The basic unit of quantum information, and why it's so much harder to build than a classical bit.
A qubit (short for "quantum bit") is a two-state quantum-mechanical system that stores one piece of quantum information. In practice, it's a physical object — usually a superconducting circuit, a trapped ion, or a photon — that physicists can put into a controllable quantum state.
That definition sounds clean. The reality is messier: a qubit must be near-absolute zero, shielded from every stray magnetic field and cosmic ray, and measured within microseconds before it loses its state. The fragility is the whole reason quantum computing is hard.
A bit lives in your laptop's transistors. It's implemented as a voltage — around 0 volts reads as 0, around 1.8 volts reads as 1. Billions of them flip billions of times per second. They barely interact with their surroundings, which is why a laptop can run for years without a single bit flipping by accident.
A qubit is a single quantum system that has two distinguishable states, usually labelled |0⟩ and |1⟩. The states could be the spin of an electron, the energy level of a trapped ion, or the direction of current flow in a superconducting loop. What makes a qubit a qubit is that it obeys quantum mechanics — specifically, superposition and entanglement.
The practical difference: a bit is one number, a qubit is a vector in two-dimensional complex space. Two bits hold four possible values, but only one at a time. Two qubits hold four amplitudes simultaneously, and operations on them interfere with each other in ways that have no classical equivalent.
A qubit in superposition is described by two complex numbers, called amplitudes, that determine the probability of measuring 0 or 1. When you measure, you get one or the other with probabilities set by those amplitudes. Before measurement, the qubit is genuinely in both states — this has been verified by Bell test experiments since the 1970s.
For computation, the trick is that you can manipulate the amplitudes with quantum gates, and the amplitudes can interfere constructively or destructively. Algorithms like Shor's and Grover's are designed so the right answer gets constructive interference and the wrong answers cancel out. That's where the speedup comes from.
There are several competing approaches, and the hardware race is essentially a race between them.
IBM and Google use these. A superconducting qubit is a tiny circuit — typically a Josephson junction — cooled to about 15 millikelvin, colder than deep space. Microwave pulses control the qubit's state, and the readout happens by measuring the circuit's resonance frequency. The upside is fast gate times (tens of nanoseconds) and standard lithography fabrication. The downside is short coherence times (around 100 microseconds) and the need for massive dilution refrigerators.
IonQ and Quantinuum use these. Individual ions (usually ytterbium or calcium) are held in place by electromagnetic fields in a vacuum chamber. Lasers cool them to near standstill and manipulate their electronic states. The upside is very long coherence times (seconds to minutes) and identical qubits. The downside is slower gate times (microseconds) and difficulty scaling beyond a few hundred qubits in a single trap.
PsiQuantum and Xanadu use these. Photons are individual particles of light. They don't need to be cooled and can travel through optical fibers. The challenge is that photons don't interact with each other naturally, so two-qubit gates are hard. PsiQuantum bets on massive integrated photonic chips; Xanadu uses a measurement-based approach called "Gaussian boson sampling."
Neutral atom arrays (QuEra, Pasqal) hold atoms in optical tweezers. Microsoft pursues topological qubits, which would be far more stable if they worked — the catch is no one has built a reliable one yet. Each approach has different tradeoffs in coherence, gate fidelity, connectivity, and manufacturability.
Every qubit is fighting a battle against decoherence — the loss of quantum information to the environment. The moment a qubit couples to anything outside its carefully controlled isolation (a stray photon, a thermal vibration, a cosmic ray), its state collapses into a classical value.
This is why quantum computers look so different from classical data centers. The IBM Quantum System One looks like a chandelier of gold-plated cylinders descending from a ceiling — that's the dilution refrigerator. Without it, the qubits would decohere in nanoseconds.
Error correction is the reason today's qubit counts aren't directly comparable to useful compute. IBM's 1,121-qubit Condor processor from 2023 had so much noise that running useful algorithms on all of them was impractical. The actual usable qubits, after error correction overhead, were far fewer. Google's Willow chip in 2024 demonstrated better error correction than previous chips, but practical advantage is still 5-15 years away for most applications.
This distinction trips up most casual coverage.
A physical qubit is the actual hardware — one Josephson junction, one trapped ion, one photon. A logical qubit is an error-corrected qubit built from many physical qubits working together. The ratio depends on the error rate of the physical hardware, but current estimates suggest somewhere between 1,000 and 10,000 physical qubits per logical qubit for useful algorithms.
When IBM announces "1,121 qubits," they mean physical qubits. When researchers talk about needing "millions of qubits" to break RSA-2048, they mean logical qubits — so the actual physical qubit count needed is in the billions. That's the gap between current machines and cryptographically relevant quantum computers.