What is quantum computing? A plain explanation for business

Quantum computing is a way of doing calculations with qubits, physical systems that follow the rules of quantum mechanics instead of the on-off logic of ordinary bits. For a small set of problems, such as simulating molecules or factoring the large numbers behind today’s encryption, a quantum computer could need far fewer steps than any known classical method. For almost everything else it is slower and less practical than the computer you are reading this on. If you only remember two things, make it these: quantum computers are specialists, not faster general-purpose machines, and the first place they will affect most companies is security, not operations.

The rest of this page explains how it works in plain language, what it is and isn’t good for, and where the technology stands in September 2026. You don’t need any physics or maths beyond school level to follow it.

How is a quantum computer different from a normal computer?

A normal computer stores information in bits that are either 0 or 1 and changes them with logic gates, one well-defined step at a time. A quantum computer stores information in qubits, whose state is described by numbers called amplitudes, and its gates change those amplitudes in ways that let different computational paths add up or cancel out.

That cancellation is what matters. Every classical program, however clever, follows paths that can only add probability. A quantum program can make paths that lead to wrong answers cancel each other. Nothing in a classical computer does that, and it is the source of every real quantum speedup.

The idea is older than most people think. Richard Feynman proposed in 1981 that simulating quantum physics would need a computer built from quantum parts. Peter Shor showed in 1994 that such a machine could factor large numbers efficiently, which is what made governments and security agencies pay attention. The engineering took another three decades, and it is still not finished.

What is a qubit?

A qubit is the quantum version of a bit: a physical object with two distinguishable states, labelled 0 and 1, that can also be in a combination of both. In practice it can be a tiny superconducting circuit, a single charged atom held in an electric trap, a neutral atom held by a laser, a photon, or the spin of an electron in silicon.

A useful picture is an arrow that can point anywhere on the surface of a globe. North is 0, south is 1. A classical bit can only point north or south. A qubit can point anywhere: toward the equator, toward Bogotá, toward Tokyo. When you measure it, you ask a yes-or-no question (“north or south?”) and it answers north or south, with a probability that depends on how close the arrow was to each pole. After the measurement, the arrow really is at the pole you got. The in-between direction is gone.

This picture is not a metaphor for something mystical. It is, with some details left out, the mathematics physicists use for a single qubit (they call the globe the Bloch sphere).

What is superposition, without the mysticism?

Superposition means a qubit’s state is a combination of 0 and 1 with specific amplitudes, not that the qubit is secretly one or the other, and not that it is “both at once” in any way you could exploit directly.

The popular image is a spinning coin that is “heads and tails until it lands.” It is misleading in one important way. A spinning coin is really in one definite state that we just don’t know. A qubit’s amplitudes can be negative or even complex numbers, so two amplitudes can cancel. Probabilities for a coin can never cancel. That difference is where interference comes from.

The other thing people get wrong is scale. Two qubits have four amplitudes (for 00, 01, 10 and 11), three qubits have eight, and 300 qubits have 2 to the power of 300, more numbers than there are atoms in the observable universe. This is why classical computers cannot simulate large quantum systems exactly. It is not the same as saying a quantum computer stores or processes that much usable information. When you measure 300 qubits you get 300 bits out. Nothing more.

What is entanglement?

Entanglement is a correlation between qubits that is stronger than any correlation classical objects can have. When two qubits are entangled, measuring one tells you something about the other, and the pattern of results across many measurements cannot be explained by the qubits having carried hidden, pre-set answers.

Experiments confirming this, in the form of tests of Bell’s inequalities, earned Alain Aspect, John Clauser and Anton Zeilinger the 2022 Nobel Prize in Physics.

Entanglement cannot be used to send messages faster than light. Each person looking at their own qubit sees random results; the correlation only appears when you compare notes through an ordinary channel. It is also not a speedup by itself. It is what lets a quantum computer build states that describe many qubits jointly, which is a precondition for useful interference across the whole machine.

Interference: where does the speedup come from?

A quantum algorithm is a sequence of operations designed so that the amplitudes for wrong answers cancel and the amplitudes for right answers add up. At the end, you measure, and the right answer comes out with high probability.

