How do you build a qubit from light, silicon, Majoranas and annealers?

A single photon, a particle of light, can carry a qubit in its polarisation (chapter 2), in which of two paths it takes, or in when it arrives. Photons hardly interact with anything, so they keep their quantum state well, and they travel down optical fibre, so linking machines is natural.

Some companies use a different kind of light state called squeezed light (page 4b), and build error-resistant qubits from it, known as GKP states, after the three physicists who proposed them in 2001.

Here's the catch. Because photons don't interact, getting two of them to perform a two-qubit gate is hard. The leading approach, called fusion or measurement-based computing, works differently from chapter 7's circuits:

1. Make lots of small entangled clusters of photons.
2. Join clusters by sending photons from each into a beam splitter and measuring them. That measurement "fuses" the clusters into a bigger entangled web.
3. Compute by measuring the web, photon by photon, in a chosen order.

Fusions only succeed some of the time, so the design has to tolerate failures. Readout uses single-photon detectors, many of which need cooling to a few degrees above absolute zero.

Strengths: chips made in existing semiconductor foundries, much of the system at room temperature, and natural networking over fibre.

Weaknesses: photons go missing (absorbed or scattered), and gates succeed only some of the time, so machines need huge numbers of components.

Builders: PsiQuantum (US and Australia), Xanadu (Canada), Quandela (France), ORCA Computing (UK) and USTC (China).

• PsiQuantum's Omega chipset (February 2025): 99.22% fusion accuracy. PsiQuantum raised $1 billion in September 2025 and broke ground on a large facility in Brisbane in June 2026. It's one of six companies in the final stage of the US defence research agency DARPA's benchmarking programme (October 2026).
• Xanadu's Aurora (January 2025): a 12-qubit machine spread over 35 chips at room temperature. Xanadu listed on the Nasdaq and Toronto stock exchanges in March 2026.

Take the same material and factories that make your phone's processor. Build a structure so small that it traps a single electron, a quantum dot. That electron's spin (chapter 2) is the qubit: up or down along a magnetic field.

Each qubit is about the size of a transistor, so in principle millions could fit on one chip, using the chip industry's existing machines.

• Gates: microwave signals or electric fields tip the spin; two neighbouring electrons interact when a gate voltage lowers the barrier between them.
• Readout: a tiny sensing transistor nearby changes its current depending on the spin's state.
• Isolation: cold, but some can run a little warmer than superconducting chips: around 1 degree above absolute zero rather than a hundredth of a degree, which would make cooling and wiring easier. Most leading results still use the colder fridges.

Strengths: manufacturing at chip-industry scale, tiny qubits, and slightly warmer operation.

Weaknesses: demonstrations are still small (a handful to a dozen or so qubits), and tiny differences between devices make each qubit behave slightly differently.

Builders: Diraq and Silicon Quantum Computing (Australia), Quantum Motion (UK), Intel (US) and Photonic Inc (Canada).

• Diraq with imec (2025): above 99% accuracy on chips made in a full-scale industrial factory. In October 2026 DARPA picked Diraq for the final, hardware-testing stage of its benchmarking programme (chapter 11).
• Quantum Motion (September 2025): installed a complete silicon quantum computer at the UK's National Quantum Computing Centre.

Every qubit so far is fragile because local noise can disturb it. In 1997 the physicist Alexei Kitaev proposed protecting quantum information through topology, the maths of shapes, and in 2001 he showed how a tiny wire could do it: store the information not in one place but in the shared state of two exotic objects at either end. Physicists call these objects Majorana zero modes. A stray bump at one end can't change the information, because it isn't at either end; it's in the relationship between them. The physics itself would suppress errors.

If it works, a topological qubit would need far less error correction (chapter 8).

Microsoft is the main company pursuing this, and the story has been bumpy.

• February 2025: Microsoft announced Majorana 1, a chip it said used topological qubits. The accompanying Nature paper came with an editors' note saying the results did not show evidence for Majorana zero modes.
• 2025 to 2026: independent physicists, led by Henry Legg at the University of St Andrews, criticised the test Microsoft used. In June 2026 Nature published his critique, which alleges errors in the analysis code and selective presentation of data. Microsoft published a reply saying it stands by its results, and announced Majorana 2, with claims that aren't yet peer reviewed.
• Earlier history: a related team retracted a 2018 Majorana claim in 2021.

Where it stands (October 2026): no independent group has confirmed a working topological qubit. The idea is sound physics; whether these devices achieve it is genuinely open. The US defence agency DARPA still has Microsoft in the final stage of its benchmarking programme. Optional page 10b sets out both sides.

Not in the sense of chapter 7. An annealer doesn't run gates. Instead you turn an optimisation problem, such as "which arrangement has the lowest cost?", into an energy landscape of hills and valleys. The machine starts in a simple quantum state and slowly changes its settings so that it settles towards the lowest valley. Quantum effects such as tunnelling through hills may help it avoid getting stuck.

It can't run Shor's algorithm or general quantum programs. Physicists debate whether it beats the best ordinary methods on real problems.

D-Wave (Canada and US) is the main company. Its Advantage2 became available in May 2025 with more than 4,400 qubits. In March 2025 D-Wave published a claim in Science that it had simulated a magnetic material beyond the reach of ordinary computers. Other researchers challenged it within days, and in May 2026 a team published ordinary-computer simulations of many of the same cases. D-Wave says its hardest cases still weren't reproduced. The claim remains contested. In January 2026 D-Wave also bought Quantum Circuits to build gate-based machines.

A nitrogen-vacancy centre is a tiny flaw in diamond: a nitrogen atom next to a missing carbon atom. Its electron spin works as a qubit, even at room temperature. It's a star of quantum sensing (chapter 12) and a candidate for quantum networks, but so far only small numbers of these qubits compute together.