Oct 27, 2026 

Article

Most quantum computing headlines are about qubits: more of them, faster ones, cleaner ones. This month's most important result is about something less glamorous. It's a resource so costly to make that, by some estimates, it could eat up to 95% of a future quantum computer's capacity. Google says it just found a much cheaper way to produce it.

What happened

On October 15, Google Quantum AI published a paper in Science titled "Magic state cultivation on a superconducting quantum processor," with 296 authors from Google and partner institutions. The team showed that a technique called magic state cultivation works on real hardware, its Willow chip. It cut errors about 40-fold compared with uncultivated states and reached roughly 99.99% fidelity.

Why "magic states" matter

An error-corrected quantum computer can do some operations easily and reliably. Others, which useful algorithms can't do without, are much harder. Magic states are specially prepared quantum states that let the machine perform those hard operations without breaking its error correction. The catch: each one is used once, and they're fragile and noisy to make.

The traditional fix, called distillation, combines many noisy magic states into fewer good ones. It works, but it's hugely expensive. Cultivation, proposed by Google researchers Craig Gidney, Noah Shutty and Cody Jones in 2024, takes a different route. It grows a state inside a small error-correcting code, checks it repeatedly, and throws away attempts that fail. Think of it as a strict quality inspector on the assembly line rather than a recycling plant at the end.

On Willow, the team prepared states in a small "color code" and then grafted the purified result into a larger surface code, the format used for broader computation. The filter is strict: ScienceAlert reports the researchers kept only about 8% of attempts in the most demanding runs.

What it doesn't prove

Google is clear that cultivation still has to work at a larger scale before it's useful for practical computing. PostQuantum's analysis adds an important number: the paper's upper bound on error is still 1,000 to 2,000 times higher than the error rate Google's own 2025 resource estimates assume for cultivated states. In plain terms, the method works, and it's still far from the quality that code-breaking scenarios require.

The rest of the race

It's been a busy month across different hardware approaches. On October 19, Quantinuum announced a logical fidelity of 99.9995% on its Helios trapped-ion computer, which it calls a world record and a step toward its 2029 fault-tolerance target. On October 25, RIKEN researchers reported controlling five silicon qubits on a shared control line at around 99.9% accuracy, a step toward simpler wiring for large processors. Each result advances a different design, and none settles which approach wins.

What builders can do now

1. Join the Kaggle challenge. Google Quantum AI is running a $150,000 competition from October 22 to January 20, 2027, using AI agents to compress fault-tolerant quantum circuits. It's a rare chance for software people to move hardware progress forward.
2. Start a post-quantum inventory anyway. No one needs to panic. But NIST approved its first three post-quantum cryptography standards, FIPS 203, 204 and 205, back in 2024. Knowing where your company relies on vulnerable encryption is cheap insurance.
3. Ignore the countdown clocks. Track error rates and logical qubit counts instead. Those are the numbers that actually move timelines.