Japan's plan to print quantum chips in a real fab is the credible kind of ambition. Then read the dates.

The idea at the center of the project Japan announced last week is genuinely beautiful, and I want to give it its due before I start asking about the calendar. A quantum bit, in this design, is not an exotic object suspended in a dilution refrigerator the size of a chandelier. It is the spin of a single electron, held in a speck of silicon a few atoms across, made on the same kind of machine that prints the processor in a laptop. If that works — and 'if' is carrying a great deal of weight in that sentence — then quantum computing would inherit the one manufacturing tradition on Earth that has reliably made things smaller, cheaper, and more numerous for sixty straight years. That is not a small prize, and the people chasing it are not naive. The reasons for hope are real. So are the reasons the useful version of this machine is further away than the roadmap's boldest number suggests.

Here is what was actually announced. Hitachi has been selected by NEDO, Japan's public research-funding body, to lead a government-backed project to build silicon quantum processors using Intel's 18A manufacturing process, in partnership with Intel's Japanese arm and AIST, the national industrial-science institute. The project covers qubit design, fabrication, packaging, and a cloud platform through which outside researchers could eventually run experiments on the hardware. Note the tense throughout: selected, to build, could eventually. What was announced is a funded program and a roadmap. Nothing was demonstrated on the day, and the distinction between what a field has demonstrated and what it has announced is, as it always is, the entire story.

The bet: that scale runs through the fab, not the lab

Underneath the announcement is a genuine scientific and industrial wager, and it is worth stating precisely because it is defensible. Most of the quantum computers you have read about — the superconducting machines from IBM and Google, the trapped-ion systems — are, in an important sense, hand-built. Each qubit is a comparatively large, carefully tuned physical structure, and scaling from dozens to millions of them runs into problems of size, wiring, and cooling that get harder, not easier, as the machine grows. The silicon-spin bet is that the way out of that trap is to stop hand-building and start manufacturing: to make qubits so small and so standard that the same extreme-ultraviolet lithography and process control that yields billions of identical transistors can yield billions of identical qubits. The claim is not that silicon spin qubits are ahead today. They are not. The claim is that they are the ones that can be mass-produced, and that mass production is what scale ultimately requires.

I find that argument serious, and I want to be clear about why, because skepticism about a timeline is not skepticism about a premise. The semiconductor industry is the most successful scaling story in the history of manufacturing. If any physical qubit can be made to ride that curve, the long-run economics change completely. Choosing Intel's 18A process — one of the most advanced nodes actually in high-volume production, with its gate-all-around transistors and backside power delivery — is a statement that this is meant to be a manufacturing program from the start, not a physics demo that someone will worry about manufacturing later. Intel has been here before, too: its research group built a twelve-qubit silicon spin device a few years ago specifically to prove these things could come off a standard 300-millimeter line. The lineage is real. The premise is sound. Now the physics.

What a silicon spin qubit actually is

Let me unpack the mechanism, because you cannot judge the timeline without it. In a silicon spin qubit, you trap a single electron in a tiny electrostatic well — a quantum dot — defined by voltages on metal gates sitting atop very pure silicon. The electron's spin, which you can picture loosely as pointing up or down, is the two-state system that stores the quantum information. You manipulate it with precisely timed microwave or electrical pulses, entangle it with a neighbouring electron by letting their wavefunctions briefly overlap, and read it out by a nearby sensor that can tell which way the spin points. The reason silicon specifically is that a spin in purified silicon is unusually well isolated from the noise of its surroundings, which buys you relatively long coherence — the time before the fragile quantum state smears into uselessness — and the structures are minute, which is what makes the dream of packing millions onto a chip physically conceivable.

All of that is demonstrated. Single silicon spin qubits work; two of them can be entangled; the best laboratories have shown the individual operations — the one- and two-qubit gates — running at fidelities above ninety-nine percent, which is the neighbourhood you need to even begin talking about error correction. This is real physics, done well, and it is why the field is taken seriously. But read that paragraph again and notice the numbers hiding in it: single, two, a handful. The demonstrations that justify the excitement live at the scale of a few qubits. The announcement asks you to think about a thousand.

The gap between a few and a thousand

That gap is not a matter of doing the same thing more times. It is where the unsolved problems live, and they are exactly the problems that manufacturing is supposed to solve and has not yet. The cruel irony of the silicon-spin approach is that the very thing it is betting on — fabrication — is also its hardest enemy at scale. Every quantum dot must behave almost identically to every other, and at the atomic scale the small variations that a fab shrugs off when it makes a transistor become the differences that make one qubit usable and its neighbour not. Uniformity and yield, in other words, are not a footnote to this program; they are the whole scientific problem, restated in the language of a production line.

