Google's Willow Chip Makes Quantum Error Correction Actually Work
105 qubits, error rates that fall as the machine grows, and a benchmark measured in septillions of years.
On 9 December 2024, Google announced Willow, a quantum computing chip with 105 qubits, fabricated at the company's new facility in Santa Barbara. Two results came with the announcement. One produced the headlines; the other is the more consequential.
The headline result was a benchmark called random circuit sampling. Willow finished it in under five minutes. Google put the equivalent runtime for one of today's fastest supercomputers at ten septillion years — a one followed by twenty-five zeros. Hartmut Neven, who founded and leads Google Quantum AI, notes that the figure exceeds known timescales in physics and far exceeds the age of the universe. It is a striking number and a narrow one: by Google's own account, random circuit sampling has yet to demonstrate practical commercial applications.
The second result is less quotable and considerably more important. Google encoded information across square grids of physical qubits — three by three, then five by five, then seven by seven — and each time the grid grew, the error rate of the encoded qubit fell by half.
Why the halving matters more than the septillions
Qubits are fragile, and the approach to dealing with that dates to 1995, when Peter Shor introduced quantum error correction: spread one unit of logical information across many physical qubits, so the machine can detect damage and repair it. The catch is that every physical qubit you add is another component that can fail. As Google puts it, using more qubits normally produces more errors and the system slides back toward classical behaviour — which has meant that a bigger quantum computer was a less trustworthy one.
"Below threshold" is the name for the point where that reverses: the qubit count rises and the error rate falls anyway, so growth improves the machine instead of degrading it. Google's three-by-three to seven-by-seven sequence is that behaviour measured rather than argued, and the company calls it an outstanding challenge in the field since Shor's 1995 work.
Two supporting details are worth more than the benchmark. The corrected arrays had longer lifetimes than the individual physical qubits did, which the announcement calls "an unfakable sign that error correction is improving the system overall". And the corrections were applied while the computation ran — Google presents this as one of the first compelling examples of real-time error correction on a superconducting system, on the straightforward grounds that a repair arriving after the computation has finished is not a repair. The underlying hardware improved as well: coherence times approaching 100 microseconds, roughly five times the previous generation, with Google stressing that gates, qubit reset and readout all have to be well engineered and integrated at the same time.
What this means if you are building something
Nothing in your stack changes this month. Willow is a research chip. Google describes a first useful, beyond-classical computation relevant to a real-world application as the field's next challenge, not as something already delivered. Any vendor arriving in the next year to sell you quantum anything for a website, a shop or a booking system is selling a word rather than a capability.
The announcement says nothing about encryption. Google's post is about error correction and benchmarks; it makes no claim about cryptography, and nothing in it suggests any encryption you use today is affected. If you follow this field for security reasons, the thing to follow is the scaling direction rather than the qubit count or the septillion-year headline — and on 9 December 2024 that direction was reported as favourable for the first time, on a 105-qubit chip that is many generations away from anything practical.
Turn it into a procurement question, not a project. At the next renewal, ask your hosting provider, CDN, payment processor and VPN supplier one question in writing: what is your roadmap for post-quantum encryption, and when do you expect to support it by default? A vendor with an answer is a vendor paying attention. A vendor with no answer has told you something useful for the price of an email.
Know which encryption is actually yours. For most businesses, the traffic encryption on your site belongs to your host or CDN, and card data belongs to the processor — their roadmaps are your roadmap. What is genuinely yours is data you encrypt and store for the long term: contracts, medical or legal records, archives you will still hold in a decade.
Our reading: no urgency, but start the paper trail. The cheap moment to ask a vendor about post-quantum plans is while it is still an ordinary question.
Sources
