Blog ·

Oct 20, 2026 · 7 min read

Are quantum computers real yet?

A grid of white dots seen at an angle, each sized by the chance of one result of a random quantum circuit. Most dots are small and a scattered few are large.

Every few months, a headline announces that a quantum computer has finished in minutes a calculation that would keep a supercomputer busy for thousands of years, or longer than the universe has existed. And yet nobody you know has used one, no company has said it found a new material or saved money thanks to one, and your bank still runs on ordinary computers. The machines are real, and they do run programs, so what have they actually done?

The honest answer is that they have set real records and passed one milestone that matters a great deal, but no quantum computer has yet solved a problem someone needed solved, faster or cheaper than an ordinary computer could. The records were set on tasks picked because quantum computers are good at them, and the results are easiest to follow in order, starting with the ones that made the biggest headlines.

The first records were set on tasks designed to be useless

The first goal was simply to prove that a quantum computer could do any calculation at all that an ordinary computer can’t. For that kind of proof, usefulness doesn’t matter. What matters is picking a task where the gap between the two kinds of computer is as wide as possible.

The task researchers settled on works like this. The quantum computer runs a random jumble of steps on its qubits, then reads them all, which gives a long string of 0s and 1s. Run it again and you get a different string, but the strings follow a pattern set by the jumble, a bit like the results of rolling a set of strangely weighted dice. For a quantum computer, rolling the dice is easy. To imitate it, an ordinary computer first has to work out how the dice are weighted, and with dozens of qubits that takes an enormous amount of calculation. The answer is no use to anyone, which is the point: the task was chosen because it is easy for a quantum computer and very hard to imitate.

  • Google, 2019. Its machine did this with 53 qubits in about 200 seconds. Google estimated that a supercomputer would need 10,000 years. Within three years, other researchers found a better ordinary method that did the same job in about 15 hours. That is still far longer than 200 seconds, but a very long way from 10,000 years.
  • Google, December 2024. A newer chip did a larger version with 105 qubits in under five minutes. Google’s estimate for the fastest supercomputer was far longer than the age of the universe. Ordinary methods have improved since, but nowhere near enough to close that gap.
  • IBM and the University of Chicago, July 2026. They ran a similar task on qubits protected by error correction (more on that below), and built it so the answer could be checked during the run.

If no ordinary computer can finish the calculation, how does anyone know the quantum computer got it right? For the earlier records, nobody could check the full-size result directly. The researchers checked smaller and simplified versions of the same task, ones an ordinary computer could still work out, using a statistical test of whether the quantum computer’s strings followed the pattern they should. The machines passed, and the researchers took that as evidence that the full-size runs, which were the same task made bigger, were working too. The 2026 test by IBM and the University of Chicago was built so the full-size answer itself could be checked, which makes it much harder to argue with.

Taken together, these records show that a quantum computer can finish a calculation no other computer can match. They don’t show it can do anything you’d want done. So the next question researchers asked was whether a quantum computer could beat ordinary computers at a task someone actually cares about, and the first place they looked was physics.

Physics simulations came next, and most of them were challenged

Quantum computers were first proposed as a way to simulate nature. The particles inside materials follow the same strange rules that qubits do, so a machine made of qubits should be naturally suited to imitating them. Simulating how a material behaves is still research, but unlike rolling weighted dice, it tells scientists something they want to know.

  • D-Wave, March 2025. D-Wave builds machines designed for one kind of physics problem, rather than for running general programs. Its machine simulated how magnetic materials respond when conditions change, and the company said a supercomputer would need nearly a million years to do the same. Within days, other researchers reproduced key parts of the result on ordinary computers. D-Wave says those researchers only reproduced the easier cases, and that the hardest cases in its study are still out of reach for ordinary computers.
  • Google, October 2025. Its chip measured how a disturbance spreads through 105 qubits and then comes back, like an echo. It did this about 13,000 times faster than the best known method on one of the world’s largest supercomputers. Unlike the earlier speed records, the result is a number that another quantum computer can reproduce and check.
  • IBM, August 2026. IBM announced three more physics simulations that it says ordinary methods can’t match. All three use simplified textbook models of materials rather than real ones, and other researchers have challenged parts of the claim.