Noise-cancelling headphones are a fair analogy. They play a sound wave shaped to be the exact opposite of the noise, and the two cancel. A quantum algorithm does the same thing to computational paths, except the “noise” is every wrong answer, and the cancellation has to be arranged in advance using mathematical structure in the problem.

That last part is why quantum computers only help with certain problems. Shor’s algorithm works because factoring can be turned into finding the period of a repeating pattern, and quantum interference is extremely good at detecting periods. Grover’s search algorithm works on any search problem, but only gives a quadratic speedup: a search that takes a million steps classically takes around a thousand on a quantum computer. That sounds large, but once you account for how slow and error-prone quantum operations are, a quadratic speedup often disappears in practice.

Why is it wrong to say quantum computers try every answer at once?

Because you only get to read one answer, picked at random. If a quantum computer simply spread itself over every possible answer and you measured, you would get a random guess, which is no better than a coin toss.

The phrase “tries all possibilities in parallel” is the most common mistake in writing about quantum computing, and it leads to wrong conclusions: that quantum computers will speed up any search, crack any password, or solve any optimization instantly. They won’t. The speedup exists only when a problem has structure that interference can use, and finding such structure is hard. There are still relatively few known quantum algorithms with large, proven speedups.

Why are quantum computers so hard to build?

Because qubits are fragile. Any stray interaction with the environment (heat, vibration, electromagnetic noise, a cosmic ray) disturbs the amplitudes and ruins the calculation. This process is called decoherence. Superconducting qubits have to be cooled in dilution refrigerators to about a hundredth of a degree above absolute zero, and even then they typically keep their quantum state for only a fraction of a millisecond.

Operations are imperfect too. The best two-qubit gates today fail roughly once in every thousand operations. Quantinuum’s Helios, launched in November 2025, reported 99.921% two-qubit fidelity, which is among the best published figures. Useful algorithms like Shor’s need billions of operations, so the raw error rate has to come down by many orders of magnitude.

The answer is quantum error correction: spreading one “logical” qubit over many physical qubits so that errors can be detected and fixed faster than they build up. For years the problem was that adding more qubits added more errors than it corrected. In December 2024 Google showed with its 105-qubit Willow chip that this has changed: as the error-correcting code grew from a 3x3 to a 7x7 grid, the logical error rate went down, not up. That result, published in Nature, is why the industry now talks about fault-tolerant machines by the end of the decade.

The physics of superconducting qubits was itself recognized in October 2025, when John Clarke, Michel Devoret and John Martinis received the Nobel Prize in Physics for showing quantum tunnelling and energy quantization in electrical circuits large enough to hold in your hand.

What kinds of quantum computers exist?

There are several ways to build a qubit, and none has won. Each trades speed, quality and scale differently.

Google’s own framing, from its March 2026 announcement of a neutral-atom team in Boulder, is useful: superconducting chips are easier to scale in the number of operations, neutral atoms in the number of qubits. The company is now working on both.

What is quantum computing good for?

Quantum computing is expected to help with three families of problems: simulating molecules and materials, some kinds of optimization, and a set of mathematical problems that includes breaking today’s public-key cryptography. Only the last has a proven, dramatic speedup, and it is the one nobody wants.

Simulation is the most promising business use. Designing a catalyst for fertilizer, a better battery electrolyte or a drug molecule means understanding how electrons behave, which is a quantum problem that classical computers can only approximate. This is where pharmaceutical, chemical, energy and mining companies may see value first, most likely after fault-tolerant machines arrive.

Optimization (routing, scheduling, portfolios) gets the most marketing. The evidence so far is weak: well-tuned classical solvers remain very hard to beat, and several claimed quantum wins have been matched by better classical methods. Our article on whether quantum computing is useful yet goes through the recent claims one by one.

The most credible demonstration to date is scientific. In October 2025 Google reported in Nature that its Quantum Echoes algorithm ran on Willow about 13,000 times faster than the best classical estimate for that task, with results that other quantum hardware can check. It is a physics measurement, not a business application, but it shows the machines can now do something useful that classical computers struggle with.

What is quantum computing not good at?

