Can quantum computers predict the weather?
No. As of August 2026, no quantum computer has ever run a weather forecast, and there is no evidence that current or near-term quantum hardware would help. The algorithms that might apply exist mostly on paper, with serious problems getting data in and out. And the atmosphere's chaos limits every computer, quantum ones included.
Paper only — with severe caveats Quantum solvers for linear systems (HHL and later versions) promise exponential speedups. But that only holds if loading the input is cheap and you never need the full solution back. A forecast needs the full solution back. Reading out a solution stored in amplitudes takes roughly as many measurements as there are numbers in it. This is the readout problem, and it erases the exponential win. Also, the atmosphere's governing equations (Navier-Stokes) are nonlinear. They must first be forced into linear form, which raises resource counts further. No start-to-finish advantage has been shown, even on paper, for a realistic forecast workload.
Toy fluid solves exist — no speedup Researchers have solved tiny benchmark versions of the incompressible Navier-Stokes equations (lid-driven cavity flows) on real noisy hardware, using hybrid quantum-classical schemes. This is proof that the plumbing works, not capability. A laptop solves the same problems instantly. The grids are microscopic next to the ~10^9 grid points of a real forecast model. And no speedup exists at any size.
Quantum ML for weather explored — no advantage shown Researchers linked to the Met Office have studied quantum scientific machine learning for weather modelling. They fitted small quantum circuits with adjustable settings (parameterised circuits) to simple atmospheric data. The work is careful and openly exploratory. It shows the approach can work at toy scale. It makes no claim of advantage over the classical ML models that already run in daily forecasting.
No operational use — exploratory research only Forecasting centres such as ECMWF are watching the field. ECMWF's centre of excellence with Atos covers HPC (supercomputing), AI and quantum computing. Its researchers say openly that it is not yet clear whether quantum computing can help at all. No credible organisation has set a date for operational quantum forecasting. Meanwhile, classical machine-learning models have delivered the decade's real forecasting gains, without a single qubit.
Chaos limits any computer — quantum included Tiny errors in the measured starting state of the atmosphere grow exponentially. This is the Lorenz butterfly effect. It is why forecasts that give one exact answer get worse after about two weeks, no matter how much computing power you use. A quantum computer takes in the same imperfect satellite and station observations as a classical one. It cannot out-compute that sensitivity to starting conditions. More computing of any kind buys sharper probability estimates and a slightly longer horizon, not certainty about rain in week three.
Why is weather such a bad fit for quantum computers?
Why care? Better forecasts save lives in storms. They help farmers, airlines and power grids plan. So people hope quantum computers will make forecasts better. But there are three separate walls in the way. Each one alone would stop it today.
Wall 1: the physics doesn't match. A quantum computer naturally runs linear quantum math. Linear means doubling the input doubles the output. Its amplitudes, the numbers that say how strongly it leans toward each result, rotate and interfere like waves. The atmosphere is different. It is a classical, turbulent, strongly nonlinear fluid, where small pushes can have huge effects. Think of stirring cream into coffee: the swirls feed back on themselves in messy ways. You can force nonlinear equations into a linear quantum form on paper, using tricks like Carleman linearisation. But that multiplies the resources needed. Often it goes past the point where the speedup survives.
Wall 2: getting data in. A forecast model starts from roughly a billion numbers describing today's atmosphere. Loading ordinary data into quantum amplitudes is expensive. The neat speedup proofs usually assume this step is free. It is not.
Wall 3: getting data out. A quantum solver does not hand you the answer. It hands you a quantum state whose amplitudes hold the answer. Each measurement gives you just one random sample. It is like learning a painting by being shown one random dot at a time. Unlike the painting, each look also uses up the state, so you must rebuild it before the next look. To rebuild a full weather map you need about as many measurements, or shots, as there are numbers in the map. Work the numbers: a billion numbers means on the order of a billion shots. That quietly erases the exponential speedup you came for. This is honest amplitude arithmetic, not pessimism. The measurement lesson explains why.
Wouldn't a quantum computer at least beat chaos?
No, because chaos is not a computing problem. Chaos means tiny differences at the start grow into huge differences later. People call this the butterfly effect. The two-week forecast limit comes from the atmosphere blowing up tiny errors in the starting state, fast. Those errors come from imperfect observations. Weather stations are spread thin, and satellites are noisy. No processor, quantum or classical, can compute its way past not knowing today's atmosphere exactly.
Here is a toy example of how fast errors grow. These are made-up round numbers, not measurements. Suppose a small error doubles every two days. Two weeks is 14 days, which is 7 doublings. So the error grows 2 × 2 × 2 × 2 × 2 × 2 × 2 = 2^7 = 128 times. A faster computer does not change the doubling. Only better starting data does.
What more computing does buy is better probabilities. That means larger sets of slightly different forecasts, called ensembles. It also means finer grids and sharper storms. That is exactly where the last decade's real revolution happened. Ordinary machine-learning models now rival traditional simulation at a fraction of the cost. So the honest comparison is not "quantum versus today's forecasts". It is "quantum versus fast-improving classical machine learning", and quantum is not on the scoreboard yet.
What would have to change for the answer to change?
Two things. First, fault-tolerant machines, which fix their own errors as they run, with enough logical qubits to run deep linear-algebra circuits. A logical qubit is a reliable qubit built from many error-prone ones. Company roadmaps put the first such systems around 2029-2030. Even those would be far smaller than solving the atmosphere's equations (PDEs) demands. Second, start-to-finish algorithms whose advantage survives after counting data loading and readout, not just the middle of the job. Neither exists today, even on paper, for a realistic forecast.
If those arrive, the likely entry points are narrow. One is linear-algebra pieces inside data assimilation, the step that blends fresh observations into the model. Another is sampling for random small-scale physics that the grid is too coarse to show (stochastic sub-grid physics). In every believable scenario, quantum hardware speeds up pieces of the pipeline. "A quantum computer predicting the weather" is not how any serious proposal reads. Track how real devices are doing on the QPU index. And see the quantum advantage assessment for how far proven capability really goes.
Classifications follow the QPU137 editorial policy: every applied label carries a date, source, scale, and hardware, or it does not render. Found an error? Report it.