Follow the paper trail; it always leads somewhere. For years, the quantum computing industry published one number on every slide: the physical qubit count. More qubits meant a bigger machine, and a bigger machine meant progress. The receipts tell the story the press release won’t — and the receipts from this August show that number has quietly been demoted. The scoreboard has changed, and the field is now graded on something harder to fake: whether the errors can be pushed down.
The scoreboard, in two documents
The first receipt comes from Quantinuum. Its Helios system, built on 98 physical qubits in an ion-trap architecture, has now demonstrated 48 logical qubits with an average two-qubit gate fidelity of 99.921 percent. The August report puts the logical fidelity in the neighborhood of “five nines” — a number that, when you see it in writing, looks like a typo and is not. The company plans to offer the system through Oracle Cloud Infrastructure, which is the part of the document a business reader should not skip: a system that reaches the market is a system that has cleared the lab.
The second receipt comes from IBM, filed jointly with the University of Chicago on July 30. The result: 70 encoded logical qubits completing 2,415 logical two-qubit operations and 468 logical T-gates. IBM calls this the satisfaction of a basic criterion for quantum advantage. Let me read that phrase the way an investigator reads an official statement — every word is load-bearing. “Basic criterion” is not “advantage achieved”. It is the threshold being crossed, deliberately and in writing.
Put the two documents side by side and the pattern emerges. Two independent labs, two different hardware architectures, two different engineering teams — and both chose to report their progress in logical qubits and gate counts rather than in raw physical bits. One document is an accident; two are a pattern.
Why the metric change is the story
Let me think about what the shift in reporting actually means, because it is easy to wave at it and easy to miss it. A physical qubit is the raw hardware unit — a trapped ion, a superconducting circuit. A logical qubit is the reliable unit that emerges when many physical qubits are arranged so that errors can be detected and corrected. The ratio between them is the cost of error correction, and it is brutal. The industry used to sell physical qubit counts because they were cheap to increase and easy to display. Logical qubits cost real engineering, and they are the only number that a future user actually cares about.
The connection between the two metrics is the real subject of these documents. Helios holds 98 physical qubits and produces 48 logical ones. The overhead is roughly two to one — a favorable ratio for a field where overhead of ten to one has been common in projections. IBM’s 70 logical qubits, by comparison, are the product of an encoding scheme whose physical cost is measured differently. I am not going to pretend the two systems can be compared column by column, because the architectures are not apples-to-apples and no honest document claims they are. What can be compared is the decision both teams made about what to report — and that decision, repeated twice in the same month, is the pattern.
Here is where the chain follows, and where the story stops being about physics. When the field’s leaders start reporting logical qubits and gate fidelity, everyone else in the industry must respond in the same currency. Benchmarks that cannot be expressed in that currency will be treated as marketing. Vendors whose physical-qubit counts look impressive will be asked the follow-up question: how many logical qubits, at what fidelity? The receipts create the standard, and the standard becomes the contract.
Let me correct myself on one framing before I go further, because I opened this piece calling the change a demotion of the physical qubit. That is not quite the accurate word. The physical qubit is not being demoted so much as it is being reclassified — from the headline metric to the cost input. It still determines everything; it is just no longer presented as the result. That is not a small correction in vocabulary. It changes what a reader should look at first when a quantum vendor makes a claim.
And here is the concrete picture that makes the change tangible. I picture an engineer’s screen in a procurement office: one tab open to a vendor slide deck announcing a large physical qubit count, another tab open to a published fidelity table. The engineer is not comparing the numbers on the two tabs as if they were the same unit. The slide deck has stopped being the document that matters. The fidelity table is the document that decides the purchase — and every vendor in the industry knows the slide deck has lost its job. That screen, two tabs, one of them obsolete, is the scoreboard change in its most ordinary form.
The cost buried in the ratio
Let me follow the paper trail one level deeper, into the ratio that sits behind every logical qubit claim. A logical qubit is not a unit you buy; it is a unit you earn, by spending physical qubits on detecting and correcting errors. The overhead is the hidden line in the ledger. Helios turns 98 physical qubits into 48 logical ones — an overhead of roughly two to one, which would have seemed utopian in the era when projections assumed ten physical qubits per logical one. IBM’s scheme reports 70 logical qubits without publishing the same simple ratio, and I will not flatten the two architectures into a single column because the documents do not support it.
But the direction of the ratio is the point, and the direction is unambiguous. The field is not just demonstrating error suppression; it is demonstrating that error suppression is getting cheaper. That is what makes the near-five-nines fidelity figure meaningful. Fidelity at 99.921 percent for a two-qubit gate is not a trophy — it is the reason the overhead can stay small enough to scale. If the errors were high, the price of a logical qubit would be astronomical, and the 2029 targets would be fantasy. The fidelity number is the quiet arithmetic that keeps the timeline alive.
