Google Quantum AI said Thursday that a technique called magic state cultivation cut errors roughly 40-fold on its Willow chip, producing magic states with a record 99.99% fidelity. The company says it has turned one of quantum computing's most promising paper tricks into something that actually works on a chip, with the results published in Science on Thursday.

Magic states are single-use resources needed to run useful quantum programs, and Google says some estimates put the cost of making them at up to 95% of a quantum computer's capacity.

Making quantum computers cheaper to build matters far beyond physics labs. It brings forward the day the machines can do useful, such as designing drugs and new materials—and, equally, the day one could break the encryption that protects bank logins and Bitcoin wallets.

Why magic states matter

Quantum computers run on qubits—quantum bits that can hold a blend of 0 and 1 at once, letting them explore many answers in parallel. They are also absurdly fragile: a little heat or electrical noise can flip a qubit and wreck a calculation, which is why the whole industry is obsessed with error correction, spreading one piece of information across many qubits so mistakes can be caught and fixed. Google's Willow chip, unveiled in December 2024, demonstrated that principle on real hardware: as the error-correcting code scales up, error rates fall rather than rise.

Error correction handles the simple operations well, but a genuinely useful program also needs a class of harder operations that cannot be protected the same way. To run them, the machine first has to manufacture magic states, which work like single-use fuel cells for those operations: each one is consumed the moment it is used.

Producing clean magic states has always been brutally expensive. Some estimates put the bill at up to 95% of a quantum computer's entire capacity, according to Google, leaving very little room for the actual computing.

From distillation to cultivation

Physicists Sergei Bravyi and Alexei Kitaev laid out the standard fix in 2004 with magic state distillation, which takes many noisy magic states and boils them down into a few clean ones. It works, but it demands sprawling qubit "factories" that do nothing else.

In 2024, Google researchers proposed a cheaper route they named magic state cultivation. Instead of a factory, the technique behaves like a filter: a rough magic state is injected into a tiny patch of the chip, a chain of checks is run on it, every attempt with a detected flaw is thrown out, and the survivors are kept growing. Their paper went as far as suggesting further distillation "may never be needed in practice."

"On Google Quantum AI's Willow processor, our protocol achieved an extraordinary 40-fold reduction in errors relative to un-cultivated states. And we got T-states with a record logical fidelity of 99.99%," Google said in its announcement. T-states are the specific kind of magic state consumed to apply the T gate, one of the harder operations error correction cannot protect on its own.

Only about 8% of attempts passed every check, meaning the chip discards more than nine out of every 10 tries to keep one clean state.

Cryptographic stakes

Cultivation is one of the tricks behind Gidney's 2025 estimate that RSA-2048 encryption could be broken with under a million noisy qubits, a 20-fold drop from the 20 million he estimated in 2019.

Google kept tightening the screws in March, when a Quantum AI whitepaper co-authored with Ethereum Foundation researcher Justin Drake and Stanford cryptographer Dan Boneh put the cost of breaking Bitcoin's elliptic curve cryptography—the math that stops anyone from deriving a private key from a public address—at fewer than 500,000 physical qubits. That followed the company's Quantum Echoes experiment last October, the first verified quantum speedup on real hardware.

The defense is moving on its own schedule: NIST finalized its first post-quantum cryptography standards in August 2024—algorithms designed to withstand attacks from large future quantum machines—and has proposed disallowing quantum-vulnerable public-key schemes such as RSA and elliptic curve cryptography in federal systems after 2035.

Still far from breaking codes

Nobody is draining wallets with this. The experiment used at most 54 of Willow's 105 qubits, and the paper notes that factoring an RSA-2048 key would require magic states with errors around one in 10 billion—roughly a million times cleaner than what Google just achieved.

What's next

Google says the next step is running cultivation at a larger scale. The team projects that halving Willow's underlying hardware noise would push magic state errors down to about two in 100,000, because errors in the process shrink with the cube of the chip's noise level. The roughly 8% pass rate is the other number to watch: the more attempts that clear every check, the less of the machine's capacity goes to tries that end up discarded.