At nearly every technology event I attended in the last year, someone described quantum computing as either imminent or fraudulent. Both positions are wrong. The second is closer to right, which is an uncomfortable thing to say about a field I think is worth working in.

Here is the honest position, and where I would put money.

No commercial quantum advantage exists

There is no quantum computer today that solves a problem anybody pays to have solved, faster or cheaper than classical hardware. Not one paying workload, anywhere.

Not in drug discovery. Not in portfolio optimisation. Not in logistics. Not in materials, which is the application with the best theoretical case. The demonstrations that have made headlines are on tasks constructed to be hard for classical computers and natural for quantum ones, principally random circuit sampling, and they are legitimate physics results. Nobody has commercial demand for samples from a random circuit.

Several of those advantage claims have subsequently been narrowed or matched by improved classical algorithms, which is what a healthy adversarial process looks like and which should make anyone cautious about the next one.

The industry vocabulary has quietly adjusted. Roadmaps that said quantum advantage now say quantum utility, and the shift is a reasonable response to the fact that advantage was not arriving on schedule.

What has genuinely been achieved

The real progress is in error correction, and it is a substantial result that deserves more attention than the advantage claims.

Physical qubits are noisy at rates far above what a useful computation tolerates. Quantum error correction encodes one logical qubit across many physical ones. The threshold theorem says that below a certain physical error rate, adding more physical qubits per logical qubit reduces the logical error rate. Above it, adding qubits makes things worse.

Google’s Willow processor demonstrated operation below that threshold, published in Nature in 2024. Increasing the surface code distance from 3 to 5 to 7 suppressed the logical error rate by a factor of about 2.14 at each step, and the distance-7 code on 101 qubits achieved 0.143 percent error per cycle. The logical qubit’s memory lifetime exceeded that of the best physical qubit in the device by a factor of 2.4.

That is the crossing of a real line. It says the engineering path to fault tolerance is open rather than blocked by physics.

The field’s metric has moved accordingly, from physical qubit count to logical qubit count and stability. Anyone quoting a headline number of physical qubits in 2026 is quoting the wrong number.

The gap in numbers

The reason none of this yields a useful machine yet is overhead.

Resource estimates for cryptographically relevant factoring illustrate the scale. Gidney and Ekerå estimated in 2019 that breaking RSA-2048 would need on the order of 20 million noisy physical qubits running for around eight hours. Gidney’s revised 2025 analysis brought that under a million qubits, which is a twentyfold improvement in six years and is genuinely impressive.

Current devices are in the hundreds to low thousands of physical qubits.

So the gap is roughly three orders of magnitude in qubit count, alongside requirements on gate fidelity, connectivity, control electronics and cryogenic infrastructure that scale with it. Three orders of magnitude is not a detail. It is also not a wall, and the direction of travel on the estimates is favourable, because algorithmic improvement reduces the requirement at the same time as hardware increases the supply.

Reasonable projections put early fault-tolerant demonstrations with a small number of logical qubits around 2027 to 2028, scientifically significant and commercially useless, and production fault-tolerant machines somewhere in the 2030 to 2035 window, conditional on several things that have not happened yet.

I would treat the far end of that with the scepticism appropriate to any technology forecast a decade out.

Why the honest version still supports working on it

If I stopped there it would read as a case for ignoring the field. I do not think that follows, and the reasons are specific.

The physics is real. Quantum mechanics is the best-tested theory we have. Superposition and entanglement are experimental facts. The question is engineering at scale, not whether the effect exists. That is a different kind of risk from a field waiting on a scientific breakthrough that may never come.

Simulating quantum systems is the application that does not need to be argued for. Feynman’s original point in 1981 was that simulating quantum mechanics on classical hardware costs exponentially, and that a quantum system would not have that problem. This remains the cleanest case. Chemistry and materials are quantum problems, and classical approximations are approximations. Cracking this changes catalysis, batteries and drug design, and unlike most claimed applications it does not require a speedup argument, only a machine.

Error correction produced transferable knowledge. The theory of protecting fragile quantum information has connections to condensed matter and to holography that are interesting independent of whether a machine gets built. Topological order, anyons, the structure of entanglement in many-body states. Some of this touches the quantum gravity questions I work on, which is not the reason to fund a quantum computer but is a real intellectual return.

The architecture question is open and it is the interesting one. Superconducting, trapped ion, neutral atom, photonic and spin qubit approaches all have live claims and different failure modes. Which architecture reaches useful scale is genuinely undetermined, and unlike the hype question it is a technical question with technical content. That is where I would put effort.

How to read announcements

Ask what the logical qubit count is and what the logical error rate is. If the release quotes only physical qubits it is a marketing document.

Ask whether the demonstrated task has a classical customer. Random circuit sampling does not.

Ask whether the classical comparison used the best available classical algorithm on modern hardware, or a convenient baseline. This is where most advantage claims have eroded.

Ask what the error rate was and whether the result was postselected.

Treat any commercial timeline inside five years as a sales position rather than an engineering estimate.

The position I actually hold

There is a version of technological honesty that says a field is either delivering now or it is a fraud. That framing would have killed most things worth building, including the transistor, which took decades from the physics to anything commercial.

Quantum computing is a long research programme with an uncertain payoff, one clearly identified application that would justify the whole enterprise on its own, a hard engineering obstacle that has recently been shown to be passable in principle, and roughly a decade of work remaining before anyone should expect useful output.

That is a completely reasonable thing to work on. It is not a completely reasonable thing to sell as available today, and the amount of the latter is doing damage to the credibility of the former.

My allocation follows from that. Effort and capital behind the architecture question, which is technical, undetermined and decidable. Nothing behind any five-year commercial timeline, which is a sales document.

Related: