What Google's Willow Chip Actually Means

Cutting through the headlines on Google's latest quantum milestone — what's real, what's relevant, and what remains unsolved

In late 2024, Google announced its Willow quantum chip had solved a benchmark computation in five minutes that would take today's fastest supercomputer 10 septillion years. The internet reacted predictably. Here's what actually happened.

Willow is a 105-qubit superconducting processor, and its announcement made two claims. The first — the five-minutes-versus-10-septillion-years benchmark — got essentially all of the coverage. The second, about error correction, got a fraction of it. The second one is the reason physicists paid attention, and understanding why requires taking both claims apart.

What the benchmark actually was

The headline number comes from a task called random circuit sampling. The quantum computer runs circuits assembled essentially at random and produces samples from the resulting output distribution. That distribution is monstrously difficult for a classical computer to reproduce — which is the entire point. The task was designed as a demonstration: something a quantum processor does natively and a classical supercomputer can only imitate through brute force. It produces no useful answer. It breaks no encryption, prices no portfolio, discovers no molecule. It is a race quantum was built to win.

There's also a precedent worth remembering. When Google claimed 'quantum supremacy' with its Sycamore chip in 2019, the claim was that the benchmark would take a classical supercomputer 10,000 years. IBM responded within days arguing it could be done in about two and a half, and subsequent improvements in classical simulation shrank the gap dramatically. The lesson isn't that these results are fake — it's that classical computing is a moving target, and any 'X years for a supercomputer' figure should be read as 'hard today', not 'hard forever'. Willow's benchmark is far more robust than Sycamore's was, but the same caveat applies to the septillion.

The result that actually mattered

Errors are the central problem of quantum computing — arguably the only problem, with everything else downstream of it. Qubits are exquisitely fragile: heat, vibration, and stray electromagnetic noise all corrupt them, and even on the best hardware roughly one operation in a thousand goes wrong. Useful quantum algorithms require billions of operations. At one error per thousand, the computation dissolves into noise long before it finishes. The only known way out is quantum error correction: spreading one 'logical' qubit's worth of information across many physical qubits, so that errors in individual components can be detected and corrected without destroying the encoded information.

The leading approach, the surface code, arranges physical qubits in a grid; the size of that grid is called the code's distance, and a bigger distance means more errors can be caught. But here's the catch that stalled the field for two decades: a bigger grid also means more physical components, each of which can itself fail. Whether growing the code helps or hurts depends on the quality of the underlying hardware. Theory says there is a threshold: if your physical error rate is below it, making the code bigger suppresses logical errors exponentially. Above it, making the code bigger makes everything worse — you're adding noise faster than you're correcting it. For most of the field's history, real hardware sat on the wrong side of that line.

The honest assessment

Willow is a genuine scientific milestone: it shows the roadmap to fault-tolerant quantum computing is viable rather than merely theoretical. It does not mean quantum computers are ready for commercial applications. A logical qubit good enough for long computations will consume on the order of a thousand physical qubits, and the algorithms that matter need thousands of logical ones. The most credible current estimate for breaking RSA-2048 encryption — published by Google's own Craig Gidney in 2025 — is a machine with just under a million high-quality physical qubits running for about a week. Willow has 105. The gap is roughly four orders of magnitude, plus years of engineering to run error-corrected operations at speed and scale.

But notice the direction of travel in that estimate. In 2019, the same researcher put the requirement at 20 million qubits. By 2025 it was under one million — a twentyfold reduction in six years, achieved almost entirely through better algorithms and error-correction techniques rather than better hardware. The finish line is moving closer from both ends: hardware is improving and the requirement is shrinking. That, not any single chip, is why security agencies treat post-quantum migration as urgent even though the machine that breaks encryption doesn't exist — a dynamic our piece on why banks are investing in quantum covers in depth.

What to watch instead of qubit counts

Raw physical qubit counts have become a marketing metric, and they tell you almost nothing — a thousand noisy, uncorrected qubits are worth less than ten good logical ones. The numbers that track real progress are logical qubit counts, logical error rates, and the depth of error-corrected computation a machine can sustain. When those numbers appear in an announcement, take it seriously; when a press release leads with physical qubits alone, be sceptical. The pattern with quantum milestones is consistent: the physics advances faster than the applications. That's not a failure — it's the normal arc of a deep technology, and the people building quantum literacy now will be the ones positioned to judge each milestone as it lands.