Pick a year and the quantum computing field will argue about it. For a long stretch the safe answer, the one researchers gave when they didn't want to overpromise, was sometime in the 2030s. Now a date is being floated that sounds almost reckless by comparison: 2028. That's the new target some are putting on useful quantum error correction, and it's close enough to feel real.

Whether you find that thrilling or suspicious probably depends on how many quantum timelines you've watched slip already.

Plenty have.

Why error correction is the whole ballgame

Here's the problem the field has been circling for years. Quantum bits, or qubits, are absurdly fragile. A stray vibration, a flicker of heat, a faint magnetic nudge from somewhere in the room: any of it can knock a qubit off its value before it finishes a calculation. The technical word is decoherence. The plain word is noise.

Classical computers don't sweat this. A bit is a one or a zero, and if a cosmic ray flips it, error-checking circuits catch the mistake and move on. Quantum information doesn't sit still so politely. You can't just copy a qubit to make a backup, because measuring it changes it. That's not an engineering inconvenience. It's physics.

So for decades the answer has been quantum error correction, or QEC. Spread the information that would normally live in one qubit across many physical qubits, arranged so the system can detect and fix errors without ever directly reading the data it's protecting. The cluster of physical qubits doing that job is called a logical qubit. One reliable logical qubit might take dozens, hundreds, even thousands of physical ones, depending on the hardware and how noisy it is.

That ratio has been the great humbling fact of the whole enterprise. You don't get a useful machine by adding qubits. You get one by adding qubits faster than they break.

What "useful" is actually doing in that sentence

Notice the word "useful." It's carrying a lot.

Researchers have already shown error correction working in the lab. Small demonstrations: a logical qubit that holds its state a bit longer than the raw physical qubits underneath it, proof that the math survives contact with real hardware. Those are genuine milestones. They're also a long way from a machine that does anything you'd pay for.

The word implies the correction scales. It implies you can chain operations together, run a real algorithm, and trust the answer at the end. It implies the error rate keeps dropping as you grow the system rather than swamping you with new failure modes. That last part is the threshold theorem in action, the idea that below a certain physical error rate, adding more qubits makes the logical qubit cleaner instead of dirtier. Cross that line and the curve bends in your favor. Stay above it and you're shoveling sand.

My read, for what it's worth: the gap between correcting an error in a demo and correcting errors well enough to compute something nobody could compute otherwise is wider than most press releases let on. A 2028 promise is really a bet that the gap closes in roughly three years. Bold. Not impossible.

The case for taking 2028 seriously

It would be easy to wave this off as another optimistic slide in another investor deck. But a few things have genuinely shifted, and they're worth laying out.

The physical qubits are getting quieter. Error rates that looked stubborn a few years ago have come down across several hardware approaches, superconducting circuits and trapped ions among them. Quieter qubits mean fewer of them per logical qubit, which means the whole machine gets cheaper and smaller to build.

The correction codes themselves have improved. The surface code, long the workhorse, demands a punishing number of physical qubits. Newer code families promise to do the same protective job with a friendlier overhead. If those hold up outside simulation, the qubit budget for a useful machine shrinks, possibly by a lot.

And the money is enormous. The biggest names in computing have poured serious resources into dedicated quantum hardware teams, and several startups have raised rounds that would have been unthinkable for the field a decade ago. Money doesn't bend physics. It does buy a lot of attempts, a lot of fabrication runs, a lot of smart people trying the next idea before the last one cools.

Put those together and a near-term target stops sounding absurd. It starts sounding like a stretch goal that somebody, somewhere, genuinely thinks they can hit. Which is different from a sure thing, and everyone involved knows it.

Why the skeptics aren't wrong either

Now the cold water.

Quantum timelines have a personality, and it's the kind that texts you "on my way" and shows up an hour late. Goals that felt imminent have a way of receding the closer you get, because each solved problem reveals two you hadn't budgeted for. The field calls this progress. Outsiders sometimes call it something less polite.

There's also a real difference between a logical qubit that survives and a logical qubit you can compute with. Holding a quantum state still is hard. Performing gates on it (error-corrected gates, in sequence, without the corrections themselves introducing new errors) is harder. Some of the trickiest operations in fault-tolerant computing remain expensive and fiddly. A 2028 machine that can store information beautifully but struggles to act on it isn't the useful machine anyone's picturing.

And "useful" invites a quiet sleight of hand. Useful for whom, doing what? A device that beats classical computers on one narrow, carefully chosen problem can be technically remarkable and commercially beside the point. The applications people actually want (cracking hard chemistry, modeling new materials, certain optimization problems) tend to demand far more clean logical qubits than the first fault-tolerant machines will have. So you could very plausibly get a 2028 system that clears the scientific bar and still leaves the killer-app question wide open.

None of that makes the target dishonest. It makes it conditional. The condition being that a stack of difficult things go right at roughly the same time, on roughly the same schedule, which is exactly the sort of coordination reality enjoys disrupting.

What this would actually change

Suppose it lands. Suppose, somewhere in 2028 or close to it, a machine runs a meaningful error-corrected computation and the result is checkable and correct. What follows?

Not a sudden new world. The first useful fault-tolerant quantum computers will be rare, expensive, and rationed, more like the room-sized mainframes of the 1960s than anything you'd keep on a desk. Access will come through the cloud, metered and queued, with a handful of operators deciding who gets time. Early users will be researchers and a few deep-pocketed companies with very specific problems, not the general public.

The louder consequence might be psychological. A working error-corrected machine, even a modest one, settles a long-running argument about whether fault-tolerant quantum computing is achievable at all, as opposed to perpetually thirty years away. That confidence reshapes funding, recruiting, and how seriously adjacent fields take the whole project. Belief is a resource here, and a credible demonstration mints a lot of it.

There's a sharper edge too. Cryptography. A large enough fault-tolerant quantum computer could eventually break the encryption that guards much of the internet's traffic and stored secrets. The first useful machines almost certainly won't be big enough for that, and security agencies have been pushing organizations toward post-quantum encryption for years already. But every step toward real error correction tightens that clock, and it's worth keeping an eye on who's quietly accelerating their migration plans. Their urgency is its own kind of forecast.

What to watch between now and then

If you want to track whether 2028 is on or slipping, ignore the qubit-count headlines. Raw qubit totals are the field's favorite vanity metric, and they tell you almost nothing about reliability.

Watch the logical error rate instead, and watch whether it keeps falling as systems grow. That's the curve that decides everything (the threshold theorem again, doing its quiet work). Watch for demonstrations where a logical qubit clearly and repeatably outperforms its physical parts, not by a whisker but by a margin that scales. Watch the gate operations, the verbs of computing, and whether the error-corrected versions become practical rather than just theoretically possible.

And watch the language carefully. There's a difference between a team saying it has done something and a team saying it expects to by 2028. The first is reporting. The second is a promise, and promises in this field have a long history of arriving late but, often enough, eventually arriving.

Three years isn't long. We'll know soon whether 2028 was ambition or accuracy.