Supremacy vs useful advantage
Two phrases get mixed up constantly, so let us pin them down. Quantum supremacy means a quantum computer performed *some* task faster than any classical computer could — even if that task is artificial and useless. It is a physics milestone: proof that the hardware can reach a regime classical machines struggle to follow. Useful quantum advantage is a much higher bar: solving a problem people actually want solved faster, cheaper, or better than the best classical method. The first has arguably been demonstrated for a contrived task. The second has not been clearly achieved for anything practical, and that gap is the whole story of this guide.
An analogy: imagine a machine that can shuffle a deck of cards in a uniquely strange way no human or computer can predict or reproduce. That is genuinely hard and genuinely novel — but nobody needed that particular shuffle. Demonstrating you *can* do the impossible-to-mimic thing is supremacy. Doing something a bank, a chemist, or a logistics planner would pay for is advantage. They are different claims, and conflating them is the single most common way quantum news misleads you.
The sampling experiments
The headline demonstrations were sampling experiments. The idea is clever: build a quantum circuit of random gates, run it, and measure the output. Because the qubits' amplitudes interfere in a complicated way, the distribution of measured bitstrings is extremely hard for a classical computer to reproduce. You are not asking the machine to compute a useful answer — you are asking it to produce samples from a distribution that is itself defined by the quantum hardware. The whole point is that classically simulating that distribution is believed to take impractically long.
Google's 2019 Sycamore experiment used 53 superconducting qubits and claimed a task that would take a classical supercomputer thousands of years took about 200 seconds. Later, photonic experiments (Jiuzhang) and larger superconducting runs made similar claims using a related task called boson sampling. These are real, careful experiments on real NISQ-era hardware — noisy machines with no error correction. That last point matters: the results lean on the *details* of the noise, not on a clean, fault-tolerant computation.
What they did and didn't prove
What they genuinely showed: a programmable quantum device can be driven into a regime where its behavior is hard to mimic classically, and where the quantum nature of the computation — superposition and interference across many qubits — is doing real work. That is a meaningful engineering and physics result. It is evidence that scaling these machines is worth pursuing.
What they did not show: any useful computation, any Shor-style threat to cryptography, or even a permanent classical-to-quantum gap. The classical pushback has been fierce and effective. After Sycamore, improved classical algorithms and clever use of large compute (and tensor-network methods) dramatically cut the estimated simulation time — in some analyses from millennia down to days or less. The honest summary is a *moving target*: each quantum claim invites a classical counterattack, and the boundary keeps shifting.
Where useful advantage might come first
If contrived sampling is not useful, where might *real* advantage appear? The most credible early candidate is quantum simulation — using a controllable quantum system to study another quantum system, like molecules and materials. This is the original idea behind quantum computing, and it plays to the machine's natural strengths: simulating quantum chemistry on a classical computer is exactly the kind of problem that scales brutally. Near-term approaches like VQE and QAOA try to extract value from today's noisy hardware, though so far they have not decisively beaten the best classical methods.
Be careful with the other commonly cited 'use cases.' Grover's algorithm gives only a quadratic speedup (roughly √N instead of N) for unstructured search — helpful in theory, but easily erased by the overhead of running a quantum computer, so it is rarely a slam-dunk. The dramatic *exponential* speedups are confined to special structured problems like factoring via Shor's algorithm, and those need large, fault-tolerant machines that do not exist yet. Most optimization and machine-learning 'quantum advantage' claims you will see are speculative or unproven.
A skeptic's checklist
You do not need a physics degree to pressure-test a quantum-advantage headline. You just need to ask a few sharp questions. The point is not to be cynical — quantum computing is real and progressing — but to separate a genuine result from marketing, and a contrived benchmark from a problem anyone actually has.
- Is the task useful, or contrived? A faster way to sample a random circuit is not the same as solving a problem someone would pay for. Ask what real question got answered.
- What is it being compared against? The claim should beat the *best known* classical algorithm on *good* classical hardware — not a naive baseline chosen to look slow.
- Quadratic or exponential? A √N (quadratic) speedup is often eaten by overhead. Only structured, exponential speedups (Shor-style) are clearly transformative — and they need hardware we don't have yet.
- Does it need fault tolerance — and do they have it? If the claim quietly assumes large numbers of error-corrected logical qubits, it is a projection about the future, not a result from today's NISQ machines.
- Has it survived classical pushback? A 'classically impossible' result is a bet. Check whether improved classical algorithms have since closed the gap — they often do.