Temperamental machinery and cosmic rays¶

Quantum computers are the most temperamental machinery humanity has built since someone decided steam engines were a good idea. They require temperatures colder than interstellar space, isolation from electromagnetic interference, and protection from cosmic radiation their classical counterparts barely notice. A classical computer survives being dropped and doused in coffee; a quantum computer experiences existential collapse from stray photons, thermal fluctuations measured in millionths of a kelvin, and the occasional cosmic ray passing through the laboratory.

Building one resembles maintaining the Unseen University’s High Energy Magic building, where reality is thin and anything might happen at unfortunate moments. Theory says quantum computers should work. Hardware says the universe has opinions about that, and the gap between the two contains most of what makes the field simultaneously fascinating and infuriating.

Colder than space¶

Most quantum computers use superconducting qubits, which only behave quantum-mechanically near absolute zero, not metaphorically near but around 15 millikelvin, colder than interstellar space by a factor of two hundred. Reaching that needs a dilution refrigerator, an apparatus resembling a gold-plated chandelier designed by someone who hates maintenance engineers. Cooling takes days, the system runs continuously because warming up loses more days, and every material inside must be chosen for how it behaves at temperatures where ordinary metals turn superconducting or magnetic. A single refrigerator costs hundreds of thousands of euros, consumes kilowatts continuously, and needs increasingly scarce helium, all before a processor goes inside it. Ion-trap and other approaches avoid the extreme cooling, but superconducting qubits currently lead because, refrigeration notwithstanding, they have advanced furthest.

Shouting over the noise¶

Quantum computers make mistakes constantly: gates are imperfect, qubits decohere, noise couples in, cosmic rays strike. Without correction, a computation dissolves into garbage after a few thousand operations. Quantum error correction fixes this by encoding one logical qubit across many physical qubits, so errors can be caught without measuring the logical qubit directly, but the overhead is brutal: hundreds or thousands of physical qubits per logical one, and physical error rates low enough that correction helps rather than harms. This is why today’s machines are called noisy intermediate-scale quantum devices, too small and error-prone for correction to buy much, so algorithms are designed to tolerate noise and produce approximate answers. A useful machine needs thousands of logical qubits, meaning millions of physical ones, which is a great many more than the dozens and hundreds that exist.

Theoretically better, practically theoretical¶

Topological qubits are quantum computing’s fusion power: always promising, perpetually a few years away, and potentially revolutionary if they ever work at scale. The idea is beautiful, encoding information in global properties of exotic states that local noise cannot disturb, which would slash the error-correction overhead. The difficulty is that the exotic states, Majorana zero modes in semiconductor-superconductor hybrids, have proven extraordinarily hard to create and verify, with more than one celebrated result later retracted or reinterpreted as noise. Meanwhile the whole apparatus stays exquisitely sensitive to its surroundings: a cosmic ray can flip qubits, electromagnetic interference leaks through every cable, and a lorry passing the building can couple in enough vibration to matter. This sensitivity is not a temporary engineering problem awaiting a clever fix. It is fundamental to quantum mechanics that quantum states are fragile, and the engineers who keep these machines running develop a philosophical resignation, calibrating meticulously, shielding carefully, and accepting that decoherence eventually wins. The goal is to extract useful computation before it does.

The clerk’s brief¶

July 2025: Europe declares a quantum ambition¶

On 2 July 2025 the Commission published the Quantum Europe Strategy, aiming to make Europe a quantum leader by 2030 and to build a sovereign quantum ecosystem across research, infrastructure, skills and dual-use technology, with a European Quantum Act proposal promised for 2026. The framing is the same sovereignty argument that runs through the chips file: eliminate structural dependencies before they harden. The clerks note that a strategy is not a machine, and that this file will judge the ambition by what gets built, not what gets announced.

June 2025: A roadmap to a real one¶

On 10 June 2025 IBM set out a roadmap to a large-scale fault-tolerant machine by 2029, codenamed Starling, aiming at around 200 logical qubits by switching from surface codes to quantum LDPC codes that cut the physical-qubit overhead. It is a plan with intermediate processors due each year, not a delivered computer. The clerks note that quantum roadmaps have a long history of sliding, and have pencilled 2029 in soft lead.

February 2025: A topological claim, and a chorus of doubt¶

In February 2025 Microsoft announced Majorana 1, a chip it said carried the first topological qubits, by press release and without the supporting data. The physics community pushed back hard: an accompanying Nature paper fell short of demonstrating the effect, its own editors noting the results were not evidence of Majorana zero modes, and one physicist called a topological qubit in 2025 a fairy tale. The clerks record this less as a milestone than as a specimen, the exact moment a vendor announcement outran its peer review, and keep it as a cautionary exhibit for the hype file.

December 2024: Below the threshold, at last¶

In December 2024 Google reported that its Willow chip had driven logical error rates down as it scaled up, a 105-qubit device crossing the long-sought “below threshold” line where adding qubits reduces rather than increases the error rate, published in Nature. This is the genuine article: the central problem of error correction, unsolved since 1995, shown to be solvable in principle on real hardware. The clerks mark it as the most important entry in the file and note, in the same breath, that it still computes nothing anyone needed.

The calibration log¶

The log records a real trajectory and a stubborn distance. The below-threshold result of December 2024 removed the deepest doubt, that error correction might never work on hardware at all; IBM’s June 2025 roadmap set a credible if slippery course to a fault-tolerant machine by 2029; and Europe’s July 2025 strategy signalled that the continent intends to build rather than only buy. Against all of it stands the topological detour, whose February 2025 flagship the field largely declined to believe, and the standing fact that no entry here is yet a useful computer. The clerks’ assessment is unchanged from the section’s founding: the machinery advances genuinely and slowly, the useful machine remains years out, and the only file where the threat is already present is the security one, for reasons that have nothing to do with whether these machines ever get built on time.