Quantum computers are a new and still experimental kind of computer. They are built for a few problems that ordinary computers cannot solve, such as predicting how a molecule will behave, which could speed the search for new drugs and materials. No quantum computer has solved one usefully yet.
Nobody yet agrees on how to build one. The difficulty is the qubit, the basic working part of every quantum computer. Qubits are fragile, and "a stray electric or magnetic field, temperature fluctuations or even a cosmic ray" can ruin a calculation, says NIST, the federal standards lab. A qubit has to be left undisturbed long enough to finish its work.
So each company must decide what to make its qubits from and how to protect them. The companies have chosen very differently, and no one knows yet which choice will work.
This issue is a field guide to six of those choices. Each entry says what the qubit is, who is building it, what it does well and where it struggles.

The six designs in this field guide, numbered as they appear below, and the companies building each. Source: company filings; DARPA.
1. Superconducting circuits
Rigetti, IBM and Google make their qubits from tiny loops of metal printed on a chip. The chip is chilled to near absolute zero, where the metal becomes superconducting, meaning it carries electricity with no resistance. The cold also keeps out the heat that would disturb the qubits.
The gold chandelier in most photographs of quantum computers belongs to this design. It is the refrigerator, and the chip hangs at the bottom, kept more than 180 times colder than deep space. These qubits work fast, but they hold their information only briefly. One on Google's Willow chip lasts about 68 millionths of a second.

Drawn from published designs, not one company's machine. Source: IBM (2026); Popular Science (2021).
2. Trapped ions
IonQ and Quantinuum use single atoms as qubits. Each atom has one electron stripped away, which turns it into an ion, an atom with an electric charge. Electric fields can then hold the ion in place, floating above a chip in a vacuum, and lasers read and steer it.
These qubits make few mistakes and hold their information for a second or more. Their weakness is speed. On Quantinuum's newest machine, one step that links two qubits takes more than a thousand times longer than the same step on Google's chip.

Drawn from published trap designs. Quantinuum's chip is 0.6 inch long, about 85 percent of a dime's width. Source: Quantinuum; US Mint.
3. Neutral atoms
Infleqtion also uses single atoms, but leaves them whole, or neutral. A focused beam of laser light holds each atom in place, and thousands of beams can hold thousands of atoms in a grid. In March, Google announced a neutral-atom program of its own.
This design's strength is size, and laboratories have already built grids of thousands of atoms. Its weakness is that atoms keep escaping from the grid.

Atoms and hair are drawn to the same scale, using the spacing in a published grid. Hairs vary widely, so the comparison is approximate. Source: Atom Computing; The Physics Factbook.
4. Photonic qubits
Xanadu makes its qubits from light. Particles of light, called photons, hardly interact with anything, so the light needs no cooling and travels well through glass fiber. Xanadu's demonstration machine is built from chips made with mature chipmaking methods, linked by about 8 miles of fiber.
The weakness comes from the same property. Light that hardly interacts with anything is hard to compute with, and some of it is lost along the way. Xanadu calls that loss "the dominant and most challenging hurdle."

The racks, the fiber and the two refrigerators are Xanadu's demonstration machine, simplified. One of its chips is drawn to scale beside the dime. Source: Xanadu's paper in Nature and its announcement (2025).
5. Topological qubits
Microsoft is trying to build a qubit that resists disturbance by design. It says its 2025 chip, Majorana 1, creates a new state of matter, called topological, in tiny chilled wires. The qubit's information would be stored in particles called Majoranas at the ends of the wires, where a stray disturbance cannot easily reach it.
Outside researchers have not confirmed that the chip creates those particles, and a paper in Nature in June argued that the test Microsoft used to detect them is flawed. Microsoft says it stands by its results.

The dashed rings mark what Microsoft claims and outside physicists have not confirmed. Source: Microsoft (2025); Nature (2026).
6. Quantum annealing
D-Wave's machine works differently from the other five. Those run a program one step at a time, and D-Wave's runs no program. It is built for one kind of problem, picking the best option out of a huge number, such as the routes for thousands of delivery trucks.
The method is called quantum annealing. The machine treats every possible plan as a point on a landscape of hills and valleys, where a lower point is a better plan, and it settles into a low valley in millionths of a second. That gives a good answer, though not always the best one. D-Wave said in 2025 that Ford Otosan, Ford's joint venture in Turkey, uses it alongside ordinary computers to set the order in which vans go down its production line.
D-Wave sells these machines today, but whether the quantum part beats an ordinary computer is disputed. D-Wave claimed in the journal Science in 2025 to have beaten ordinary computers at a physics simulation, and outside researchers challenged the claim there in 2026.

The landscape pictures the idea; a real problem has thousands of dimensions. The machine's size is from D-Wave's system specifications. Source: D-Wave.
How close anyone is
No quantum computer has yet been shown to do useful work that an ordinary computer could not. What some have done is beat ordinary computers at test problems chosen for the purpose, a milestone called quantum advantage. IBM said on November 11 that researchers reached it this year. Such claims get challenged, and one that IBM made in July was simulated on ordinary computers about two weeks later.
Useful work depends on error correction, a way of spreading one qubit's information across many qubits so the machine can catch its own mistakes. IBM is aiming for 2029 for its first machine built that way.
Washington's bets
Washington has not picked a design either. In May, the Commerce Department proposed about $2 billion for nine companies, in return for a stake in each, and it has since signed agreements with most of them.
It is also preparing for a risk. A large enough quantum computer could break the codes that protect email and financial records, and encrypted data captured today can be saved until then, a tactic called harvest now, decrypt later. So NIST says it will disallow today's vulnerable encryption after 2035.
DARPA, the Pentagon's research agency, is testing which designs can work. It estimates it is "more likely than not" that at least one company builds an industrially useful quantum computer, one worth more than it costs, by 2033. That leaves seven years to learn which kind of qubit can be left undisturbed long enough to finish the job. The gold chandelier may turn out to be the answer, or it may end up in a museum.
Anthony
The Luniter Observatory
Disclosure: I own shares of IonQ (IONQ), D-Wave (QBTS), Alphabet (GOOG), Microsoft (MSFT) and NVIDIA (NVDA).
IonQ (IONQ), Quantinuum (QNT), Rigetti (RGTI), Infleqtion (INFQ), Xanadu (XNDU), D-Wave (QBTS), Alphabet (GOOG) and Microsoft (MSFT) are among the 200 companies Luniter follows.
Research and educational analysis only — not financial advice. Every investment carries risk.
