Live signals Morning Signal 07:00 GMT

Become a member
Quantum Computing

Superposition in quantum computing: what it means and what it doesn’t

A qubit in superposition carries amplitudes, not two answers at once. How that powers quantum algorithms, why errors matter and where hardware stands in 2026.

Christopher Parkmond
By Christopher ParkmondUpdated 7 min read
A blurry image of a computer mouse and a keyboard

Superposition means a qubit’s state is described by two numbers, called amplitudes, which set the odds of reading 0 or 1 when the qubit is measured. It does not mean a quantum computer tries every answer at once and reads off the best one; useful algorithms work by steering those amplitudes so that wrong answers cancel and right ones become likely.

That distinction explains both what quantum computers could do well and why they are so hard to build. Hardware details below are as of 3 November 2026.

What a qubit’s state actually is

A classical bit is either 0 or 1. A qubit’s state is written as α|0⟩ + β|1⟩, where α and β are complex numbers called probability amplitudes. As Microsoft’s quantum training material sets out, a measurement finds 0 with probability |α|² and 1 with probability |β|², and every valid state must satisfy |α|² + |β|² = 1.

Two points matter here. First, a measurement always returns a plain 0 or 1, and after it the qubit is left in that state, so you never read the amplitudes directly. Second, because amplitudes are complex numbers, they carry a phase as well as a size. Phase has no effect on a single measurement, but it decides how amplitudes add up or cancel when operations combine them.

For one qubit, every possible state can be drawn as a point on the surface of a sphere, the Bloch sphere. The north and south poles are |0⟩ and |1⟩; points elsewhere are superpositions, with latitude setting the measurement odds and longitude setting the phase. Gates are rotations of that point. The picture is useful but limited: it does not extend to several qubits, where entanglement appears.

Classical bits and qubits compared

Property Classical bit Qubit
State 0 or 1 α|0⟩ + β|1⟩, with complex amplitudes α and β
Reading it Returns the stored value without changing it Returns 0 or 1 at random, with probabilities |α|² and |β|²; the superposition is lost
Copying Can be copied freely An unknown state cannot be copied (the no-cloning theorem)
Describing n of them n values Up to 2ⁿ amplitudes in general
Correlation between units Classical correlation only Can be entangled, with correlations no classical system reproduces
Main source of error Rare bit flips, easily corrected Bit flips, phase flips and gradual loss of coherence
Error correction Copy and vote Spread one logical qubit across many physical qubits and measure parity checks

Why “0 and 1 at the same time” misleads

The popular line that a qubit “is 0 and 1 at once”, and that n qubits therefore test 2ⁿ answers in parallel, gets the size of the state right and the conclusion wrong. It is true that describing n qubits can take 2ⁿ amplitudes. But a measurement gives you only n bits, chosen at random according to those amplitudes. Put every possible answer into an equal superposition and measure, and you simply get a random answer.

A better description is that a superposition is a single, definite state with no classical equivalent, and an algorithm’s job is to arrange operations so that most of the probability ends up on useful outcomes.

Interference and entanglement: how algorithms use superposition

Interference is the mechanism. Because amplitudes have phases, two routes to the same outcome can reinforce each other or cancel out, much as waves do. Algorithms are designed so that paths leading to wrong answers cancel and paths leading to right ones add up.

Entanglement is what makes the state space large and useful. When qubits are entangled, the state of the group cannot be written as separate states for each qubit, and measurement results are correlated in ways classical bits cannot match. Without entanglement, a set of qubits could be simulated efficiently on an ordinary computer.

Two well-known algorithms show the pattern:

  • Grover’s search (1996). Lov Grover showed that a quantum computer can find a marked item among N unsorted entries in about √N steps, against roughly N/2 checks classically. Each round flips the phase of the target’s amplitude and then reflects all amplitudes about their average, so the target’s probability grows step by step. The gain is quadratic, not exponential.
  • Shor’s algorithm (1995). In his paper on polynomial-time algorithms for prime factorisation and discrete logarithms, Peter Shor showed that a quantum computer could solve both problems efficiently. It uses superposition to evaluate a function over many inputs, then a quantum Fourier transform, where interference concentrates probability on values that reveal the function’s period. Those two problems underpin RSA and elliptic-curve cryptography.

