Ask three serious people when quantum computers will be useful and you will hear three dates. IBM says 2029. The French State's programme, which raised its target on 22 May 2026 from 128 to 1000 logical qubits, says 2032. The OECD says a far larger machine is plausible by 2040, and adds, in brackets, that it is not guaranteed. A reader who lines those dates up on one axis will conclude that the experts disagree by eleven years. That reader would be wrong, and the reason is the subject of this piece: the three dates do not measure the same thing. One is a company's delivery promise, one is a government purchasing target, one is an outlook on the size of machines. Most of the loud arguments inside quantum computing have this shape. They sound like disagreements about the future, and turn out to be disagreements about what is being counted.
This piece belongs to the quantum chapter of the CareerOn Industry Atlas, and two earlier pieces set its ground. The first followed the money and found the French State acting as planner and buyer. The second followed the hardware underneath and found three layers France does not yet fully supply. This one takes the three questions every candidate in the field is asked to have a view on: which kind of machine is ahead, when a useful advantage arrives, and whether error correction is still a scientific question or now an engineering one. We print one dated document per side, we score nothing, and we say where a disagreement dissolves once the terms are read.
The three great arguments inside quantum computing are narrower than they sound: no dated document says a hardware route has won and the French State has built its procurement around that uncertainty, the competing dates for a useful machine measure a company promise, a public target and an outlook on scale rather than one milestone, and no admitted document says error correction cannot work, only how much engineering remains; for a candidate, the useful skill is reading which claim a document is making, and the work sits where the remaining engineering is.
We read eight dated documents. Five were already in the series and were read again for what they argue, not what they fund: the Cour des comptes audit of July 2026, the PROQCIMA page of the national quantum programme as updated after the acceleration of 22 May 2026, the OECD map of the global quantum ecosystem of December 2025, the OECD policy primer of January 2025, and the European Quantum Flagship's research and industry agenda of November 2022. Three are new to the series. The first is the paper in which Google's quantum team and its collaborators report error correction below threshold, posted on 24 August 2024 and later published in Nature. The second is John Preskill's paper on the noisy intermediate-scale era, posted on 2 January 2018, still the reference text for the cautious side. The third is IBM's press release of 10 June 2025 announcing a large fault-tolerant machine. We label that last document for what it is: a company describing its own plan. It is cited, never weighed as if it were an audit or a peer-reviewed result.
We refused two documents. IBM's roadmap page is a living product page with no edition date; the dated release says the same thing and is admitted instead. The US National Academies' report on progress and prospects is often quoted on timing, but the page we could reach is a catalogue entry and our probe could not read the text behind it. A sentence we cannot check against its document is not printed.
Which machine is ahead
There are five main ways to build a qubit, and every one has a champion. Superconducting circuits are fast and well funded. Trapped ions and neutral atoms hold their state longer. Photonic machines work closer to room temperature. Spin qubits in silicon promise to borrow the chip industry's factories. Each camp publishes milestones that make its route sound like the one to bet on, and a candidate reading only the press releases would think the race had been called several times over.
The public documents do not call it. The OECD's map of December 2025 is the plainest: "Several very different approaches are competing to solve the same problems, and no single platform has emerged as the clear frontrunner." Its policy primer of January 2025 goes further and allows that none may succeed: "it remains unclear which one, if any, may ultimately lead to fault-tolerant quantum computers." Neither is a cautious footnote. Both are the institution's summary of the field, written for governments deciding where to spend.
The French State has turned that uncertainty into a method. The Cour des comptes describes PROQCIMA as a competition between five companies, selected for their capacity to produce a fault-tolerant machine, and notes that the strategy opened the way to several qubit technologies, "présentant chacune des avantages et inconvénients." The programme's own page states the intent without ambiguity: it organises a competition to select progressively, on objective criteria, the technologies that will lead to large-scale quantum computing. The choice of route is not missing from French policy. It is deferred on purpose, and the procurement is designed to make it later, on evidence.
