Quantum computing quietly crossed the threshold it has chased since 1995

The number that mattered was never the qubit count

Qubit counts made good headlines because they are a single rising integer, which is what a headline wants. They were also close to meaningless on their own. A machine with a great many noisy qubits computes nothing useful, because the errors accumulate faster than the answer does.

The number that decides everything is the error rate of a two-qubit gate, compared against the threshold of whatever error-correcting code you are running. Below that threshold, adding more physical qubits makes the encoded logical qubit better. Above it, adding qubits makes things worse. The whole field is organised around which side of that line you are on, and for the surface code the line sits at roughly one percent.

1995: proving the errors could be corrected at all

For a while it was reasonable to think quantum error correction was impossible in principle. Classical error correction works by copying bits, and quantum states cannot be copied. Worse, you cannot look at a qubit to check it, because measurement destroys the superposition you were trying to protect.

Peter Shor's 1995 code showed the way out, with Andrew Steane arriving at a related construction independently: spread one logical qubit across several physical ones, then measure only the relationships between them. You learn that an error occurred, and which kind, without ever learning the state itself. The threshold theorems that followed in the late 1990s established the rest of the argument, that if the per-gate error is below some finite value, arbitrarily long computations are possible. That turned a philosophical objection into an engineering target, and the target took thirty years.

2026: the crossing

Two-qubit gate fidelity now runs from roughly 99.5 to 99.99 percent depending on the hardware platform, which puts error rates below one percent across the major modalities rather than in a single favoured one. The more meaningful result is what that enables downstream: a Nature paper this year reported logical error rates improving by factors of 11 to 800 against physical baselines on a trapped-ion processor, using a twelve-qubit code inspired by Knill and a sixteen-qubit tesseract colour code.

The hardware work continues alongside it. D-Wave published research in August on a dual-rail qubit gate reaching about 99.9 percent fidelity with gate times near 500 nanoseconds, with hardware-level error detection built in. The claim that matters in all of this is not any single fidelity figure but the direction of the curve: logical error rates falling as the system grows, which is the behaviour the threshold theorem predicted and the one the field needed to observe.

The sensors arrive before the computers

Quantum technology is already leaving the lab through a side door. On 27 August, Q-CTRL demonstrated GPS-free quantum-gravimetric maritime navigation in the Coral Sea, a system called Ironstone Opal that fixes position by reading the gravity field rather than by listening to satellites. That is a working deployment of quantum sensing, not a laboratory result, and it is useful precisely where satellite navigation is jammed or unavailable.

The computers remain further out, and the roadmaps say so honestly. Xanadu published a technical roadmap on 31 August targeting more than 1,000 logical qubits by 2031, with a quantum data centre in the 2029 to 2030 window. Five years is a long time, but it is the first era in which those dates are engineering schedules rather than expressions of hope.