Every quantum computing roadmap rests on a single, quietly demanding piece of math. It says that error correction can drive a machine's failure rate as low as you like, but only if the underlying hardware is already good enough to begin with. Cross that line and adding more qubits makes things better. Fall short of it and adding more qubits just adds more ways to fail. This is the threshold theorem, and it is the reason engineers obsess over gate fidelities that already sound impressive.
Why more qubits can hurt
Quantum error correction works by spreading the information of one protected logical qubit across many physical qubits, then constantly checking for signs of trouble. The catch is that the checking machinery is itself built from noisy qubits and noisy gates. Every extra component you add to watch for errors is another component that can introduce one.
That sets up a tug of war. Error correction removes mistakes, but the operations that perform the correction create fresh mistakes. Whether you come out ahead depends entirely on how reliable each individual operation is. If your gates and measurements are clean enough, correction wins and the logical error rate shrinks as you pour in more physical qubits. If they are too sloppy, the correction machinery generates errors faster than it can catch them, and bigger codes perform worse than smaller ones.
The number that decides everything
The threshold is the fidelity level that separates those two regimes. Above the threshold, the payoff compounds: a modest improvement in code size buys an exponential drop in logical errors. Below it, the whole scheme is upside down. There is no single universal value. The threshold depends on the error-correcting code you choose, the noise model your hardware suffers from, and the way you schedule the correction circuit. For the popular surface code, the commonly quoted figure sits near one percent, meaning physical operations must fail less than roughly one time in a hundred.
That sounds forgiving until you remember it applies to every operation in the correction cycle, including the notoriously error-prone measurement and reset steps. It also assumes errors are reasonably independent. Correlated bursts, like a cosmic ray dumping energy across a whole chip at once, can violate the assumptions the theorem relies on. So hardware teams do not aim to just barely clear the threshold. They aim to sit far below it, because the further below you are, the fewer physical qubits each logical qubit needs.
Distance, and the price of safety
The lever engineers pull is code distance, roughly a measure of how many things must go wrong simultaneously before a logical error slips through. Increasing the distance means using more physical qubits per logical qubit. When you are comfortably under the threshold, each step up in distance suppresses the logical error rate by a large factor. That is the exponential magic that makes fault tolerance believable.
It is also why the physical-to-logical ratio is so brutal. To reach the error rates needed for something like breaking encryption or simulating a complex molecule, estimates often call for hundreds or even a thousand physical qubits per logical qubit. The closer your hardware creeps toward the threshold from the wrong direction, the steeper that ratio climbs, because you need larger codes just to make any progress at all.
Crossing the line
For years the threshold was a theorem without a demonstration. The milestone the field has been chasing is a break-even experiment, where a larger code visibly outperforms a smaller one on the same hardware. That is the fingerprint of being below threshold: scale up the code and the logical qubit lives longer, not shorter. Reaching that point does not mean a useful machine exists yet, but it confirms the ladder is pointing the right way.
This is what gives the theorem its outsized influence on strategy. A company touting a new qubit type, a cleaner gate, or a faster measurement is usually making the same underlying argument: we are moving further below threshold, so our error correction will cost fewer qubits and scale faster. The threshold theorem, in other words, is not just a proof about the possibility of fault tolerance. It is the scoreboard that tells you whether a given machine has any hope of getting better as it gets bigger.