No document says a route has won, and the State plans for that
| What the document prints | Stance | Publisher, date |
|---|---|---|
| No single platform has emerged as the clear frontrunner | Open field | OECD, December 2025 |
| It remains unclear which platform, if any, leads to fault tolerance | Open field | OECD, January 2025 |
| A competition between five companies for a fault-tolerant machine | Several routes funded | Cour des comptes, July 2026 |
| Technologies selected progressively, on objective criteria | Choice deferred by design | France 2030 quantum programme, May 2026 |
Sources 2 Organisation for Economic Co-operation and Development · 3 Organisation for Economic Co-operation and Development · 1 Cour des comptes · 5 France 2030
The counter-case deserves a hearing. A strategy that backs several routes can also be read as a strategy that cannot choose, spreading money thin while better-funded American programmes concentrate on one. That is a fair argument about budgets, and the first piece of this chapter covered it. It is not an argument that a route has won. No admitted document makes that claim, and a reader who meets it in a pitch deck should ask which document it rests on.
When the machine becomes useful
The timing argument is where the arithmetic goes wrong most often, so we set the dates side by side and say what each one dates. IBM's release says: "Delivered by 2029, IBM Quantum Starling will be built in a new IBM Quantum Data Center in Poughkeepsie, New York." The same release says the machine will run 100 million operations on 200 logical qubits. That is a promise about one company's machine, made by the company. The French programme raised its own target on 22 May 2026, from 128 to 1000 logical qubits by 2032. That is a purchasing objective, the level the State intends to buy at. The Cour des comptes, in July 2026, had summarised the programme as aiming for a fault-tolerant machine by 2030; the programme's page now prints 2032. We print both as their publishers printed them and do not reconcile them. The OECD map says it is "plausible (but not guaranteed) that by 2040 quantum computers with around one million physical qubits could exist." That is an outlook on hardware scale, counted in physical qubits, not logical ones.
Four dates, and none of them measures the same thing
| Date | What is dated | Kind of claim | Publisher, date |
|---|---|---|---|
| 2029 | A machine running 200 logical qubits | Company claim | IBM, 10 June 2025 |
| 2030 | A fault-tolerant machine, as the auditor summarised the programme | Audit summary of a target | Cour des comptes, July 2026 |
| 2032 | 1000 logical qubits | Procurement target, raised in May 2026 | France 2030 quantum programme, May 2026 |
| 2040 | Around one million physical qubits, plausible but not guaranteed | Outlook | OECD, December 2025 |
Sources 8 IBM · 5 France 2030 · 1 Cour des comptes · 2 Organisation for Economic Co-operation and Development
Read across the table rather than down it and the eleven-year disagreement disappears. A logical qubit is built from many physical ones, so 200 logical qubits and a million physical qubits are not two estimates of the same machine. A delivery promise and a procurement target are not two forecasts of the same event. None of the four dates is a forecast of useful advantage at all, which is the thing a candidate actually wants to know.
On that question the documents are more modest than the headlines. The European Flagship's agenda separates two milestones that are often merged: it calls a demonstrated advantage in simulation "an important milestone, but not an application as such." Preskill, writing in January 2018, drew the line that still organises the cautious side: "We may feel confident that quantum technology will have a substantial impact on society in the decades ahead, but we cannot be nearly so confident about the commercial potential of quantum technology in the near term, say the next five to ten years." His window has now largely passed, and we print his view as the view of its date, not as a statement about today. What has not changed is the distinction he drew between impact over decades and commercial value within a few years. None of the admitted documents dates the second.
The counter-case here is strong and we state it at full strength. A company that commits to a dated machine in public, with a named site in New York, is taking a risk an outlook never takes, and IBM's earlier roadmaps have mostly been met on qubit counts. A candidate should take the 2029 date seriously as a statement of intent. What the date cannot do is stand in for an audited result or for a forecast of commercial value. It is a different kind of document, and the table prints it as one.
Is error correction science or engineering
The third argument is the deepest, because everything else depends on it. Physical qubits make errors far too often to run long calculations. Error correction spreads one logical qubit over many physical ones so that errors can be detected and undone, and the scheme only works if the physical error rate is below a threshold. For years the open question was whether real hardware could get below it. The cautious view of 2018 was blunt: Preskill described fault-tolerant quantum computing as "still a rather distant dream."
Google's paper of August 2024 is the document that moved the argument. Its abstract is short: "In this work, we realize surface codes operating below threshold on two superconducting processors." The logical errors fell as the code grew, which is the behaviour the theory predicts and the experiment had to show. It is one team, on one route, on processors the same team built, and the paper is honest about the distance left: "Despite realizing below-threshold surface codes, orders of magnitude remain between present logical error rates and the requirements for practical quantum computation."
