Types of Quantum Computers: 5 Ways to Build a Qubit

A qubit is not one object. It is a job. Anything that can hold two states, be nudged by gates, be linked to other qubits and be read out can be a qubit. Many things can do that job. None of them does it perfectly.

Think of how many ways there are to store music. Records, tapes, CDs and files all hold songs. Each has its own strengths. Quantum hardware is like that, except no one knows yet which format will win.

This page walks through the five main types. For each one, you get how it makes a qubit, what it is good at, what it struggles with, and real machines from our catalog. Every figure is copied from our sourced records. Almost all of them are vendor-reported, meaning the maker published them.

How the qubit is made. A superconducting qubit is a tiny loop of metal on a chip. It is cooled until electricity flows with zero resistance. At that point the loop acts like an atom that we designed ourselves. Its lowest energy level is 0 and the next one up is 1. Microwave pulses act as the gates.

What it needs. Deep cold. Google's paper on Sycamore says the chip ran below 20 mK. That is 20 thousandths of a degree above absolute zero, colder than outer space.

Strengths. Gates are very fast. Google lists a two-qubit gate on Willow at about 42 nanoseconds. Chips are made with methods close to normal chip making.

Weak spots. Qubits forget their state fast. Willow's chip 1 lists a T1 of 68 ± 13 microseconds. T1 is how long a qubit keeps a 1 before it decays to 0. Also, each qubit links only to its near neighbors. Far-apart qubits need extra moves, which our lesson on connectivity costs explains.

Examples in our data: Google Willow (105 qubits), IBM ibm_boston (156 qubits, Heron r3), Rigetti Ankaa-3 (84 qubits), and IQM Garnet (20 qubits). A different twist is the cat qubit in Alice & Bob Boson 4, which is built to resist one kind of error.

How the qubit is made. An ion is an atom with one electron removed, so it has an electric charge. Electric fields hold a row of ions floating in a vacuum. Two energy levels inside each ion are the 0 and the 1. Lasers or microwaves act as the gates. Every ion of one kind is exactly the same, so the qubits are all alike.

Strengths. Very low error rates and long memory. Quantinuum's Helios data sheet lists a typical two-qubit error of 8e-4. That means about 8 failures in 10,000 gates. IonQ lists a T1 of 10 to 100 seconds for Forte. Ions can also be moved or linked so that any qubit can reach any other. This is called all-to-all connectivity.

Weak spots. Gates are slow. A peer-reviewed paper on a 30-qubit Forte chain gives an average two-qubit gate time of about 900 microseconds. Compare that with Willow's 42 nanoseconds. Do the arithmetic: 900 microseconds is 900,000 nanoseconds. Divide by 42 and you get about 21,000. So one ion gate takes as long as about 21,000 of Willow's gates. The two figures come from different kinds of sources, so treat this as a rough size, not an exact ratio.

Examples in our data: Quantinuum Helios (98 qubits, barium ions), Quantinuum H2 (56 qubits, ytterbium ions), IonQ Forte (36 qubits), and AQT IBEX Q1 (12 qubits). The course chapter on trapped-ion qubits covers the hardware in depth.

How the qubit is made. Neutral atoms have no charge. So instead of electric fields, focused laser beams hold them. Each beam is called an optical tweezer. Hundreds of tweezers make a grid of single atoms. To link two atoms, lasers push them into a large, puffed-up state called a Rydberg state, where nearby atoms feel each other strongly.

Strengths. Size. Atom Computing reports 1,180 qubits on a 1,225-site array for its second-generation system. Atoms can also be moved in the middle of a program, which changes which qubits can talk.

Weak spots. Runs are slow to repeat. QuEra lists Gemini-class at 1 shot per second. Atoms can also be lost from their traps.

Examples in our data: QuEra Gemini-class (260 qubits), QuEra Aquila (256 qubits, analog), Pasqal Orion Beta (100 qubits maximum, analog), and Infleqtion Sqale (100+ qubits). We go deeper in neutral-atom quantum computers.

How the qubit is made. Here the qubits are made of light. A photon is a single particle of light. A qubit can be stored in which of two paths a photon takes, or in when it arrives. Mirrors, beam splitters and chips with tiny light channels act as the gates. Detectors count photons at the end.

Strengths. Light barely reacts with its surroundings, so it keeps its state well while it travels. Much of the gear can run warm. Xanadu's paper on Aurora says its light-based parts run at room temperature, across four server racks. Light also travels well down fiber, which helps link machines.

Weak spots. Photons do not easily react with each other, so two-qubit gates are hard. Photons also get lost. For many photonic machines, key figures are missing. Our record for Aurora lists its two-qubit gate fidelity as not publicly disclosed. PsiQuantum has no public machine, and its qubit count is not publicly disclosed.

Examples in our data: Quandela Ascella (6 photonic qubits), Xanadu Aurora (12 physical qubit modes per clock cycle, 35 chips), and ORCA PT-2. Some photonic machines are not general computers at all. Xanadu Borealis did one special task called boson sampling. For PT-2, a University of York study found that classical methods did better at the settings they tested. Read more in the chapter on photonic quantum computing.

Yes. The four types above are gate-based. You give them a circuit, step by step. A quantum annealer works another way.

How it works. You turn your problem into a landscape of hills and valleys. The best answer is the lowest valley. The annealer starts all its qubits in a simple state. Then it slowly changes the rules until the qubits settle into a low valley. Think of a ball rolling down a bumpy hill as it slowly cools. D-Wave's annealers use superconducting loops, cooled like other superconducting chips. D-Wave lists its Advantage machine Advantage_system4 at 15.4 mK.

Strengths. Very large qubit counts for this one kind of task. D-Wave reports 4,400+ qubits for Advantage2. Each qubit is coupled to 20 others.

Weak spots. It runs only this one kind of problem. It cannot run a normal quantum circuit such as Grover's search. Gate fidelity does not apply, and D-Wave does not publish it. Also, as of 2026 there is no shown quantum speedup on real-world optimization. See quantum optimization.

So never line up an annealer's qubit count against a gate machine's. They are counting different things. The chapter on annealing vs gate-based explains why.

This table is a plain summary, not a score. We do not rank these types, and we do not name a "best" machine. The right choice depends on your circuit. A circuit that links many far-apart qubits may suit an all-to-all ion machine. A circuit that needs many fast runs may suit a superconducting chip.

Start with the same question for both: what does my circuit need? Then look at the numbers that matter for it. Check how each number was measured. A "best pair" figure and a "median" figure are not the same thing.

Our compare page puts machines side by side and warns you when two numbers used different methods. Our guide to reading hardware specs shows what each field means. For a full tour, the Inside Quantum Processors course covers every type in depth.