Scott Aaronson's 'qombie' label for the HSBC/IBM bond trading paper is a working template for evaluating the next quantum advantage claim you read, and the public money that follows it.
In November 2025, HSBC and IBM published a paper describing a 34% relative improvement in predicting whether a corporate bond would trade at a quoted price, a small step inside the algorithmic machinery large banks use to fill customer orders. HSBC framed the work in a press release as the "world's first known quantum-enabled algorithmic trading" experiment. Six months later, the paper has a problem: one of the most cited complexity theorists working today has read it, written a public response, and labeled the central claim a "qombie."
Scott Aaronson, who holds the Schlumberger Centennial Chair in Computer Science at the University of Texas at Austin and has spent two decades working on the theoretical limits of quantum machines, published a detailed critique of the HSBC/IBM result. His label: a qombie, a quantum advantage claim that walks like a result and talks like a result, but does not survive the kind of scrutiny that distinguishes a genuine computational speedup from a benchmark artifact. The word has stuck. The dispute, technical as it is, is now the cleanest working case study in years for what a real quantum advantage claim has to clear before the field should treat it as one.
The claim, on its own terms
The HSBC/IBM experiment targets a small but real problem in bond markets. When a bank receives a customer order to buy or sell a corporate bond, it has to estimate the probability that the bond will actually trade at the quoted price, what traders call fill probability. Get that estimate right, and the bank can quote tighter prices and risk less capital on every trade. Get it wrong, and the bank either loses the trade or ends up holding inventory it does not want.
The paper's central claim is that a machine learning model trained on inputs that have been pre-processed by a quantum circuit does this job 34% better, in relative terms, than a classical model trained on the same raw data. That 34% is the number HSBC chose to put in headlines. It is also the number Aaronson spent his response pulling apart.
Where Aaronson says the claim goes wrong
Aaronson's critique is mechanism-level, not vibes-level. He is not arguing that quantum computing can never help with bond trading. He is arguing that this paper, on its own evidence, has not shown that it does.
The first problem is the relationship between the claimed advantage and the noise. Quantum computers in 2025 are noisy. Qubits decohere, gates misfire, and results have to be averaged over thousands of runs. The HSBC/IBM experiment, like almost every quantum machine learning paper so far, runs on a device where the signal sits close to the noise floor. The result, as Aaronson notes, is that the largest reported gains appear precisely in the regime where the hardware is least reliable, which is also the regime where any classical model would do worst on the same raw data. When the hardware is loud, "the quantum circuit is transforming the input" and "the quantum circuit is scrambling the input" can look similar on a benchmark. The paper does not, in his reading, separate these two explanations.
The second problem is the baseline. A 34% improvement only means something if you are comparing against the right classical model. The classical models used in the paper, by Aaronson's reading, were not the strongest classical models that could be brought to bear on the same task. A reader cannot tell from the paper whether a well-tuned classical machine learning pipeline would close the gap, or erase it, on the same data. That is the standard quantum advantage claims in machine learning are expected to meet, and it is the standard the HSBC/IBM paper does not, in Aaronson's argument, meet.
The third problem is the role of noiseless simulation. A meaningful chunk of the paper's evidence comes from running the quantum circuit as a noiseless classical simulation, not on the actual hardware. That is a reasonable thing to do during development. It is not, however, evidence that the hardware is doing something a classical computer cannot. The simulated circuit is, by construction, a classical computation. A reader who skims the press release and walks away with "quantum beats classical" is reading past a distinction the paper itself is more careful about.
None of this is a claim of fraud. It is a claim that the standard of evaluation is too weak to support the headline.
Why the label matters
The qombie label, a term Aaronson has used for years to describe quantum advantage claims that look alive on press release day and fail to reproduce under closer reading, is doing real work here. Quantum computing is a field under unusual pressure. Quantum machines are expensive. Public money is starting to flow toward them in earnest: in May 2026, Forbes reported that the US government is preparing a roughly $2 billion equity investment in quantum computing firms, taking stakes in the same companies whose advantage claims will, eventually, determine whether the public got its money's worth. The pull of headlines is real, and so is the pull for the labs and vendors issuing them to overstate what a noisy current-generation device can do.
A reader does not need to take a position on whether quantum advantage in trading is possible in principle. The question the HSBC/IBM paper raises is narrower, and more useful: did this particular experiment, evaluated against the strongest classical baselines, on a noisy device, in a regime where noise and signal are hard to disentangle, actually show what its press release said it showed? Aaronson's argument is that it did not. That is a methodological dispute, not a verdict on the field.
What to watch
Two things will tell you whether the qombie label sticks. The first is whether HSBC and IBM run a follow-up comparison against a tuned classical model on the same bond data, and report the result either way. The second is whether the broader literature on quantum machine learning for finance starts treating "qombie" as a checklist rather than a slur, a set of conditions a paper is expected to satisfy before a press release goes out.
Meanwhile, the operational story is moving on a separate track. Oxford Quantum Circuits and Equinix have announced a co-located quantum computer inside an Equinix data center, with OQC positioning direct quantum-as-a-service access as a way to get real workloads onto real hardware and away from the kinds of benchmark experiments the qombie critique targets. That is the constructive version of the same argument: if the field wants the public money following the work to land on something real, the path runs through tighter evaluation and tighter access to devices, not through more dramatic press releases.
The HSBC/IBM paper is not the only quantum advantage claim you will read this year. It is, however, the one with the cleanest public dissection. Read it that way: not as a verdict on quantum computing, and not as a scandal, but as a template for what asking the right questions looks like.