Most things. A quantum computer will not make your website faster, run your ERP, speed up a database query, or replace GPUs for training most AI models. The reasons are specific.

Loading data is slow. A quantum computer has no efficient way to take in a large dataset, so “big data” problems usually lose any speedup at the input stage.

Reading results is limited. Each run gives you a few hundred bits at most. Problems that need large outputs are a poor fit.

Operations are slow and expensive. A quantum gate takes much longer than a classical instruction, and error correction multiplies the cost. Modest speedups, like Grover’s quadratic one, are often wiped out by that overhead.

Most problems have no quantum algorithm. Sorting, text processing, standard business logic and most statistics have no known quantum advantage at all.

Symmetric encryption is mostly fine. AES with 256-bit keys and modern hash functions are not seriously threatened; the quantum risk is concentrated in public-key cryptography.

The practical model is a quantum processor working as an accelerator next to classical computers, handling one hard piece of a larger calculation, much as GPUs handle graphics or neural networks today.

How do companies use quantum computers today?

Through the cloud. Almost no company buys a quantum computer. IBM offers its processors through the IBM Quantum Platform, Amazon Braket and Microsoft Azure Quantum resell time on machines from several makers, and in 2026 Quantinuum announced access to Helios through Oracle Cloud Infrastructure. Most services include simulators and free or low-cost tiers for learning.

Programming happens in Python with open-source toolkits such as Qiskit (IBM), Cirq (Google) and others. A developer with good linear algebra can run a first circuit on real hardware in an afternoon. Building something useful takes much longer, mostly because it requires understanding which problems fit.

What companies do with this access today is learn, benchmark and prepare. They train a few engineers, test small versions of their problems, and build the classical baselines they will need to judge quantum results later. A quantum training program for teams is often the most useful first spend, and a structured use case discovery keeps those experiments honest.

Where does quantum computing stand in 2026?

The field has moved from “can error correction work?” to “how fast can we scale it?”, but no one has yet built a fault-tolerant quantum computer, and no quantum computer has yet solved a commercially important problem faster than the best classical methods.

The main systems in use are IBM’s Nighthawk and Heron processors, Google’s Willow and Quantinuum’s Helios, which has 98 trapped-ion qubits and supports 48 error-corrected logical qubits. IBM’s public roadmap targets Starling in 2029, a fault-tolerant machine running 100 million operations on 200 logical qubits. Google, Quantinuum and Microsoft also point to 2029 for large error-corrected systems. Our quantum hardware roadmaps article compares the four plans and lists the milestones to watch.

Those dates are targets, not promises. Quantum computing has a history of optimistic timelines, and independent physicists have challenged some high-profile claims, especially Microsoft’s topological qubits. A reasonable reading is that useful, error-corrected machines for chemistry and materials are plausible within five years and uncertain beyond that.

Why does quantum computing matter for security now?

Because a quantum computer large enough to run Shor’s algorithm would break RSA and elliptic-curve cryptography, which protect most web traffic, VPNs, banking systems and digital signatures, and the data being encrypted today may still need to be secret when that machine arrives.

The resource estimates keep falling. In May 2025 Craig Gidney of Google estimated that RSA-2048 could be broken with fewer than a million noisy qubits in under a week, about twenty times fewer qubits than his 2019 estimate. That machine does not exist yet, but attackers can record encrypted traffic now and decrypt it later, a strategy known as harvest now, decrypt later.

The fix is already standardized. NIST published its first post-quantum cryptography standards (FIPS 203, 204 and 205) on 13 August 2024, and its draft transition plan, NIST IR 8547, would deprecate RSA and elliptic-curve algorithms after 2030 and disallow them after 2035. Replacing cryptography across a large organization takes years, which is why this is the part of quantum computing with a deadline. The post-quantum cryptography migration page explains what that work involves.

Where to go from here

If you are deciding what quantum computing means for your company, start with two questions: which of your data must stay secret for more than five to ten years, and which of your hard problems involve chemistry, materials or large-scale optimization. The first determines how urgent your cryptography work is. The second tells you whether there is an opportunity worth tracking. A quantum readiness assessment answers both in a few weeks, and our glossary covers the terms you will meet along the way.