Set beside that, the OECD primer of January 2025 still describes fault-tolerant machines as "the theorised large-scale and stable devices capable of reliably performing quantum computations over extended periods." IBM's release, a company announcing its own plan, describes a path to such a machine and gives it a date. Read together, the documents do not split into believers and sceptics. No admitted document says error correction cannot work. They disagree on distance: a lab result with orders of magnitude to go, a policy summary that still calls the machine theorised, and a company that says it knows the route.
Error correction: everyone agrees it can work, not on how far it is
| What the document prints | Distance it implies | Publisher, date |
|---|---|---|
| Surface codes operating below threshold on two superconducting processors; orders of magnitude still to go | Shown in the lab, far from use | Google Quantum AI, 24 August 2024 |
| The fault-tolerant machine is still described as theorised | Not yet built | OECD, January 2025 |
| Fault-tolerant computing still a rather distant dream | Far, at the time of writing | John Preskill, 2 January 2018 |
| A path to a large-scale, fault-tolerant machine | Company claim of a dated plan | IBM, 10 June 2025 |
Sources 6 Google · 3 Organisation for Economic Co-operation and Development · 7 John Preskill, California Institute of Technology (arXiv preprint, later published in Quantum) · 8 IBM
The counter-case is that "orders of magnitude" may hide a scientific surprise, a new error source that appears only at scale, and that calling the rest engineering is premature. That is possible, and it is the reason we reframed our own starting hypothesis rather than declare a winner. But no admitted document names such an obstacle. Until one does, the honest description of the field is that the science of error correction has passed its first experimental test on one route and the engineering of scale has barely begun.
What the three arguments have in common
In each case the loud version of the argument mixes two kinds of document. A route is said to have won because a company says so. A date is said to be the date because it is the earliest one printed. Error correction is said to be solved, or hopeless, depending on which sentence of the same paper is quoted. The quiet version, the one the dated public documents support, is narrower and more useful. The routes are open and the State has planned for that. The dates measure different things. Error correction works in the lab and is far from use.
That narrowness is not a lack of conviction. It is what a careful reader can defend, and it is the position we will hold until a dated document gives us a reason to move. Three things would change it: an audit or public funder stating that one route is preferred, a dated forecast of commercial advantage from an institution with no machine to sell, or an experiment on a second route showing errors falling below threshold. Any of the three would be reported here with a recorded correction.
What this means for your career
Every quantum interview asks for a view on these arguments, and most candidates answer with a route and a date. A stronger answer names the kind of claim. Saying that IBM promises 200 logical qubits in 2029 while the French programme buys toward 1000 logical qubits in 2032, and that the two are not the same milestone, shows an interviewer that you can read a roadmap the way a procurement officer or an investor does. That skill is rarer than knowing the physics, and it transfers to every deep technology field.
The second lesson is where the work is. If error correction has passed its first test and the distance left is engineering, the roles that grow are the ones that close it: qubit design and characterisation, the control systems that run correction in real time, and the integration that makes a thousand logical qubits more than a slide. Because the routes are open, a candidate who learns the engineering of one route while reading the others is better placed than one who bets a career on a single press release.
The simulation Superconducting qubit engineer: designing and characterising a cat qubit is placed in this piece, against the paper that ran error correction on superconducting processors, which is the route the simulation rehearses. One honest caveat: that paper uses surface codes, not cat qubits. Cat qubits are a different answer to the same problem, designed so that one kind of error is suppressed in the hardware itself, and the simulation lets you practise that design. The simulation Cryogenic electronics engineer: integrated quantum control at 4 K was placed in the second piece; none of the eight documents in this one names control electronics, so we print it as unplaced here. We have no simulation yet for error correction decoding or for systems integration, and both are recorded as gaps in our catalogue.
One question to ask a prospective employer, made possible by this piece: when you say your machine will be useful, which milestone do you mean, and which document dates it? The answer tells you how the company thinks, and whether its plan is a promise, a target or an outlook.
What we could not prove
We found no admitted document that dates a commercially useful advantage, and no dated forecast from an institution without a machine to sell. We could not read the National Academies' report against its text, so its often-quoted timing view is not printed. We found no experiment on a route other than superconducting that reports error correction below threshold among our admitted documents; such results may exist, and if a dated paper is admitted we will add it and say what it changes. The gap between the Cour des comptes' 2030 and the programme's 2032 is printed, not explained; we have not found a document that states why the summary and the page differ.
The next piece of the chapter turns from the arguments to the people: which quantum jobs exist in France today, which are announced, and which a candidate can rehearse on CareerOn now.