Quantum Computing in 2026: How It Works, What Has Been Achieved, and Why It Matters

A quantum computer is a machine that uses the rules of quantum mechanics — superposition, entanglement and interference — to process information in ways a classical computer cannot efficiently copy. Where an ordinary computer stores information as bits that are either 0 or 1, a quantum computer uses quantum bits, or qubits, which can hold a blend of both at once. That does not let the machine simply try every answer at the same time, a common misunderstanding. It lets a well-designed algorithm steer the odds so that the right answer is the one most likely to appear when the qubits are read, and only some problems can be steered that way.

For decades that description was a theoretical ambition. In 2026 it describes working machines, with a caveat that matters: they have run a handful of carefully chosen calculations that the best classical methods struggle to match, and nearly every such claim has been challenged, sometimes within weeks. This is how quantum computing works, what has actually been achieved and disputed, and what it might mean for medicine, cryptography and the future of computing itself.

Beyond the announcements, a quieter measure is what researchers title their papers. Searching arXiv titles on 1 October 2026 turned up 6,310 papers with “quantum computation” in the title, 998 on quantum error correction and 368 on quantum advantage, each a little higher than a month earlier (6,234, 976 and 350).

The ratio matters more than any single processor announcement. For every paper claiming an advantage over classical machines, roughly three are about keeping the machine working long enough to have one.

Hardware numbers date within months. The shape of the literature is a slower and more honest indicator of where this technology actually is.

How Quantum Computing Actually Works

To grasp quantum computing, it helps to start with what classical computing does — and where it hits a wall. A classical computer works by flipping bits, tiny switches that are either on (1) or off (0). Everything it does, from loading a page to modelling a molecule, is a sequence of these binary operations. That works superbly for most tasks. But for problems involving vast numbers of possible combinations, it meets a hard limit.

Consider chemistry. The amount of information needed to describe the electrons in a molecule exactly roughly doubles with every extra electron orbital the calculation includes, so exact simulation of anything but small molecules quickly outruns any classical computer. Chemists rely on clever approximations instead, and those approximations are weakest exactly where electrons interact most strongly. No amount of extra processing power fixes an exponential; the architecture itself is the constraint.

Quantum computers compute differently. A single qubit can hold a superposition of 0 and 1. Two qubits can describe four combinations at once, ten qubits over a thousand, and a few hundred qubits more combinations than there are atoms in the observable universe. That does not make a quantum computer a faster version of an ordinary one. It gives certain problems, such as simulating molecules or factoring large numbers, a structurally different and potentially far shorter route to an answer, while leaving most everyday computing exactly where it is.

Superposition, Entanglement, and Interference

Superposition is the property that lets a qubit exist in multiple states at once. The usual image is a spinning coin — neither heads nor tails until it lands. A qubit in superposition is similarly undecided until measured, contributing to a calculation as though all its possible values exist together. This is the source of a quantum computer’s power, and also its fragility: any stray interaction with the environment — heat, vibration, electromagnetic noise — can collapse the superposition prematurely and introduce errors. That effect, called decoherence, is one of the central engineering challenges of quantum computing.

Entanglement — explored in depth in our article on quantum entanglement — links two qubits so that the state of one instantly determines the state of the other, no matter how far apart they are. Inside a quantum computer, entanglement ties qubits together so that operations on one ripple through others in coordinated ways, letting the machine handle complex relationships between many variables at once.

Interference is the quieter third ingredient. Because qubits behave like waves, their possibilities can reinforce or cancel one another. A well-designed quantum algorithm choreographs this interference so that the wrong answers cancel out and the right ones add together, leaving the correct result far more likely to appear when the qubits are finally measured. These are not magic — they are precise, well-understood physical phenomena that happen to be extraordinarily useful for computation.

The Many Kinds of Qubit

There is no single way to build a qubit, and the competition between designs is one of the most important stories in the field. Superconducting qubits, favoured by Google and IBM, are tiny circuits chilled to near absolute zero — fast, but delicate. Trapped-ion qubits, used by IonQ and Quantinuum, hold individual charged atoms in electromagnetic fields and manipulate them with lasers, trading speed for exceptional stability.

