We may earn a commission when you buy through links on this page. It doesn't change what we recommend.
Google says it has shrunk one of quantum computing’s costliest overheads, by about 40 times. The fix has an odd name: magic state cultivation. It was published October 15 in the journal Science, from Google’s Quantum AI team, on the 105-qubit Willow chip. Progress on an old bottleneck, yes. But narrow, too, and Google itself says much more work is needed before it matters outside a lab.
This explainer covers what magic state cultivation is, what Google actually showed, what the result doesn’t prove, and what it means for encryption and Alphabet’s stock.
What magic state cultivation is
Quantum computers make errors constantly, so useful ones will need error correction. Most operations can be protected fairly cheaply. A few can’t, including a type of gate called a T gate. Those need special ingredients called magic states. They’re hard to make accurately, and each one can be used only once. Google’s blog cites estimates, not measurements on Willow, that making them could eat up to 95 percent of a quantum computer’s capacity. That is the overhead cultivation is meant to shrink.
A rough analogy helps. Think of magic states as single-use parts on an assembly line. If most of the factory’s floor space goes to making those parts, little is left for building the product. Anything that makes the parts cheaper frees up room for the actual calculations, which is why a bottleneck like this draws so much attention.
The older fix was distillation, which uses dedicated qubit factories to turn many rough magic states into a few good ones. Magic state cultivation works more like a filter. Craig Gidney, Noah Shutty and Cody Jones of Google proposed it in 2024. A rough state is placed in a small patch of the chip, coordinated measurements check for errors, and any attempt that fails is thrown away. The survivors are cleaner, and they can then be expanded into a larger form for real computation.
What Google showed with magic state cultivation on Willow
Until now, cultivation had been studied mainly in theory and simulation. Google’s blog and Science paper report that it works on hardware. Errors fell about 40-fold compared with states prepared without cultivation, and the team reported a logical fidelity of 99.99 percent for T-states, a record. That is about one error in 10,000.
Getting there took work. The Quantum Insider’s report says the team used reinforcement learning as a final calibration step, with about four hours of training before data collection, and that it improved results by roughly an order of magnitude. Early measurements also looked wrong. Standard tests could overstate quality because of slow electrical drift and temperature noise, so the researchers built a new benchmarking method, called kickback tomography, that gives a conservative lower bound.
The team also moved purified states from one error-correcting code, a distance-3 color code, into a larger distance-5 surface code, in a setup involving 54 qubits. That step, which Google calls grafting, shows the states survive being handed from one layer of a machine to another.
What the magic state cultivation result does not show
Several caveats come from Google’s own paper and from the coverage of it. The 40-fold figure applies to this state-preparation task, not to Willow’s overall performance. The quality applies only to attempts that pass the checks, and Decrypt reports that only about 8 percent did. If so, over 90 percent of attempts are discarded, which is a real cost at scale.
After grafting, the logical error rate was about seven times higher than simulations predicted, according to The Quantum Insider. The likely cause is leakage, where a qubit drifts out of its intended states, and the circuits had no leakage-removal steps. Errors while storing the states were also far above simulations. Independent confirmation of the results is still pending, and the experiment used at most 54 of Willow’s 105 qubits.
For context, Decrypt notes that this follows the company’s “Quantum Echoes” result from October 2025, which it describes as a verified quantum speedup on real hardware. Google says cultivation must still be shown at larger scale. It offers no timeline for commercially useful machines. Decrypt says Google projects that halving Willow’s hardware noise would bring errors to about 2 in 100,000.
Encryption and Alphabet’s stock
Quantum computing news often raises one question: can it break encryption? Not yet. Decrypt notes that factoring a standard 2048-bit RSA key would need magic states with errors near 1 in 10 billion, roughly a million times cleaner than what was achieved. The same article says nobody can drain crypto wallets with this work. Estimates of the hardware needed have been falling, though, which is why security researchers keep watching. Software flaws are a nearer threat, as our report on XBOW’s six-day Chrome exploit showed.
For Alphabet, investingLive judges that any lasting effect on the share price is likely to be limited, since quantum is small next to search, cloud and AI. Smaller quantum-only stocks, it adds, tend to react more to headlines like this. That’s the publication’s analysis, not a forecast, and this article is not investment advice.
What to watch next
Watch for three things. The first is a larger demonstration, using more qubits and more rounds of error correction. The second is outside replication, since results like this carry more weight when another lab reproduces them. The third is how much of the overhead shrinks once leakage removal and better storage are added, because those are the gaps Google lists in its own results. Until then, treat a headline like this as a progress report, not a finish line.
Editor’s take
Magic state cultivation is a good example of how quantum progress usually looks: a specific, measured improvement on a specific bottleneck, with the caveats printed in the paper itself. It doesn’t mean useful quantum computers arrive next year. It does mean one of the biggest line items in the cost of building them just got smaller, on real hardware. Read headlines that say “cracked” with care, and trust the ones that tell you what’s still missing.