And there are more. Every qubit needs its own control wiring, and threading thousands, then millions, of control lines into a chip kept a fraction of a degree above absolute zero is an interconnect problem no one has solved. The classical electronics that generate and time the pulses have to move closer to the cold chip without cooking it, which is its own research field. None of this is reason to dismiss the effort. It is reason to be honest that between a beautifully working six-qubit device and a thousand-qubit machine sits roughly a decade of engineering that is mostly unsolved, most of it in manufacturing rather than in the underlying physics. That is not a criticism. It is the actual shape of the work.

The thing this program bets on — fabrication — is also its hardest enemy. At the atomic scale, the tiny variations a fab shrugs off in a transistor are the difference between a usable qubit and a dead one. — On why uniformity is the whole problem

The number just past the money

Now the dates, which are where I earn my reputation for spoiling the mood. The roadmap has three markers. In fiscal 2027, an initial cloud service so outside researchers can run experiments on the hardware through AIST's quantum-computing center. In fiscal 2028, a prototype hundred-qubit silicon machine implementing quantum error-correcting codes. In fiscal 2030, a thousand-qubit-scale, three-dimensionally integrated processor. It is a clear, staged plan, and I credit it for being staged rather than gestured at. But hold the last two numbers up against a detail that appeared in none of the headlines: the funded project period runs to March 2029. The hundred-qubit prototype sits just inside the money. The thousand-qubit milestone — the number built to travel, the one in the headlines — sits a full year past the end of the current funding. It is not a commitment. It is an aspiration printed next to commitments, in the same font, and the reader is meant not to notice the seam.

I keep a private archive of technology's 'by [year]' promises and the years they were actually made, and quantum computing has been filling its own shelf for a decade now — the useful machine has spent most of that time staying a receding number of years away, in the same way fusion's grid connection does. So when a headline milestone lands just beyond the funded horizon, I read it not as a lie but as a tell: the honest part of the plan is the part someone has already agreed to pay for, and the exciting part is the part they have not. Watch the fiscal-2028 hundred-qubit prototype. That is the marker with money behind it, and hitting it on time would be a genuine achievement. The thousand-qubit line is a direction, not a date.

On what timescale — and to what end

There is a deeper reason not to let 'a thousand qubits by 2030' do the work the phrase wants to do, and it is the single most important thing to understand about this entire field. A thousand physical qubits is not a thousand useful ones. Because real qubits make errors, a fault-tolerant machine has to spend many physical qubits to protect a single reliable, error-corrected 'logical' qubit — with the most established error-correcting schemes, the ratio can run to roughly a thousand physical qubits for one logical qubit. Newer, more frugal codes could improve that dramatically, and I wrote here recently about a search that turned up promising ones — but those remain, for now, results in simulation rather than on hardware. Apply today's honest overhead, and a thousand-qubit silicon machine is a landmark scientific instrument and a superb research platform. It is not a code-breaker, and it is not a computer that will factor anything you care about. Even hitting every milestone on schedule delivers a proof of manufacturability, not a useful quantum computer.

So the timescale question, asked out loud the way I like to ask it: fault-tolerant, genuinely useful quantum computing is, by the sober reckoning of most people who work on it, a proposition for the 2030s and beyond, and that is true across every hardware platform, not just this one. Nothing announced last week changes that horizon. What the Japanese project could change — and this is the part worth being genuinely hopeful about — is whether, when the rest of the field is finally ready, there exists a manufacturing pathway capable of producing qubits by the million instead of by the dozen. That is a supporting-industry question, and supporting industries are built on exactly this kind of unglamorous, multi-year, government-funded tooling work: design kits, packaging, three-dimensional integration, a process a fab can actually run.

Which is why, in the end, I think the most credible thing about this announcement is also the least quotable. It is not really a promise to have a quantum computer by a certain year. It is a promise to figure out how to make the parts — to get yield and uniformity up on a real production line — and those are the right things to be working on, measured in the right units. The wrong question to carry away from it is 'will Japan have a working quantum computer by 2030.' The right question, the one the engineers themselves will be living inside, is quieter and harder: can a fab be made to print ten thousand identical qubits and have them all work. That is a manufacturing problem, on a manufacturing timescale, and the honest answer is that no one knows yet — which, coming from a field that has spent a decade being told the answer was almost here, is a refreshingly truthful place to start.