Newer approaches are gaining ground. Neutral-atom machines arrange atoms in reconfigurable grids held by “optical tweezers.” Photonic designs encode information in particles of light. And Microsoft’s topological qubits take the most radical path of all, storing information in the collective geometry of an exotic quantum state rather than in any single particle. Each design balances speed, stability, and scalability differently, and no one yet knows which will win — or whether several will coexist.

Whatever the design, the engineering is extreme. Superconducting quantum computers must be chilled inside elaborate dilution refrigerators to around a hundredth of a degree above absolute zero — colder than the depths of interstellar space — and shielded from the faintest stray vibration or radio wave. The golden, chandelier-like structures so often shown in photographs are not the computer itself but the intricate cooling and wiring needed to keep a handful of qubits calm enough to compute.

The Algorithms That Made It Matter

Hardware is only half the story. A quantum computer is useless without algorithms designed to exploit superposition and interference, and the field’s foundations were laid long before the machines existed. Feynman supplied the motivating idea in 1981. In 1985, David Deutsch formalised the concept of a universal quantum computer, proving such a machine could in principle simulate any physical process.

The turning point came in 1994, when Peter Shor devised an algorithm able to factor large numbers exponentially faster than any known classical method — the result that first made governments and banks take quantum computing seriously, because it threatened the encryption guarding the modern world. Two years later, Lov Grover found a quantum algorithm that speeds up searching through unsorted data. Together these proved that quantum computers were not merely faster classical machines but a genuinely different kind of tool — and they set the targets the hardware has chased ever since.

What IBM, Google, and Microsoft Have Achieved

The past eighteen months have delivered milestones the field had chased for decades.

Google Willow and Quantum Echoes

In December 2024, Google unveiled Willow, a 105-qubit superconducting chip that marked a real advance in quantum error correction. Google reported that Willow ran a benchmark called random circuit sampling in under five minutes, a task it estimated would take one of the fastest supercomputers 10 septillion (10 to the power 25) years. The benchmark was chosen because it is hard for classical computers, not because its answer is useful.

In October 2025 Google followed with Quantum Echoes, published in Nature. Willow measured a quantity physicists call an out-of-time-order correlator, which tracks how information spreads and scrambles inside a quantum system, in about two hours; Google estimated that the best classical algorithm on a leading supercomputer would take some 13,000 times longer. Google calls the result verifiable because it is a measurable number that another quantum computer should reproduce. The practical payoff is not here yet. A companion demonstration applying the technique to molecules, of the kind studied with nuclear magnetic resonance, was in Google’s own words “not yet beyond classical”.

IBM Nighthawk and the Path to Fault Tolerance

IBM has pursued systematic scaling against a public roadmap. In November 2025 it announced Nighthawk, a 120-qubit processor with 218 tunable couplers, up from 176 on its predecessor Heron, able to run circuits of about 5,000 two-qubit gates, with 7,500 targeted by the end of 2026. Alongside it came Loon, an experimental chip that IBM says demonstrates every key processor component fault tolerance needs, and a decoding milestone: classical hardware that identifies errors in IBM’s codes in under 480 nanoseconds, ten times faster than the previous leading approach and a year ahead of schedule. In August 2026 IBM released Nighthawk r2, which adds a reset element to every qubit and runs more than 100,000 circuits a second, and joined two of its cryogenic modules, cooled below 15 millikelvin, into a single environment. Its roadmap names Kookaburra for 2026; Starling for 2029, which IBM expects to be the first fault-tolerant quantum computer, running 100 million gates on 200 qubits; and Blue Jay, with a billion gates on 2,000 qubits, from 2033.

Microsoft and the Majorana Connection

Microsoft has taken a fundamentally different route, based on the Majorana fermion — a particle that is its own antiparticle, first theorised by the physicist whose haunting story we tell in our article on Ettore Majorana. In February 2025 Microsoft announced Majorana 1, which it described as the first processor built on topological qubits, in which information would be stored in the shape of an exotic quantum state rather than in any one particle and so be naturally shielded from noise. The peer-reviewed paper published alongside it in Nature is more careful. It reports a fast, single-shot measurement of a property called fermion parity in a nanowire device, a necessary step towards such a qubit, and its authors discuss both ordinary and topological explanations for what they measured. On its own it does not establish that the Majorana states a topological qubit needs are present.

