Google Quantum AI said Thursday it has solved a problem that has slowed quantum computing research for two decades, demonstrating a technique called magic state cultivation on real hardware for the first time. The work, published in the journal Science, reduced errors in a critical quantum resource by roughly 40 times and produced what the company calls a record logical fidelity of 99.99% for states known as T-states.
"We've shown that magic state cultivation works on Google Quantum AI hardware," wrote research scientist Emma Rosenfeld and quantum software engineer Craig Gidney in a company blog post.
The experiment ran on Google's Willow processor, which the accompanying research paper describes as containing 105 superconducting qubits arranged in a square lattice.
Why magic states have been such an expensive problem
Quantum computers need more than just protected information to be useful. They also need to run a specific set of operations that ordinary error correction cannot handle on its own, and magic states provide the resource that makes those operations possible. The catch is that creating error-corrected magic states consumes enormous hardware capacity. According to Google's post, some estimates suggest producing sufficiently reliable magic states could eat up as much as 95% of a quantum computer's total capacity, leaving little room for actual computation. Adding to the difficulty, a magic state gets used up the moment it's consumed, so a computer running any lengthy calculation has to keep manufacturing replacements continuously.
Researchers have chipped away at this cost since 2004. Google's new approach, developed in 2024 by Gidney alongside colleagues Noah Shutty and Cody Jones, works like a filtering process. A noisy quantum state gets injected into a small region of the chip, then a coordinated sequence of measurements checks for errors and discards any flawed attempts, leaving only the purified result behind. This filtering step, known as postselection in the research literature, means the reported 99.99% fidelity applies only to attempts that pass the checks, so the proportion of attempts actually retained matters just as much as the final fidelity number when estimating the real-world cost of running this at scale.
A measurement problem that looked like broken physics
Getting the theory to work on hardware required solving an unexpected problem along the way. Rosenfeld described encountering data that initially seemed to defy the laws of physics while first testing the protocol. After months of simulations and consistency checks, the team traced the issue to their measurement technique itself, called logical state tomography, which combines measurements taken in different ways to estimate a quantum state.
The research paper explains that in the presence of certain noise, specifically low-frequency phase noise from sources like electrical equipment drift, residual magnetic fields and temperature variation, those different measurement approaches could retain different subsets of attempts. Combining them produced estimates that did not correspond to any physically possible state, and in some cases the same flaw could overstate fidelity without producing an obviously impossible result. The team built a new technique called kickback tomography to get around this, using the cultivation process itself to verify its own accuracy through a self-consistent measurement approach.
Calibration leaned on reinforcement learning, a form of machine learning that adjusts system controls based on feedback. For an experiment targeting an error rate of roughly one part in 10,000, training ran for about four hours immediately before data collection, and the added calibration step improved both the circuit's logical error probability and the proportion of retained attempts by roughly an order of magnitude, according to the paper's supplementary material.
Testing whether the purified states can actually be used
A result that only works in isolation wouldn't help build a real computer, so Google also tested whether these purified states could transfer into the error-correction scheme meant to support larger-scale computation. The team moved a cultivated state from what's called a distance-3 color code into a distance-5 surface code, a process they call grafting. Code distance refers to how many physical qubit errors must combine in a specific pattern before they cause an undetected logical error, so a higher-distance code offers stronger protection. The distance-5 grafting experiment involved 54 qubits, according to the paper's supplement.
The transfer worked, but a follow-up test exposed more work ahead. Researchers extended the number of error-correction cycles after grafting to see how well the purified state held up over time, and the observed logical error rate came in about seven times higher than simulations had predicted. The team attributed this mainly to leakage, where qubits drift outside the two states used to encode quantum information, a factor the simulations didn't account for and that the experimental circuits lacked dedicated operations to remove. Google says future work will focus on better calibration, leakage removal, simpler circuit arrangements and testing at larger scale.
Why this result matters beyond the physics lab
Google's March 2026 cryptography research estimated that a sufficiently powerful quantum computer could break the elliptic curve cryptography protecting Bitcoin using fewer than 500,000 physical qubits, a roughly 20-fold reduction from earlier resource estimates thanks to optimized quantum circuits designed to recover private keys from exposed public keys. Magic states are needed to support some of the operations in that kind of attack, which is part of why this week's cultivation result drew attention well outside the quantum physics research community, even though Willow's 105 qubits sit nowhere near the scale that cryptography study describes. Google's own framing is careful to separate the two results. This week's experiment focused on preparing and verifying quantum states, not attacking any cryptographic system, and the company has not said when its cultivation method might reach the scale needed for anything resembling a real cryptographic attack.
Google has previously urged organizations to begin preparing for post-quantum cryptography standards before capable machines exist, a message the company has repeated across its recent quantum security research. For now, the Science paper stands as a demonstration that a long-theorized fix for one of quantum computing's most resource-draining bottlenecks actually works outside of a whiteboard, with Google itself acknowledging that plenty of technical challenges remain before the approach supports anything resembling practical, large-scale quantum computation.
Gunel Ismayilova
Disclaimer: All materials on this site are for informational purposes only. None of the material should be interpreted as investment advice. Cryptocurrency investments involve significant risk of loss. Please note that, despite the nature of much of the material created and hosted on this website, HODL FM operates as a media and informational platform, not a provider of financial advisory services. The opinions of authors and other contributors are their own and should not be taken as financial advice. If you require advice, HODL FM strongly recommends contacting a qualified industry professional.
