Quantum computing is having a moment, and most of the coverage is either breathless or baffling. The useful version sits in between. It is a real technology that is genuinely advancing, mostly in ways that will not touch your business for years, with one exception you should act on now. What makes the story interesting in 2026 is not quantum on its own, but how it and artificial intelligence have started to lean on each other. This is a plain-English guide to how the two fit together, what is real, and what a sensible business owner should actually do.
The short version, before the detail:
- Quantum computers are specialist tools, not faster versions of the machines you already own. They help with a narrow set of problems, chiefly simulating molecules and materials.
- The hardware is real but early. The best machines today have hundreds to a thousand noisy qubits and cannot beat an ordinary computer on any task that pays.
- The most productive quantum-and-AI story right now runs the other way round: AI is being used to make quantum hardware work better.
- For nearly every business, the honest advice is prepare, do not deploy. The one urgent action is moving to quantum-safe encryption.
On this page
- First, what a quantum computer actually is
- Where the technology really is (2025 to 2026)
- The part most coverage misses: AI is already making quantum better
- Quantum for AI: keep your scepticism handy
- Supremacy, advantage, and why the words matter
- What this actually means for your business
- The grounded take
First, what a quantum computer actually is
An ordinary computer chip is, at heart, billions of switches that are either on or off, a 1 or a 0. Everything your laptop does comes down to shuffling these bits. A quantum computer plays by different rules, borrowed from the physics of atoms and particles, and stores information in qubits.
Two strange properties do the work. Superposition lets a qubit behave as a blend of 0 and 1 at once, so a handful of qubits can represent an enormous number of combinations together. Ten qubits span 1,024 combinations at the same time; three hundred span more combinations than there are atoms in the observable universe. Entanglement links qubits so their states stay correlated, letting the machine explore those combinations in a coordinated way rather than one at a time. A well-designed quantum algorithm then uses interference, like ripples on a pond meeting crest to crest or crest to trough, to reinforce the paths that lead to the right answer and cancel the ones that do not.
This is why a quantum computer is not simply a faster computer. It offers a dramatic advantage only for specific problems with the right mathematical shape, chiefly simulating other quantum systems such as molecules and materials, factoring very large numbers, and some optimisation. For email, spreadsheets and the AI tools you use today, it offers nothing at all.
There is one universal catch: noise. Qubits are exquisitely fragile, and the slightest heat, vibration or stray signal scrambles the calculation. Today’s machines are called NISQ devices, for Noisy Intermediate-Scale Quantum, because errors pile up almost as fast as useful work gets done. The fix is quantum error correction, which bundles many physical qubits into a single reliable logical one, with the rest constantly spotting and fixing errors. The overhead is brutal: hundreds to thousands of physical qubits for each logical qubit, which is why a machine capable of breaking encryption may need a million or more physical qubits in total.
Where the technology really is (2025 to 2026)
This is a field mid-transition from science to engineering. The leading machines have on the order of one hundred to a thousand physical qubits, and only a handful of error-corrected logical ones. The milestones are genuine, but they are engineering milestones, not commercial ones.
The standout result came from Google. Its 105-qubit Willow chip, unveiled in December 2024 and published in Nature, hit a target the field had chased for nearly thirty years: as engineers made a logical qubit larger, the error rate fell instead of rising. In October 2025 Google went further, reporting in Nature the first verifiable quantum advantage with its Quantum Echoes algorithm, running a physics task about 13,000 times faster than the best classical method on a leading supercomputer. Worth keeping in perspective: IEEE Spectrum noted that while founder Hartmut Neven expects real-world applications within five years, the researchers themselves cautioned the results are not yet beyond the reach of classical machines for anything commercially useful.
The other serious players are lined up behind their own roadmaps and claims. It pays to read them with a cool head.
| Who | 2025 to 2026 headline | Reading it honestly |
|---|---|---|
| IBM | A public roadmap to Starling, a fault-tolerant machine with about 200 logical qubits, by 2029. | The most detailed plan in the industry, and a firm date to judge everyone against. |
| Quantinuum | Helios, 98 trapped-ion qubits with better than 99.9% fidelity, launched November 2025. | High quality, fully connected qubits, but still small in number. |
| IonQ | A world-record two-qubit gate fidelity above 99.99%, the so-called four nines, in October 2025. | Excellent accuracy on a small system; the projected leap to millions of qubits by 2030 is the ambitious part. |
| Microsoft | The Majorana 1 chip, a bet on inherently stable qubits, February 2025. | High risk, and disputed: one physicist called claiming a topological qubit in 2025 “selling a fairytale”. |
The money has followed the noise. McKinsey’s 2026 Quantum Technology Monitor put investment in quantum start-ups at 12.6 billion dollars in 2025, more than six times the year before, and estimates the technology could create between 1.3 and 2.7 trillion dollars of value worldwide by 2035. Treat the top of that range as a hope, not a forecast. Quantum is still under 1% of global venture funding, and the headline numbers rest on optimistic assumptions.
The part most coverage misses: AI is already making quantum better
Here is the genuinely interesting bit. Ask most people how quantum and AI connect and they will say quantum will one day supercharge AI. That may come, but it is speculative. The relationship that is already delivering runs the other way: modern AI is being used to tame quantum hardware today.
