Is quantum computing useful yet? Real results and overclaims

As of September 2026, quantum computers are useful for a few scientific tasks and for learning, and not yet for your business problems. In 2025 two experiments did things classical machines could not match and survived scrutiny. Several business-flavored claims did not. If you run a bank, a mine or a hospital in Latin America, the part of quantum that needs your budget this year is security, not computing.

The table sums up the five results that got the most attention, with the strongest named critique of each.

What counts as “useful” in quantum computing?

A quantum result is useful when it does a task someone needs, better than the best classical method, and someone else can check it. Most disputes in 2025 were about the second condition. “Better than a classical computer” means little if the comparison is against an old algorithm or an untuned solver.

Three words get mixed up. Quantum supremacy or advantage means beating classical computers on some task, even a contrived one; Google’s 2019 random circuit sampling was the first claim. Useful advantage means the task matters to someone outside the lab. Verifiable advantage means the answer can be checked, either by another quantum computer or by a classical calculation. Google’s October 2025 announcement leaned on that last word for good reason.

Did Google’s Quantum Echoes show real advantage?

Yes, on a physics task, and Google was careful about the limits. Its Quantum Echoes algorithm measures out-of-time-order correlators, a quantity that describes how quantum information spreads in a chaotic system. Google reported on 22 October 2025, with a paper in Nature, that Willow ran it in about two hours, and estimated a classical supercomputer would need about 13,000 times longer.

Two caveats come from Google’s own post. First, the speedup is measured against known classical algorithms, so a better one could narrow the gap. Second, the follow-up that applied the method to learn molecular structure through nuclear magnetic resonance was, in Google’s words, “not yet beyond classical,” and that paper was still headed for peer review. So this is a real, checkable advantage on a scientific measurement, with an application in sight but not in hand.

Is certified randomness a real use case?

It is the closest thing to a verified, practical output so far, though it is a small niche. In a Nature paper published on 26 March 2025, JPMorganChase, Quantinuum and researchers from US national laboratories and universities used the 56-qubit H2-1 trapped-ion processor to generate more than 70,000 random bits and certify them with over 1.1 exaflops of classical computation.

Certified randomness matters for some cryptographic protocols, lotteries and audits, where you need proof that nobody could have predicted the numbers. It does not speed up any calculation. One of the co-authors was Scott Aaronson, whose protocol it builds on, which is worth noting because he is also the field’s best-known critic of overclaims.

What went wrong with the HSBC quantum trading claim?

The result was real data but not a quantum speedup. On 25 September 2025 HSBC announced, with IBM, “up to a 34 percent improvement” in predicting whether a European corporate bond trade would be filled at the quoted price, using IBM Heron processors in a hybrid quantum-classical model.

The same day Scott Aaronson published a post titled “HSBC unleashes yet another ‘qombie’,” his word for an undead claim of business quantum advantage. His main point was that the improvement appeared when the quantum hardware’s noise was part of the pipeline and vanished in noiseless simulation, so it had “nothing really to do with quantum mechanics.” A noisy random feature generator can help a classifier, but so can a classical random number generator. Our article on quantum computing in finance looks at what banks can test today instead.

Did D-Wave achieve quantum supremacy on a useful problem?

D-Wave says yes, and researchers at EPFL dispute it. On 12 March 2025 D-Wave published a paper in Science reporting that its Advantage2 annealer simulated the quantum dynamics of programmable spin glasses (a magnetic materials problem) in minutes. It estimated the same simulation would take nearly one million years on the Frontier supercomputer.

Around the same time, Linda Mauron and Giuseppe Carleo at EPFL posted an arXiv paper showing that a classical method, time-dependent variational Monte Carlo, could simulate annealing dynamics on systems of up to 128 spins with resources that scale polynomially, and argued the advantage frontier sits further out than D-Wave claimed. The debate is not settled, and that is the point: a supremacy claim that classical specialists are already contesting on the day it is published is not something to build a business case on.

What about Microsoft’s Majorana 1?

It is a hardware bet, not a usefulness claim, and the evidence is disputed. On 19 February 2025 Microsoft unveiled Majorana 1, an eight-qubit chip it described as the first built on topological qubits, which in theory would be far more stable than today’s qubits.

The accompanying Nature paper did not demonstrate a topological qubit; it showed one type of measurement but not the second one needed. Henry Legg of the University of St Andrews called the data “incredibly unconvincing,” Sergey Frolov of the University of Pittsburgh said the data shown were “just noise,” and Kartiek Agarwal of Argonne said the device “certainly can’t be used as a qubit in its present state.” If topological qubits work, they could shorten the road to large machines. Until independent groups reproduce them, treat Majorana 1 as unproven.

What changed in 2026?

The 2026 news was less about new advantage claims and more about companies acting as if large machines are closer. On 25 March 2026 Google set 2029 as the deadline for its own post-quantum migration, citing faster hardware progress, better error correction and lower resource estimates for factoring. On 30 June 2026 Microsoft moved its own target to 2029 as well, with Azure CTO Mark Russinovich saying cryptographically relevant quantum computers “could arrive sooner than previously expected.” The Global Risk Institute’s expert survey, published on 9 March 2026, put the chance of such a machine within ten years at 28% to 49%.

None of that makes today’s quantum computers more useful for your operations. It does mean the two questions have split. “Can quantum help my business?” is still mostly “not yet.” “Can quantum hurt my business?” now has dates attached.

So what is quantum computing good for in 2026?

Today it is good for four things, in rough order of how solid they are.

1. Physics and chemistry research on quantum systems, where the machine simulates something that is itself quantum. Quantum Echoes is the best example.
2. Certified randomness, a small but verified niche.
3. Learning and preparation. Teams that learn the tools now will be ready when error-corrected machines arrive. Quantinuum’s Helios (98 qubits, 48 error-corrected logical qubits, launched November 2025) and IBM’s roadmap to Starling in 2029 give a rough timeline.
4. Quantum-inspired classical methods. Tensor networks and other techniques that came out of quantum research run on normal hardware and sometimes help with optimization and simulation today.

What it is not good for yet: routing trucks, pricing portfolios or designing drugs better than the best classical methods. Our pages on logistics and transport and healthcare and pharma describe what those sectors can test in the meantime, mostly with classical and quantum-inspired tools.

The one thing you should not wait for is security. The hardware progress that makes these experiments possible is also what shortens the timeline to a machine that can break RSA and elliptic-curve cryptography. Our post on when Q-Day might arrive covers those estimates.

How can you tell a real result from hype?

Four questions filter out most overclaims:

• What is the classical baseline, and did classical specialists try to beat it? If the comparison is against “a classical computer” with no named algorithm, be careful.
• Is the result verifiable, and was it peer reviewed? A press release and a Nature paper are different things.
• Does the advantage survive without noise? The HSBC case shows why this matters.
• Does the task matter to anyone outside the lab? Many true advantage results are on tasks designed to be hard for classical machines.

IBM, Algorithmiq, the Flatiron Institute and BlueQubit launched an open quantum advantage tracker in November 2025 to test claims in public, and IBM has said it expects verified advantage by the end of 2026. That tracker is a better place to check progress than any vendor’s announcement, including IBM’s.

If you want to find out whether any of your problems is worth testing, quantum use case discovery starts with the classical baseline and tells you plainly when the answer is “not yet.” For the security side, a quantum readiness assessment comes first. Our guide on choosing a first quantum use case covers the same ground if you want to try it yourself.