Quantum computing explained: why qubits defy every intuition

01The spinning coin: what superposition actually means

A classical bit is a coin lying on the table: heads or tails, zero or one, settled and unambiguous. A qubit is a coin spinning in the air. Until it lands, it is not heads and it is not tails - it is in a superposition of both possibilities, described not by a simple maybe but by two numbers called amplitudes, one for each outcome. Quantum mechanics gives each amplitude a size and a phase, and the rules for combining amplitudes are richer than ordinary probability - they can cancel as well as accumulate.

The payoff comes from scale. Two qubits carry amplitudes for four outcomes; ten carry a thousand; fifty carry quadrillions. An n-qubit register represents two to the power n amplitudes simultaneously, and a quantum operation can transform all of them in one pass. Three hundred qubits would describe more configurations than there are atoms in the observable universe - a growth rate the chart below makes visible.

The catch, however, is the source of nearly every misconception about the field: reading the register does not hand you those quadrillions of values. Measurement collapses the superposition to a single classical outcome. The spinning coin lands, and it lands on one side only. Superposition is a computational resource, not a warehouse you can raid.

02Entanglement and the myth of instant communication

Entanglement is the second ingredient, and the most misrepresented. Two qubits are entangled when their joint state cannot be described by describing each qubit separately - the pair holds correlations with no classical counterpart, and measuring one immediately constrains what the other can yield, regardless of the distance between them.

Almost every popular account then takes a wrong turn: if measuring here instantly determines something there, surely we could send messages faster than light. We cannot. The no-communication theorem shows that the local statistics an observer sees - the probabilities of their own measurement outcomes - are unchanged by whatever happens to the distant qubit. The correlations only become visible when the two observers later compare notes over an ordinary classical channel, which obeys the speed of light like everything else in physics.

Entanglement is still genuinely useful. It is the raw material for quantum error correction, a resource that quantum algorithms consume by design, and part of what separates a quantum computer from a very cold classical one. But it is a correlation engine, not a communication device.

03Why measuring destroys the answer you wanted

If an n-qubit register holds two-to-the-n amplitudes, why not just read them all? Because measurement is destructive. Each qubit you measure collapses to zero or one, and it takes the whole register with it. You get one classical bit string out per measurement, and if you prepared a perfectly uniform superposition, that string would be random - all of the enormous parallel representation evaporates into a single sample.

This is why a quantum computer is not a computer that tries every answer at once. The art of quantum algorithms is interference: arranging the evolution of the amplitudes so that paths leading to wrong answers cancel each other out while paths leading to right answers reinforce. The machine does not hide a huge pile of answers behind a curtain; it choreographs amplitudes so the measurement is overwhelmingly likely to land on a useful value.

Designing such choreography is brutally hard, which is why decades of research have produced a short list of algorithms with proven, decisive advantages - and a vast graveyard of problems for which no quantum speedup is known.

04Algorithms that justify the machines: Shor and Grover

Two algorithms carry the field's economic case. Shor's algorithm factors large integers in polynomial time. Factoring is believed to be intractable for classical computers at the scales used in RSA, and RSA-style public-key cryptography underpins a great deal of modern secure communication. A sufficiently capable quantum machine would render those keys breakable - which is why the threat is taken seriously now, even though the machines do not yet exist at anything like the required scale.

Grover's algorithm is more modest and more broadly applicable: a quadratic speedup for unstructured search, examining a space of size N in roughly the square root of N steps. Quadratic is real but not revolutionary - it turns a trillion-step search into a million-step one, not into one step. In cryptography, its effect is to shave effective key lengths rather than destroy whole schemes outright.

Both algorithms exploit interference, and neither uses quantum parallelism as a shortcut to brute-force enumeration. That narrowness is worth keeping in mind whenever a vendor claims a quantum machine will solve logistics or climate modeling next quarter.

05Decoherence: the enemy that is physics, not engineering

The obstacle standing between these algorithms and reality is decoherence. Qubits hold their delicate amplitudes only while isolated from the environment; stray heat, vibration or electromagnetic noise entangles with the qubit and destroys the phase information that interference depends on. A decohered qubit is a spinning coin that has been bumped - it lands early, and its quantum character is gone.

Isolation is why the hardware looks the way it does. Superconducting qubits sit in dilution refrigerators near absolute zero; trapped-ion machines hold atoms in electromagnetic traps; photonic and neutral-atom approaches take different routes to the same requirement. Every platform is a different answer to the same question: how do you let a machine touch a qubit without letting the rest of the world touch it?

The scale of the fix is sobering. Error rates per physical operation remain far too high for direct computation, so quantum error correction bundles many physical qubits into one logical qubit - by widely cited estimates, hundreds to thousands of physical qubits per reliable logical one. A machine that needs a few thousand logical qubits therefore needs physical hardware on a scale no flagship chip approaches today.

06Where quantum stands in 2026 - and what it will not do

By 2026, the field's leading superconducting machines hold on the order of a thousand physical qubits at most. IBM's Condor generation crossed the thousand-qubit mark, and Google's Willow chip demonstrated that error correction can push logical error rates down as more qubits are added - the first time scaling in that direction worked rather than hurt. Neutral-atom arrays have reached thousands of atoms, though operating them as fully connected, error-corrected machines remains a research problem. The chart below shows public headline numbers that tell a subtler story than a simple upward line: Willow has fewer qubits than Condor, because raw count stopped being the metric that matters.

Meanwhile the classical world is not waiting. In 2024, NIST finalized its first three post-quantum encryption standards, and migration has begun against the threat of harvest-now-decrypt-later attacks, in which encrypted data captured today is stored for decryption by a future machine.

What quantum computing will not do is equally important. It will not replace classical computing for everyday tasks, it will not accelerate matrix mathematics the way GPUs already do, and it will not break all cryptography on arrival. It is a specialized instrument for problems whose structure - factoring, unstructured search, simulating quantum systems - matches what interference can express. The spinning coin is real, and it is still learning to land where we want it.