The clearest example is error decoding. When a quantum computer runs, it constantly throws off a stream of clues about where errors are creeping in, and something has to read those clues in real time and work out what went wrong. Google DeepMind’s AlphaQubit, published in Nature in November 2024, is a neural network that does exactly this, and does it more accurately than the hand-built methods before it: 6% fewer errors than tensor-network decoders and 30% fewer than correlated matching in the largest tests. A follow-up in December 2025 made it fast enough to keep up with the chip in real time.
It goes further than reading errors. A 2026 Nature paper showed a reinforcement-learning agent, the same broad technique that mastered board games, can continuously tune Google’s Willow processor while it runs, squeezing out roughly 20% further suppression of the logical error rate and keeping the machine markedly steadier as it drifts, replacing slow manual recalibration. DeepMind has also used AI to redesign the circuits quantum computers will run, cutting out the most expensive operations. Nvidia, sensing where this is heading, launched NVQLink in October 2025 to wire quantum processors directly to its GPUs. People call this a virtuous cycle: AI helps build better quantum machines, which might one day help train better AI.
Quantum for AI: keep your scepticism handy
The reason people dream of quantum helping AI is cost. Training a frontier model is staggeringly expensive; Stanford’s AI Index estimated the compute for GPT-4 at around 78 million dollars and for Google’s Gemini Ultra at roughly 191 million. If a quantum machine could shortcut even part of that, it would be a big deal.
The catch is that this is the least certain corner of the whole field. The approaches, known as quantum machine learning, are experimental on today’s hardware, and several early claims of huge speed-ups have quietly collapsed. In 2018 an 18-year-old undergraduate, Ewin Tang, found an ordinary classical algorithm that matched a celebrated quantum one, a process now called dequantisation that has since deflated several others. The computer scientist Scott Aaronson has long warned, in a piece pointedly titled Read the Fine Print, that many quantum machine-learning speed-ups come with caveats that swallow the advantage. If a vendor promises quantum will transform your AI soon, that is the claim to probe hardest.
Supremacy, advantage, and why the words matter
Two phrases get used loosely, and the difference matters when you read a headline. Quantum supremacy means doing some task, however useless, that no classical computer can feasibly match; Google first claimed it in 2019 on a contrived benchmark. Quantum advantage is the meaningful bar: beating classical machines on a genuinely useful problem. As of 2026, no company has demonstrated a commercial quantum advantage in a real production setting. When you see the word advantage, check whether the useful problem is real or the benchmark is a stunt.
What this actually means for your business
For almost every organisation, the right posture is to prepare, not deploy. Quantum is unlikely to change how you operate before the early 2030s, and there is a real risk of a “quantum winter” if the hype outruns delivery. Even Nvidia’s Jensen Huang suggested in early 2025 that practical quantum computing could be 15 to 30 years away, though he softened that later in the year. None of that means ignore it. It means match your effort to the stage.
| Horizon | What a sensible business does |
|---|---|
| Now (0 to 2 years) | Deal with the encryption threat. This is the one concrete, urgent action. Find where you rely on today’s encryption, and start planning a move to the new standards. |
| Now (0 to 3 years) | Build literacy and run tiny, low-stakes pilots on cloud quantum services, in your most relevant area: chemistry and materials, or optimisation for logistics and finance. Treat them as learning, not production. |
| Medium (3 to 7 years) | Watch the metrics that matter: logical qubit counts and logical error rates, not the raw physical qubit numbers vendors love to quote. Judge everyone against IBM’s 2029 target. |
| Long (7 to 10 years) | Move business-critical work to quantum only once there is repeatable, independently verified advantage in your specific domain. |
The encryption point is the one genuinely time-sensitive item, so it is worth spelling out. A future quantum machine could break the encryption that protects almost everything sensitive online. Google’s Craig Gidney estimated in May 2025 that the standard RSA encryption could be cracked by a machine with fewer than a million noisy qubits, a twentyfold drop from his own 2019 estimate. That machine does not exist yet, but the threat is already here, because adversaries can “harvest now and decrypt later”, quietly storing your encrypted data today to unlock once the hardware matures. Anything that must stay secret beyond about 2035 is already exposed. The reassuring news is that the defence is ready: America’s standards body, NIST, finalised post-quantum encryption standards in August 2024, and the large platforms are already rolling them out.
The grounded take
Quantum computing is real, genuinely exciting for chemistry and materials, and years away from your day-to-day. The most useful thing happening right now is the quieter one: AI is making quantum hardware work, a partnership that could compound progress in both. For your business, the moves are simple. Get your encryption on a path to the new standards, stay literate, run the odd small experiment, and judge the grand promises by logical qubits and independent results rather than press releases.
That is the same principle we apply to any shiny technology: start with the problem, measure the outcome, and stay vendor-neutral. If you want a grounded read on the AI that can help you today rather than in a decade, our AI-readiness checklist is the place to start, and if you would rather talk it through, that is exactly what we do. And for the uncomfortable version of what AI has already done to professional work, including my own, there is the most automatable job in Britain.