Each time, a company announces a quantum result, other researchers go looking for a faster ordinary method, they find one for at least part of the problem, and the gap narrows. The results that have held up are the ones designed so the answer can be checked. Until a claim can be checked, it’s safest to read it as a press release.

Quantum computers have started to fix their own mistakes

All of these results run into the same limit: qubits make a lot of mistakes. Every step a quantum computer takes is slightly off, and over a long program the small errors pile up, like a message passed along a game of telephone. (Our post on why quantum computers make so many mistakes goes through where the errors come from.)

The planned fix is error correction. Instead of trusting one qubit, the machine spreads one qubit’s worth of information across many qubits and keeps checking them against each other, so that a mistake in one can be spotted and undone. But every extra qubit can make mistakes of its own. If each qubit is too error-prone, adding more of them brings in more mistakes than the checking removes, and the bigger code does worse than the small one. Error correction only starts to help once each qubit is good enough on its own.

In December 2024, Google showed that its chip had passed that point. Each time it made its code a step bigger, the mistakes roughly halved. This is the most important result on the list, even though it made fewer headlines than the speed records, because every route to a useful quantum computer depends on error correction working at much larger sizes than this. The error correction lesson explains how it works.

A quantum computer has made random numbers that come with proof

There is one result that comes close to being useful in its own right. In March 2025, JPMorgan Chase, the company Quantinuum and US government laboratories used a 56-qubit machine to produce about 70,000 random bits.

Random numbers are easy to come by, so the interesting part is the proof. Anyone can hand you a list of numbers and say they’re random, and you have to take their word for it. These came with a way to check: two supercomputers tested them and confirmed that only a real quantum computer could have produced them. Numbers that nobody could have predicted or rigged are valuable for lotteries and for encryption, where a predictable number is a weakness.

The researchers said the margin isn’t yet wide enough for high-stakes uses. Even so, it is the closest any quantum computer has come to making something a customer could want.

Our own tests show how quickly mistakes pile up

We run small programs on a real quantum computer every month, and the results match everything above. Programs with only a few steps come out almost perfectly. As programs get longer, the game of telephone takes over. A search through 16 possible codes, which took 161 steps, found the right code less than half the time. Factoring 15, the job of finding the two numbers that multiply to give it (3 × 5 = 15), took 1,266 steps when done the long way, and the machine did no better than guessing.

What quantum computers haven’t done yet

That gap between short and long programs is why the list of things that haven’t happened is still longer than the list of things that have.

  • Breaking encryption. Much of today’s encryption relies on how hard it is to factor very large numbers, and Shor’s algorithm would do that on a big enough quantum computer. So far, no quantum computer has even factored 21 (3 × 7 = 21) with Shor’s algorithm without a shortcut that builds the answer into the program. The Q-Day page follows how that gap is closing.
  • Useful chemistry. No quantum computer has calculated anything about a real molecule better than an ordinary computer can. The medicines post covers what has been tried.
  • Faster planning and scheduling. Delivery routes, timetables and investment choices are often named as uses, but there’s no proof that quantum computers are faster at them, and no case where one has beaten good ordinary software.
  • Money from solved problems. The companies selling quantum computers earn most of their money from research contracts and governments. For now, the surest customer for a quantum computer is someone who wants to study quantum computers. The earnings post has the figures.

None of this means progress has stalled. Since 2024, the machines have improved faster than many skeptics expected, especially at fixing their own mistakes. When the first useful result does arrive, it will most likely be a physics or chemistry simulation whose answer can be checked, because that’s where every lasting result so far has been.

Sources

Read next