Quantum Is A Supply Chain Problem

Key Takeaways

• Error correction stopped being theoretical. Real machines now run logical qubits, in small numbers.
• The bottleneck moved from physics to industry. Refrigeration, helium-3, cryogenic wiring, fabrication capacity.
• Helium-3 costs on the order of a thousand times gold by weight, and a supplier signed a contract to source it from the Moon.
• Governments are taking equity stakes to secure capacity, which tells you how they classify this.
• Read roadmaps as promises. The gap between an announced target and a delivered result is where most of the commentary goes wrong.

Most quantum coverage is a qubit count and an adjective. That is a bad way to understand an industry, because the number that gets announced is rarely the number that constrains it.

I have spent the last year looking at this the way I look at any company: what does it take to build one more unit, who supplies that, and what happens if they cannot.

The answers are surprisingly physical.

First, what actually got delivered

Three results from the last two years are real, published and peer reviewed, and they are why the tone changed.

Google's Willow processor demonstrated below threshold error correction on 105 qubits, meaning the logical error rate roughly halves with each increase in code distance. Adding physical qubits started making the logical qubit better rather than worse, which is the precondition for everything else. Google followed it in October 2025 with a verifiable quantum advantage result on a physics problem. (Google Research)

Quantinuum launched Helios commercially in November 2025 with 98 physical qubits, two qubit gate fidelity of 99.921 percent, and 48 error corrected logical qubits. (Quantinuum)

IBM's Nighthawk arrived the same month, and alongside it IBM demonstrated real time decoding of an error correcting code on FPGAs, fast enough to keep up with the machine, which they described as being a year ahead of their own schedule. (arXiv, October 2025)

Notice what those have in common. None of them is a qubit count. They are about error correction working, being verifiable, and being decodable in real time.

And notice what is not on that list. Microsoft's topological claim from early 2025 remains contested, with Nature's own editorial note saying the results do not represent evidence for topological modes. (APS Physics, March 2025) A contested result is not a delivered one.

Now the boring part, which is the real part

To run a superconducting quantum computer you need to hold a chip near absolute zero, deliver thousands of control signals into that cold space without carrying heat in with them, and read signals back out. That is an industrial problem with a small number of suppliers.

Helium-3. Dilution refrigerators depend on it. It trades somewhere in the range of two thousand to fifteen thousand dollars per litre, which works out to roughly a thousand times the price of gold by weight. (The Quantum Insider, September 2025)

It is not mined. In the US it is recovered as a decay product of tritium, which means supply is a function of a nuclear weapons programme's schedule rather than of demand. (US DOE Isotope Program)

In September 2025, Bluefors, one of the main refrigerator manufacturers, signed an agreement to source up to 10,000 litres a year of helium-3 from the Moon, starting in 2028. (Bluefors)

Read that again. A supply agreement, with delivery dates, for an isotope extracted from lunar regolith. Whatever you think about the timeline, the fact that a serious manufacturer signed it tells you how tight they expect terrestrial supply to be.

Cabling. Each qubit needs control lines running from room temperature into the cold stage. Coaxial cable does not scale: too much physical space, too much heat, too many failure points. One supplier in this niche describes flexible superconducting cabling giving roughly eight times the channel density at similar cost, with a roadmap from 256 channels per loader now to 1,024 in 2027 and 4,096 in 2029. (Physics World, September 2025)

Those numbers are the actual ceiling on qubit count for that architecture. Not the chip. The wiring.

Fabrication. Quantum chips need fabs, and the good ones are busy making something else. Which is why the moves this year have been about owning capacity rather than renting it.

Governments are buying equity, not writing grants

This is the part I find most telling as an operator.

In May 2026 the US Department of Commerce announced letters of intent worth about 2.013 billion dollars to nine companies, including a billion to IBM for a quantum foundry, with a minority non controlling equity stake in each company as a condition of the money. (NIST, May 2026)

Governments do not usually take equity in exchange for industrial support. When they do, it is because they have classified the capability as strategic rather than commercial.

DARPA's benchmarking initiative advanced 11 companies to its second stage in November 2025, and the programme's stated question is whether any approach reaches utility scale by 2033. (DARPA) The UK committed two billion pounds in March 2026, including an advance market commitment to actually buy large scale machines. (NQCC) The EU's quantum strategy funds chip pilot lines and has a Quantum Act on the calendar, with supply chain security named explicitly. (EU Quantum Europe Strategy)

An advance market commitment is not a subsidy. It is a purchase order for something that does not exist yet, which is how you de risk a supply chain rather than a laboratory.

Roadmaps are marketing until they are not

Every major player publishes a roadmap with logical qubit counts and dates several years out. Treat those as intentions.

The useful corrective is the Australian project with PsiQuantum, announced with close to a billion Australian dollars behind it, which by mid 2026 had changed site, changed chief executive, had no major development plan lodged, and had its delivery date reported as at risk. (Forbes Australia)

No physics failed there. Permits, sites and construction did. Which is exactly the point of this post.

What this means if you are not building a quantum computer

Three things, and none of them require a physicist.

1. Your cryptography timeline is set by this industry's timeline, and this industry's timeline is now industrial. Delays in refrigeration and fabrication push the threat out. Better error correction and decoding pull it in. Both are happening at once.
2. The suppliers are more interesting than the headline names. In any deep tech wave, the people selling the picks have better margins and more certainty than the people digging. Refrigeration, cryogenic wiring, control electronics and isotope supply are small markets today with structural demand behind them.
3. Watch what governments buy, not what they announce. Equity stakes and purchase commitments are a stronger signal than strategy documents, and they arrived this year.

The verdict

Quantum computing stopped being only a physics story and became an industrial one at the moment error correction started working. From here, the things that decide the pace are helium, refrigerators, cables, fabs, permits and procurement.

That is less romantic than a breakthrough narrative. It is also how you tell, from the outside, whether a timeline is real.