The July 2026 Advantage Claim, and the 37-Minute Reply

IBM had promised quantum advantage by the end of 2026 and defined it strictly: a computation beyond what classical computing can do alone, with a result that can be rigorously checked. On 30 July 2026 it said the bar had been met, pointing to three papers. The headline one, written with researchers at the University of Chicago, introduced doped Clifford sampling: circuits built mostly from operations a classical computer handles easily, seeded with a few hundred harder operations called T gates that make the whole circuit hard to simulate. The structure lets the experiment carry its own error checks and certify, from the hardware’s own measurements, a lower bound on how faithfully it ran.

The checking problem is the heart of the matter. If a quantum computer really does something no classical computer can, then no classical computer can confirm the answer directly. Earlier demonstrations leaned on smaller or simplified circuits that could be simulated, and assumed the large version behaved the same way. IBM’s approach tries to build the check into the experiment itself, so that the machine’s own error measurements set a floor on how trustworthy its output is. As the reply below showed, that cannot stop a cleverer classical algorithm from catching up; it means everyone can agree on what the quantum computer actually did.

The first version of the paper described a 70-qubit circuit, 70 layers deep, with 468 T gates, encoded in 97 physical qubits, with a certified fidelity of at least 0.284. Sixteen days later, researchers at the Singapore University of Technology and Design and NVIDIA posted a reply. By exploiting the circuit’s one-dimensional layout, they computed the exact probabilities of all 2,051 output bitstrings IBM had published, using 256 NVIDIA H100 graphics processors for 37.3 minutes. Their classical estimate of the fidelity, about 0.35, agreed with IBM’s certified bound.

That result cut both ways. It removed the advantage for that particular circuit, and at the same time it independently confirmed that the quantum hardware had performed as well as IBM said. By early September IBM had revised the paper around a different 64-qubit circuit, 73 layers deep with 314 T gates on 76 physical qubits and a certified fidelity of at least 0.349, and the three classical researchers had joined it as co-authors. Two further papers, a 74-qubit simulation of a periodically driven magnet by Qedma, RIKEN and BlueQubit, and a 56-qubit study led by Algorithmiq, argue that leading classical methods gave inconsistent answers where the quantum results held steady. Whether the revised claims survive the same scrutiny is the open question at the time of writing.

The Error-Correction Breakthrough That Matters Most

Beneath the headline speed records sits the achievement that quantum computing has needed all along: error correction. Because qubits are so fragile, a useful machine cannot rely on physical qubits alone. Instead, many physical qubits are woven together to form a single, far more reliable “logical” qubit, with the group constantly checking and correcting its own errors.

For years there was a haunting worry that adding more physical qubits might introduce errors faster than correction could remove them. Willow’s central result was to show the opposite — that scaling up the code made the logical qubit less error-prone, crossing below the critical threshold. That is the moment the field had waited for: proof that the road to large, fault-tolerant quantum computers is genuinely open, not blocked by a wall of noise.

The numbers, published in Nature in 2025, are worth having. On a distance-7 code spread over 101 qubits, Willow’s protected qubit made an error in 0.143 per cent of correction cycles, and each step up in code size cut the error rate by a factor of 2.14. The protected qubit also outlived the best individual physical qubit on the chip by a factor of 2.4. Codes are improving as well: an IBM design published in Nature in 2024 protects 12 logical qubits with 288 physical ones for nearly a million rounds of checks at a 0.1 per cent physical error rate, where the standard surface code would need nearly 3,000.

