Begin with skepticism about the headline number. Ten septillion years is the kind of figure that does not really land. It is too large to picture, too clean to interrogate, and too useful as marketing to take entirely at face value. What it produced, in December 2024, was a wave of coverage announcing that Google’s Willow quantum chip had outperformed the fastest classical supercomputer by a factor of roughly one quadrillion on a calculation that took the chip about five minutes. What it largely obscured was a separate result, buried lower in the announcement, that quantum computing researchers consider far more important than any benchmark comparison.
The benchmark in question is not a useful calculation. The supercomputer comparison rests on assumptions that have been challenged before and may be challenged again. And the real news from the Willow announcement — the part that vindicates nearly thirty years of theoretical work — has almost nothing to do with the headline statistic. The disparity between what got reported and what actually mattered is itself worth examining.
According to Google’s official announcement of the chip on the company’s research blog, Willow is a 105-qubit superconducting quantum processor, built using transmon qubits (a specific type of superconducting circuit that has become the dominant quantum-computing hardware approach over the past decade). The benchmark calculation Willow performed is called Random Circuit Sampling, or RCS. It is a test developed specifically to measure whether a quantum computer can perform calculations that no classical computer can match in any reasonable timeframe. RCS is, in the words of Hartmut Neven, the founder and lead of Google Quantum AI, “the classically hardest benchmark that can be done on a quantum computer today.” The 10-septillion-year figure represents Google’s estimate of how long Frontier would need to reproduce the result. A simultaneous paper in Nature presented the underlying technical detail.
What the speed comparison actually means
The first thing to understand about the 10-septillion-year figure is that it does not mean Willow can solve useful problems a quadrillion times faster than Frontier. It means Willow can solve this specific benchmark problem a quadrillion times faster than Frontier. Random Circuit Sampling is a calculation that has no practical application beyond demonstrating that quantum computers can do things classical computers cannot. The output of an RCS computation is a list of probabilities for various quantum measurement outcomes. That is useful for verifying that the quantum computer is functioning as designed, but not useful for designing a new drug, optimising a supply chain, breaking an encryption key, or any of the other applications that have been proposed for quantum computing.
The benchmark is, in this sense, somewhat circular: it measures whether a quantum computer can do something that is hard for a classical computer, by selecting a specific task that is constructed to be hard for a classical computer. As reported by HPCwire’s analysis of Google’s quantum-advantage claims with Willow, the physics community typically distinguishes between two milestones in the development of quantum computers: “quantum advantage,” which requires solving a practical problem faster or more accurately than any classical computer ever could; and the weaker condition of demonstrating that a quantum computer can do something beyond the capabilities of classical computers, even if that something is artificial. Willow’s RCS performance establishes the weaker condition, which Google has called a “beyond-classical” regime. It does not, by the more rigorous definition, establish quantum advantage in the practical sense.
The 10-septillion-year figure has also been disputed by other researchers in the field, including IBM, which has historically argued that Google’s classical-supercomputer comparisons are based on conservative classical algorithms that have since been improved. When Google announced quantum supremacy with its 53-qubit Sycamore chip in 2019, the original claim was that Sycamore had completed a calculation in 200 seconds that a classical supercomputer would have required 10,000 years to match. IBM responded within days with a paper arguing that a better classical algorithm could complete the same calculation in approximately 2.5 days, not 10,000 years. Whether the 10-septillion-year figure for Willow holds up to similar scrutiny will depend on whether better classical algorithms can be found over the next several years.
The other breakthrough
The achievement that has attracted less popular attention but more serious enthusiasm from quantum computing researchers is Willow’s demonstration of below-threshold quantum error correction, a result that the field has been working toward for nearly three decades. As documented by Live Science’s detailed technical coverage of the announcement, quantum computers are inherently noisy. Qubits, the quantum equivalent of classical bits, are extraordinarily sensitive to interference from heat, electromagnetic fields, cosmic rays, and small imperfections in their physical construction. Every time a qubit performs an operation, there is some probability that an error will occur. The error rate for a typical superconducting qubit is approximately one error per thousand operations, comparable, in classical terms, to one bit-flip per thousand-bit calculation.
By contrast, classical computers experience approximately one error per quintillion operations. The factor of 10^15 difference in reliability is the central reason quantum computers have not yet become practically useful: a calculation requiring thousands of qubits and millions of operations would accumulate so many errors that the result would be indistinguishable from random noise. The theoretical solution to this problem, proposed by Peter Shor in 1995 and elaborated by Alexei Kitaev in 1997, is quantum error correction. The technique combines multiple physical qubits to form a single “logical qubit” that can tolerate errors in its individual components. The catch is that error correction only works if the underlying physical error rate is below a specific threshold. Above the threshold, adding more qubits makes the total error rate worse, not better. Below the threshold, adding more qubits makes the total error rate exponentially better. The threshold had never been demonstrated in physical hardware until Willow.
What Willow actually demonstrated
Per The Quantum Insider’s coverage of the technical achievement, Willow demonstrated the below-threshold regime by building logical qubits of increasing size and measuring how the error rates changed. As reported in the Nature paper, the logical error rate decreased by a factor of approximately 2.14 with each two-step increase in code distance, with the distance-3 logical qubit showing approximately 0.65 percent error per cycle, the distance-5 logical qubit showing approximately 0.31 percent error per cycle, and the distance-7 logical qubit (built from 101 physical qubits) showing 0.143 percent error per cycle. The pattern is exactly what theory predicts for below-threshold operation: errors decrease exponentially as the logical qubit size increases, opening the path toward arbitrarily large quantum calculations with arbitrarily small total error rates.
This is the result that quantum computing researchers have been working toward for the better part of three decades. It does not, by itself, produce a useful quantum computer. Reaching practical applications will require scaling up to thousands or millions of physical qubits per logical qubit, building larger and larger arrays, and solving a long list of engineering challenges that current hardware does not yet address. But it does establish, for the first time in physical hardware, that the basic theoretical pathway to fault-tolerant quantum computing is real. The 1995 prediction has been verified experimentally. The exponential error reduction that error-correction theory has long promised is now something that has been demonstrated, not merely modelled.
What comes next
The implications of below-threshold error correction extend well beyond Willow itself. If errors really do decrease exponentially as qubit count increases, then the development trajectory of quantum computing changes fundamentally. The field can, in principle, now begin building progressively larger systems in the confidence that the engineering problems are tractable rather than insurmountable. The estimated timeline for useful quantum computers, for applications like drug discovery, materials simulation, cryptographic analysis, and optimisation problems, has accordingly shifted from “perhaps in the indefinite future” to “perhaps within the next 10 to 20 years,” depending on which researcher one asks. Google has stated that its next chip will aim to demonstrate a logical qubit with a substantially lower error rate than has been achieved by any classical computer, which would be the first useful quantum computation in the strict sense: solving a problem that has practical value, faster or more accurately than the best classical alternative.
It is worth being direct about which of Willow’s two results will matter in a decade. The speed comparison will not. Random Circuit Sampling is a constructed benchmark, the 10-septillion-year figure is the kind of claim that previous Google announcements have seen substantially revised within weeks, and even if the number survives, it describes a calculation no one needs done. The error-correction result is the one that changes the field. It converts fault-tolerant quantum computing from a theoretical pathway into an engineering programme, and it does so on the basis of measured exponential scaling rather than projection. The headline number will probably continue to dominate the press cycle for as long as Willow is in it. The historical record, when it is written by people who care about what the machine actually did, will record the error-correction demonstration as the moment the field crossed from promise into evidence.
