Why qubits are so fragile
On this page
If you've read Parts 1 to 4, you've seen the same caveat again and again: qubits are fragile. It's the reason today's machines can't yet run the famous algorithms, and it's the reason estimates for breaking encryption run to a million qubits rather than a few thousand. This piece explains what fragile actually means, and how engineers are fighting it.
What "fragile" means
A qubit is only useful while it holds its blend of 0 and 1, with its amplitudes intact. The trouble is that almost anything can disturb that blend. A little heat, a vibration, a stray electrical or magnetic field, even a single passing particle of light can interact with the qubit.
Each of those interactions acts like a tiny, accidental measurement. The outside world "learns" a bit about the qubit's state, and in doing so blurs it. The careful pattern of amplitudes leaks out into the surroundings and the qubit drifts towards behaving like an ordinary bit, or like random noise. Physicists call this process decoherence.
A good picture is a soap bubble. It's beautiful and does something remarkable, but touch it, breathe on it or simply wait, and it's gone. Quantum computing is the art of doing useful work inside the bubble before it pops.
How long do qubits last?
It depends on the type of qubit:
- Superconducting qubits (used by IBM and Google) keep their state for fractions of a millisecond. That's short, but their operations are also very fast, so a lot can happen in that window.
- Trapped ions and neutral atoms can hold their state for seconds or longer. Their operations are slower, so the trade-off is different rather than simply better.
To buy time, engineers isolate qubits as much as they can. Superconducting chips sit inside refrigerators that cool them to around a hundredth of a degree above absolute zero, far colder than outer space. Ions and atoms are held in an ultra-high vacuum, away from stray air molecules. Wiring, shielding and control signals are all designed to keep noise out.
Errors in every step
Decoherence isn't the only problem. Every operation the computer performs on a qubit, called a gate, is a little imperfect. The best hardware today gets a typical two-qubit operation right around 99.9% of the time. That sounds excellent until you do the arithmetic.
At 99.9%, you expect roughly one error per thousand operations. A useful run of Shor's algorithm needs billions of operations. Without some way to catch and fix mistakes, the errors would pile up and the final answer would be noise long before the calculation finished.
The gap between "one mistake per thousand steps" and "billions of steps" is the gap between today's machines and useful ones.
Why you can't just copy a qubit
Ordinary computers deal with errors easily. The simplest trick is to store three copies of each bit and take a majority vote: if one copy flips by accident, the other two outvote it.
Quantum computers can't do that, for two reasons. First, there's a rule of quantum physics, called the no-cloning theorem, that says you can't make an exact copy of an unknown qubit state. Second, you can't simply look at a qubit to check whether it's still correct, because looking is a measurement and destroys the blend.
For a while in the 1990s, some physicists suspected these rules made large quantum computers impossible.
Quantum error correction
The breakthrough was realising you can protect information without copying it or looking at it directly. Quantum error correction spreads one qubit's worth of information across many physical qubits in an entangled pattern. Extra helper qubits then check whether neighbouring qubits still agree with each other, without ever measuring the information itself. It's a bit like checking that the pieces of a puzzle still fit together without looking at the picture.
When a check fails, the pattern of failures points to which qubit went wrong and how, and the computer can correct it. The group of physical qubits working together this way is called a logical qubit. Logical qubits are the ones algorithms actually use.
An everyday picture helps. Imagine a precious message split across a team of proofreaders, each holding only a fragment and none able to read the whole thing. Every few seconds, neighbouring proofreaders compare notes on whether their fragments still line up. If two neighbours disagree, the team can work out who made the slip and fix it, and at no point does anyone read the message itself. Quantum error correction runs these checks constantly, millions of times a second, throughout a calculation. A fast ordinary computer sits alongside the quantum chip, reading the check results and deciding what to correct, which is a demanding engineering problem in its own right.
The catch: it takes a lot of qubits
Error correction is expensive. With the most studied approach, the surface code, each logical qubit can need hundreds to a thousand or more physical qubits, depending on how good the hardware is and how reliable the result must be. Newer codes promise to cut that overhead substantially, and companies including IBM are building their roadmaps around them.
There's a crucial condition, too. Error correction only helps if the physical qubits are already good enough, below a certain error rate called the threshold. Above the threshold, adding more qubits adds more errors than it fixes. Below it, bigger codes make the logical qubit steadily more reliable.
In 2024 Google reported a milestone on exactly this point. On its Willow chip, making the error-correcting code bigger made the logical qubit's error rate go down, which is the first clear sign that scaling up works as theory predicts. It was still far from a useful machine, but it was the kind of evidence the field had been waiting decades for.
Physical qubits versus logical qubits
This distinction is the single most useful thing to remember when you read quantum news.
- Physical qubits are the raw hardware. Today's largest machines have hundreds to a few thousand. Headlines usually quote this number.
- Logical qubits are error-corrected qubits built from many physical ones. Machines today have, at most, a handful working together.
Breaking today's encryption needs over a thousand logical qubits running for days, which is why a 2025 estimate put it at under a million physical qubits. A press release announcing "a 1,000-qubit chip" is announcing 1,000 physical qubits, which is a long way from 1,000 logical ones.
What this means for timelines
Fragility is the main reason nobody can give a confident date for a useful quantum computer, or for one that breaks encryption. Progress isn't one number going up; it's error rates falling, qubit counts rising and error-correction overheads shrinking, all at once. A surprise improvement in any one of them can pull the date closer.
That's exactly why security planners don't wait for certainty. Preparing takes years, and the engineering is moving in one direction.
- Any interaction with the outside world blurs a qubit's state. That's decoherence.
- Error correction combines many physical qubits into one reliable logical qubit.
- Headlines count physical qubits. Useful algorithms need logical ones, and lots of them.
Sources and further reading
Tags show what kind of source each one is. A standard or government guidance is an official document; a peer-reviewed paper has been checked by other experts; a preprint has not been peer-reviewed yet; an experiment reports a real-world demonstration; a company announcement is the company's own account. Dates and figures were checked against these sources on 11 October 2026. Spotted an error? Email hello@plainquantum.com and it will be corrected, with a note.
- Peer-reviewed paperQuantum Computing in the NISQ era and beyondJohn Preskill, Quantum, 2018
- Peer-reviewed paperScheme for reducing decoherence in quantum computer memoryPeter Shor, Physical Review A, 1995
- Peer-reviewed paperA single quantum cannot be clonedWootters and Zurek, Nature, 1982
- Peer-reviewed paperSurface codes: towards practical large-scale quantum computationFowler et al., Physical Review A, 2012
- ExperimentQuantum error correction below the surface code thresholdGoogle Quantum AI and collaborators, Nature, 2024
- PreprintHow to factor 2048 bit RSA integers with less than a million noisy qubitsCraig Gidney, 2025