The race map: how to read the news
Lesson 19 of 23 in the course “Quantum”. Free, with animations and a quiz.
Who builds on what
Let’s put the whole level on one map. Superconductors: Google, IBM, Rigetti and China’s USTC. Ions: Quantinuum and IonQ. Neutral atoms: QuEra, Atom Computing and France’s Pasqal. Photons: PsiQuantum and Xanadu. Silicon spins: Diraq, Quantum Motion, SQC. Topology: Microsoft. There’s no winner yet, and that’s normal: in the 1950s people also argued about what to make transistors from. What matters is that each technology has a weak spot, and now you know which.
China in the race
China isn’t falling behind. In March 2025 the USTC team published results in Physical Review Letters for the Zuchongzhi 3.0 chip with 105 superconducting qubits. Using 83 of them, it solved a random-circuit task in seconds that, by their estimate, would take the Frontier supercomputer about 6.4 billion years. Sounds huge, but it’s the same genre as Google in 2019: a record on a task designed specifically to be hard for classical machines. It doesn’t do useful calculations. China is also strong in quantum communication and has the Jiuzhang photonic machines.
Numbers that beat qubit count
The headline “1,000 qubits!” says almost nothing on its own. Remember IBM’s Condor: 1,121 qubits in 2023, and then IBM itself switched to smaller but more accurate chips. Look at four things. Two-qubit gate fidelity: 99.9% means one error per thousand. How many of the qubits are logical. Whether the system runs below threshold, i.e. gets better as the code grows. And how many operations in a row it survives: thousands, millions or billions. These numbers separate progress from marketing.
Records on useless tasks
Why stay calm about loud “supremacy” claims? Remember Sycamore in 2019: 200 seconds versus “10,000 years” on a supercomputer. By 2022 researchers with a new classical algorithm solved the same task in about 15 hours on 512 ordinary graphics cards. Classical computing gets smarter too, and random-circuit records regularly get “eaten.” So the industry’s real goal is tasks with a useful, checkable answer. Google’s Quantum Echoes in 2025 is closer to that: its result can be checked against a real experiment.
How much money is here
Now a sober look at the money. According to McKinsey (2025 report), quantum computing companies together earned just $650–750 million in 2024. That’s less than the revenue of one mid-sized IT company. Meanwhile, investment in quantum start-ups grew 50% in a year, to nearly $2 billion. The forecast for the quantum computing market in 2035 is $28 to $72 billion. Note the range: the top figure is 2.5 times the bottom one. That’s what a market looks like when expectations still exceed sales.
A checklist for any headline
The level’s takeaway is a tool you’ll use every week. See a quantum headline? Ask five questions. Physical or logical qubits? What fidelity, how many operations? A useful task or a benchmark record? Peer-reviewed, or just a press release with a preprint? Already done, or only on a roadmap? Try it on Majorana 2: a preprint, not peer-reviewed, no working qubits yet, so interesting but early. Willow below threshold: published in Nature, detailed measurements, peer review passed, so it’s serious.
Check yourself
- A company announces “1,000 qubits!” What should you ask first?
Correct answer: What’s the gate fidelity and how many are logical. Qubit count without fidelity or logical qubits means almost nothing: the 1,121-qubit Condor wasn’t a breakthrough. - Why doesn’t a random-circuit record prove a computer is useful?
Correct answer: The task is picked to be hard for classical machines but has no use, and classical sometimes catches up. Random circuits are a fair test of power, not useful work. Sycamore was “caught” in 15 hours on 512 GPUs. - Which sign points to the most reliable news?
Correct answer: A peer-reviewed paper with data that others confirmed. Peer review and replication by others are the best filter. Willow’s below-threshold result passed review in Nature; Majorana 2 hasn’t yet.