optimization and simulation | General-purpose quantum algorithms |
| Qubit counts | Thousands | Dozens to around a thousand |
| Error correction | Not designed for full error correction | Path to fault tolerance |
| Can run Shor's algorithm | No | Yes, at sufficient scale |
| Key players | D-Wave | Google, IBM, Quantinuum, IonQ |
Annealers are specialized optimizers; gate-based machines aim to be general-purpose quantum computers.
For analysts
If you're exploring quantum annealing, benchmark it against your best classical solver, not your current spreadsheet. I'd only scale a pilot that beats the strongest classical method on cost, speed, or quality.
History & origin
In 1998, physicists Tadashi Kadowaki and Hidetoshi Nishimori formally proposed quantum annealing, building on earlier ideas. D-Wave Systems, founded in 1999 and based in the Vancouver area of British Columbia, sold the first commercial quantum annealer, the D-Wave One, to Lockheed Martin in 2011. Google and NASA bought later systems. Academic studies over the 2010s questioned whether the machines provided quantum speed-ups. D-Wave's Advantage system launched in 2020 with more than 5,000 qubits, and Advantage2 became generally available in 2025. That same year, D-Wave published a study in Science claiming beyond-classical performance for simulating magnetic materials, which some researchers challenged with classical methods. In my view, quantum annealing's story shows how hard it is to prove an advantage, even with real hardware in customers' hands.