Drug Discovery: The Promise and the Record

Designing a medicine means predicting how a candidate molecule will behave and bind, and the most accurate predictions depend on the quantum behaviour of its electrons. Chemistry is still the application that justifies much of the investment, for the reason Feynman gave: molecules are quantum systems, and the shortcuts classical chemists rely on are weakest where electrons interact most strongly, as they do in many catalysts. The honest record is short. The closest real demonstration is Google’s 2025 molecular experiment, which its authors say is not yet beyond classical, and Boston Consulting Group concluded in July 2024 that quantum computing then provided no tangible advantage over classical computing in either commercial or scientific applications. The same logic applies to materials science — designing new superconductors, batteries and catalysts — where a material’s properties are set by exactly the quantum electron behaviour that classical computers struggle to simulate at scale.

Where Quantum Computing Will Change the World

Cryptography and cybersecurity. The most urgent near-term implication is cryptographic. Much of internet security rests on problems, such as factoring very large numbers, that are infeasible for classical computers; a large enough quantum computer running Shor’s algorithm could solve them. How large keeps shrinking on paper. In May 2025 Craig Gidney estimated that a 2048-bit RSA key could be broken in under a week by a machine with fewer than a million noisy qubits, down from 20 million in a 2019 estimate he co-authored, assuming error rates of 0.1 per cent. No machine is close to that. Even so, the US National Institute of Standards and Technology published its first three post-quantum encryption standards on 13 August 2024 and urges organisations to start switching now, partly because data harvested today could be decrypted later, a threat known as “harvest now, decrypt later”. Its draft transition plan proposes deprecating RSA and elliptic-curve methods after 2030 and disallowing them after 2035.

Climate and energy. Quantum chemistry simulations could speed the design of catalysts for carbon capture, more efficient solar cells, and better electrolysers for green hydrogen. In many clean-energy technologies the bottleneck is not funding but a lack of precise understanding of the quantum chemistry involved — exactly what quantum computers are built to provide.

Artificial intelligence. Quantum machine learning is still early-stage, but researchers have identified classes of optimisation problems in AI training where quantum methods show theoretical promise, potentially trimming the enormous energy cost of training large models. The convergence with advanced AI and the pursuit of artificial general intelligence could prove especially powerful, with AI helping to design and interpret quantum experiments and quantum hardware accelerating the science underneath AI.

Finance and logistics. Banks and logistics companies are among the earliest commercial explorers, drawn by problems that are essentially giant optimisations — balancing investment portfolios, pricing complex risk, or routing fleets through millions of possible combinations. Quantum approaches to these optimisation and Monte Carlo problems could, in principle, reach better answers faster. No such advantage has yet been shown on a real portfolio or delivery network.

The Honest Picture: What Quantum Computing Cannot Yet Do

It is important to be accurate about where things stand, because the field has a long history of overpromising. Today’s processors are what researchers call NISQ devices — Noisy Intermediate-Scale Quantum machines, a label introduced by the physicist John Preskill in 2018. They are real quantum computers, but imperfect ones. The largest fault-tolerant tasks are expected to need far more: Gidney’s estimate for breaking RSA assumes almost a million physical qubits. The leading superconducting processors have around a hundred high-quality qubits (Willow has 105, Nighthawk 120), and a Caltech neutral-atom array has trapped more than 6,100 atoms with record coherence, though without yet running computations across all of them.

The 2025 and 2026 milestones are genuine and significant. They prove the physics works and the engineering path is tractable. They do not mean quantum computers will replace classical ones, or that they are ready for general commercial use. For most everyday computing, classical machines remain far superior. The realistic near-term future is hybrid: quantum processors handling the specific subroutines where they offer a true advantage, wired into classical infrastructure that handles everything else.

Why Feynman’s 1981 Idea Is Now Reality

In 1981, the physicist Richard Feynman gave a lecture in which he argued that because classical computers cannot efficiently simulate quantum systems, the solution was obvious: build a computer that is itself quantum. He sketched the conceptual outline of such a machine, and at the time it was treated as an intriguing observation from a characteristically provocative thinker.