Decoherence and error correction

Superpositions are fragile. Any unwanted interaction with the environment, from heat, stray fields or imperfect control pulses, leaks information about the qubit and blurs the phases that interference depends on. This is decoherence. Today’s physical qubits make errors far more often than useful algorithms can tolerate, so the field’s long-term plan is quantum error correction: encode one “logical” qubit across many physical qubits and repeatedly measure checks that reveal errors without reading out the protected data.

Error correction only helps if physical error rates are below a threshold; above it, adding qubits adds more errors than it removes. Google Quantum AI’s “Quantum error correction below the surface code threshold”, later published in Nature, reported that on its Willow chip a distance-7 surface code using 101 qubits had a logical error rate of 0.143% per correction cycle. Each step up in code size cut the error rate by a factor of about 2.14, and the logical qubit outlasted the best physical qubit by about 2.4 times.

Error-corrected qubits also need special resources, called magic states, to run a full set of operations. On 15 October 2026 Google reported in Science that a technique called magic state cultivation, run on Willow, cut errors in these states about 40-fold and reached a logical fidelity of 99.99%.

Where real hardware stands, as of 3 November 2026

Google Quantum AI

Willow has 105 qubits. Beyond the error-correction results above, Google reported in October 2025 a “Quantum Echoes” experiment on Willow that it described as a verifiable beyond-classical result, estimated at about 13,000 times faster than a classical supercomputer for that specific task. That is Google’s own estimate, for a physics benchmark rather than a commercial workload.

IBM

IBM’s roadmap, published in June 2025, targets Starling in 2029 at a new data centre in Poughkeepsie, New York: 200 logical qubits running 100 million quantum operations. The same plan listed Kookaburra, its first module for storing and processing encoded information, for 2026, and Blue Jay, with 2,000 logical qubits, after Starling. IBM is building around a different family of error-correcting codes (qLDPC) from Google’s surface code.

On 30 July 2026, IBM and the University of Chicago claimed a quantum advantage using 70 logical qubits, 2,415 logical two-qubit operations and 468 logical T gates, with logical error rates about ten times lower than physical ones. The run took about 15 minutes. The claim rests on the team’s own assessment of classical difficulty; the circuits were posted to an open, community-run tracker so others can try to beat them classically.

Quantinuum

Quantinuum, which uses trapped ions, launched Helios on 5 November 2025 with 98 physical qubits, a two-qubit gate fidelity of 99.921% and 48 error-corrected logical qubits. On 19 October 2026 it reported a logical fidelity of 99.9995% on Helios using a colour code with up to 37 physical qubits per logical qubit, and said its fault-tolerant Apollo system is expected by 2029. The company describes this as a record; that comparison is its own.

What it means for encryption

No quantum computer today can break RSA or elliptic-curve cryptography. The concern is the future, and estimates of what it would take keep falling. In May 2025 Google researcher Craig Gidney estimated that 2048-bit RSA could be factored in under a week with fewer than one million noisy qubits, down from 20 million in a 2019 estimate. That assumes a 0.1% gate error rate and a one-microsecond error-correction cycle, well beyond the size of any current machine.

Data intercepted now could be decrypted later, so the migration has already started. On 13 August 2024 the US National Institute of Standards and Technology published its first three post-quantum standards: FIPS 203 (ML-KEM) for general encryption, and FIPS 204 (ML-DSA) and FIPS 205 (SLH-DSA) for digital signatures. These run on ordinary computers and are designed to resist attacks from quantum ones.

The short version

A qubit in superposition holds amplitudes that set the odds of each outcome, not two hidden answers, and algorithms use interference and entanglement to push those odds towards the right result. The hard part is keeping those amplitudes intact; as of late 2026 Google, IBM and Quantinuum have each shown error-corrected qubits that perform better than the physical qubits they are built from. IBM and Quantinuum both target large-scale, fault-tolerant machines by 2029.

Sources

More in Quantum Computing