Ask a quantum engineer why the field bothers with error correction at all, and eventually the conversation lands on a number. Not a qubit count, not a company valuation, but a probability: the fraction of operations a machine is allowed to get wrong. That number has a name, the fault-tolerance threshold, and it is the quiet hinge on which the entire enterprise turns.
What the threshold actually promises
Quantum error correction works by spreading the information of one clean "logical" qubit across many noisy physical ones. The idea sounds circular. If every physical qubit is faulty, and you use faulty qubits to detect and fix faults, why doesn't the error correction machinery introduce more errors than it removes?
The threshold theorem, proved in the late 1990s by several groups working independently, answers exactly that worry. It says there exists a critical physical error rate. As long as your components sit below it, you can suppress logical errors to any level you like by adding more physical qubits and more layers of encoding. Push the physical error rate down a little, or scale the code up a little, and the logical error rate falls dramatically. Above the threshold, the opposite happens. Every additional qubit you throw at the problem introduces more noise than it corrects, and the logical error rate climbs. Bigger machines get worse, not better.
That single fact reframes the whole race. It means quantum computing is not about reaching some heroic level of perfection in a single qubit. It is about clearing a bar, and then letting scale do the work.
Why the number is slippery
There is no universal threshold value etched into physics. It depends on which error-correcting code you use, how the qubits are wired together, what noise looks like, and how much you trust the classical decoder that interprets error signals. Early estimates for general codes were brutally demanding, hovering near one error in ten thousand operations. That looked hopeless given the hardware of the era.
The surface code changed the mood. Because it needs only nearest-neighbor connections on a two-dimensional grid, it fits the layout of real superconducting and neutral-atom chips, and it tolerates a comparatively generous error rate of roughly one percent per operation. Suddenly the target looked reachable rather than fantastical. Trapped-ion and neutral-atom systems, with their higher gate fidelities and flexible connectivity, can aim at even friendlier codes.
The catch is that hitting the threshold is necessary but not sufficient. Sitting just below it buys you almost nothing, because the overhead explodes. To get a meaningful reduction in logical errors near the boundary, you might need thousands of physical qubits per logical qubit. The further below threshold your hardware operates, the cheaper each logical qubit becomes. That is why fidelity improvements matter so much: a factor-of-two drop in physical error rate can slash the qubit budget for a useful machine.
Crossing the line
For years the field lived in a frustrating limbo. Chips were good enough to demonstrate the pieces of error correction but not good enough to show it actually helping. The signature everyone wanted was simple to state and hard to achieve: build a logical qubit that lives longer than its best individual physical qubit, and show that making the code larger extends that lifetime further. That is the fingerprint of operating below threshold.
In recent years, experiments on superconducting and neutral-atom platforms have started to show exactly that behavior. Growing the code shrinks the logical error rate, which is the qualitative proof that the machines have crossed to the right side of the line. It is a modest result in absolute terms and a profound one in principle. It confirms that the theorem is not just a blackboard abstraction, and that the path from noisy prototypes to fault-tolerant computers is an engineering slog rather than a search for new physics.
Why it dominates every roadmap
Once you understand the threshold, the industry's obsessions make sense. The relentless focus on two-qubit gate fidelity, the fights over benchmark methods, the investment in faster classical decoders, all trace back to putting distance between the hardware and that critical line. A company that reports 99.9 percent gate fidelity instead of 99 percent is not bragging about a rounding error. It is describing an order-of-magnitude cut in the number of physical qubits it will need to build anything useful.
The threshold theorem is the reason the field believes a million-qubit machine could ever compute anything reliable. It turns an impossible demand for perfection into a demand for good-enough, repeated at scale. Everything else is bookkeeping on the size of the bill.