Forty-five years later, the machines Feynman imagined exist. They are imperfect and limited against the field’s long-term roadmaps, but they are real, operational, and already producing results that the best classical methods struggle, and sometimes fail, to match. The idea has become engineering, and the engineering is becoming technology. Boston Consulting Group estimated in July 2024 that quantum computing could create 450 to 850 billion dollars of economic value by 2040, sustaining a market of 90 to 170 billion dollars for the companies that build the hardware and software. It is a forecast, and quantum timelines have slipped before.

What Researchers Say

Google’s own account of its 2025 molecular experiment says it is not yet beyond classical. IBM’s definition of quantum advantage requires not just a hard computation but one whose result can be rigorously validated, which is why its July 2026 claim was built around a certified fidelity, and why a classical reply that confirmed that fidelity was, in its way, part of the validation. Meanwhile superconducting circuits, trapped ions, neutral atoms, photons and topological designs are all being pursued at once, and that spread of approaches reduces the risk that a single obstacle could halt progress.

There is something quietly astonishing in all of this. To compute with a quantum machine is to press the strangest features of reality — a coin that is both heads and tails, particles linked across space, waves that cancel and reinforce — into practical service. Feynman’s provocation was that nature is not classical, so if we truly want to understand it, our machines had better not be either. Four decades on, that bet is paying off, one fragile, frozen qubit at a time.

Frequently Asked Questions

Sources

• Google — Meet Willow, our state-of-the-art quantum chip
• IBM Quantum — Roadmap and processor announcements
• Microsoft — Majorana 1 topological quantum chip (Feb 2025)
• NIST — Post-Quantum Cryptography Standards (2024)
• Preskill — Quantum Computing in the NISQ Era and Beyond, Quantum 2, 79 (2018)
• DiVincenzo — The Physical Implementation of Quantum Computation, arXiv:quant-ph/0002077 (2000)
• Google Quantum AI — Quantum Error Correction Below the Surface Code Threshold, Nature 638, 920 (2025)
• Google Quantum AI and collaborators (22 October 2025) — Observation of constructive interference at the edge of quantum ergodicity. Nature 646:825-830.
• Google Research (22 October 2025) — A verifiable quantum advantage.
• IBM (12 November 2025) — New quantum processors, software and algorithm breakthroughs on path to advantage and fault tolerance.
• IBM (30 July 2026) — Quantum advantage through trusted quantum computation.
• Martiel, S. et al. (2026) — Sampling hard circuits with verifiably high fidelity, arXiv:2607.25941 (v1 28 July, v3 2 September).
• Manabe, H., Gu, H. & Pan, F. (13 August 2026) — Classical simulation and design frontiers for IBM’s doped Clifford sampling experiment, arXiv:2608.13110.
• Leviatan, E. et al. (2026) — Resolving structure in prethermal Floquet dynamics with precision quantum computation, arXiv:2607.24937.
• Barron, S. V. et al. (2026) — Observable estimation in the absence of classical verification, arXiv:2607.25998.
• IBM (31 August 2026) — IBM Quantum Nighthawk r2.
• IBM (19 August 2026) — IBM connects its first modular cryogenic systems.
• IBM — Quantum roadmap.
• IBM (November 2025) — Scaling for quantum advantage and beyond (Nighthawk: 218 couplers, up from Heron’s 176).
• Pan, F., Chen, K. & Zhang, P. (2022) — Solving the sampling problem of the Sycamore quantum circuits. Physical Review Letters 129:090502.
• Microsoft Azure Quantum (19 February 2025) — Interferometric single-shot parity measurement in InAs–Al hybrid devices. Nature 638:651-655.
• Bravyi, S. et al. (2024) — High-threshold and low-overhead fault-tolerant quantum memory. Nature 627:778-782.
• Manetsch, H. J. et al. (2025) — A tweezer array with 6,100 highly coherent atomic qubits. Nature 647:60-67.
• Gidney, C. (May 2025) — How to factor 2048 bit RSA integers with less than a million noisy qubits, arXiv:2505.15917.
• NIST (13 August 2024) — First three finalised post-quantum encryption standards.
• NIST IR 8547 (initial public draft) — Transition to post-quantum cryptography standards.
• Boston Consulting Group (18 July 2024) — The long-term forecast for quantum computing still looks bright.