Google Quantum AI announced on October 15, 2026, that its Willow processor had demonstrated magic-state cultivation, a technique for preparing useful quantum states with fewer errors. The result advances one piece of future fault-tolerant computing; Google’s Bitcoin-related figures come from separate models of a hypothetical future machine. Google Quantum AI Science
How magic-state cultivation works
Google describes cultivation as a way to filter noisy input states through measurements. Attempts that fail error checks are discarded, leaving a smaller set of states for further computation. That makes the quality of the retained states—and the number of attempts that survive—important parts of the result.
What the Willow experiment measured
The Science study reports that cultivation reduced errors by a factor of 40 compared with states that were not cultivated. It also reports fidelity of at least 0.9999(1) for the retained states, while retaining 8% of attempts. The fidelity figure applies to that selected group, not to every attempt. Science
Google describes the result as 99.99% logical fidelity. The 40-fold figure concerns errors in cultivated states; it is not a measure of the total resources or cost of a Bitcoin attack. Google Quantum AI
What the code transfer and measurement work add
Google reports that the experiment transferred cultivated states from a distance-3 color code to a distance-5 surface code. The result therefore covers both state preparation and a transfer between two error-correcting code arrangements. Google Quantum AI
Google also says electrical drift and temperature variation affected a benchmarking assumption, prompting the team to develop a different measurement approach. The company says scaling cultivation to larger systems and addressing other technical challenges remain part of the work.
What Google’s Bitcoin estimates model
In a separate paper published March 31, 2026, Google Research estimated resources for circuits using Shor’s algorithm to solve ECDLP-256, the elliptic-curve problem relevant to Bitcoin signatures. One circuit uses fewer than 1,200 logical qubits and 90 million Toffoli gates; another uses fewer than 1,450 logical qubits and 70 million Toffoli gates. Under the paper’s stated superconducting-hardware assumptions, Google estimated that the circuits could run on a future machine with fewer than 500,000 physical qubits. Google Research
Those resource estimates describe a modeled future system, not Willow’s experimental setup. In a separate scenario reported by CoinDesk reporter Shaurya Malwa, a modeled attack after a public key became available took about nine minutes and required roughly 500,000 sufficiently reliable magic states per second. The scenario depends on its model and assumptions. CoinDesk
For Bitcoin, the practical challenge also involves migration. CoinDesk’s Shaurya Malwa reports that some older holdings have exposed public keys; moving coins to stronger protections requires the private keys, so owners who have lost them cannot make that transfer. CoinDesk