There is a date attached to this arithmetic that I keep returning to. IBM’s Starling target of 200 logical qubits and 100 million operations by 2029 is not a vague ambition; it is a multiplication problem with a deadline. A machine that performs 100 million operations must run gates at error rates low enough that the error-corrected layer holds up across the whole run. That is the difference between a demonstration and a tool. The documents from this month do not prove Starling will arrive on schedule. They prove the error budget is no longer the obvious reason it would fail.
What the market sees in these documents
The commercial layer of the trail deserves its own look, because the documents carry it in plain sight. Quantinuum plans to offer Helios through Oracle Cloud Infrastructure. That one sentence moves the conversation from the lab notebook to the invoice. A system available over the cloud is a system that potential customers can benchmark with their own workloads instead of trusting a vendor’s summary — which is exactly the kind of verification the quantum industry has needed and resisted.
Let me think about what that availability does to the competitive field. Once a 48-logical-qubit machine is reachable through a cloud console, the standard of proof shifts. A rival’s claim of higher physical qubit counts becomes answerable: run the workload, look at the fidelity, compare the logical result. The cloud listing is the public ledger for the whole argument. It turns marketing into a testable proposition, and the industry is better for having it — assuming, as always, that the performance in the brochure survives contact with an actual customer workload.
The Treasury working group closes the loop in a way that completes the document trail. A financial regulator does not create a working group to organize migration to post-quantum cryptography because it expects the machines to stay in the lab. It does so because it has read the same receipts I am reading — logical qubits rising, error rates falling, cloud availability arriving — and drawn the same conclusion: the day when these machines can break what the current systems protect is no longer a thought experiment. It is a scheduling problem with a date on it.
The third document
There is a third document in the chain, and it connects the two hardware receipts to something broader. On August 25, a research group using data from Google’s Willow processor reported that optimizing measurement timing could reduce logical error rates by up to 40 percent. That is not a hardware announcement. It is an operating-system result — an improvement extracted from how the machine is run rather than how it is built. It matters because it shows that the error-suppression story has not one but two levers: better hardware and better control. An industry that can squeeze 40 percent out of scheduling is an industry where the ceiling keeps moving.
The same day carried a different kind of document, from the United States Treasury, which announced a quantum-ready working group established under Executive Order 14412 to coordinate the federal financial system’s migration to post-quantum cryptography. The connection is not a coincidence. The same week that labs publish their strongest error-suppression results, a finance regulator begins organizing against the day those machines get dangerous. The paper trail runs from the ion trap to the clearinghouse, and it does not stop anywhere in between.
What follows the chain into 2029
The documents also contain a date, and dates are the part of a document that cannot be negotiated. IBM has stated its target of delivering the Starling fault-tolerant system by 2029, supporting 200 logical qubits and 100 million quantum operations. Let me set that target next to the July result and read them as a single chain: 70 logical qubits and 2,415 operations today, 200 logical qubits and 100 million operations in three years. That is not a linear climb. That is an order-of-magnitude claim on two axes at once, and it is written down where audits can find it.
I will be honest about what the paper trail does not contain. It does not contain a working large-scale fault-tolerant machine on a commercial schedule. It does not contain a public price list for logical qubit time. It does not contain a signed contract that proves any of the 2029 targets will be met. What it contains is the next best thing an investigator can ask for: a written, dated, public commitment, in a unit of measurement that has just become the industry standard. When a promise is made in the same currency that everyone will use to grade it, the promise is checkable.
That is the quiet power of this particular paper trail. The era of unverifiable claims in quantum computing is not over, but it is on notice. A company that reports physical qubit counts without logical fidelity is now answering a question it did not previously have to answer. A claim about error rates can now be checked against a published methodology. The receipts are out there, they are dated, and they follow a chain that leads from the lab bench to the Federal Reserve’s cryptography agenda.
The verdict, on the record
Let me state the verdict plainly, the way one closes a file. The quantum industry’s scoreboard has changed this month, and the change is documented on both major hardware tracks at once. Physical qubit counts are no longer the unit of credibility; logical qubits and gate fidelity are. Quantinuum and IBM have both filed results in the new currency — 48 logical qubits at near-five-nines fidelity, and 70 logical qubits crossing a defined threshold — and a third result shows error suppression improving through better control as well as better hardware. Meanwhile, a finance regulator has already begun preparing for the day the machines matter to the money system. One document is an accident; six are a pattern. The trail has led somewhere, and the somewhere is this: the competition that matters in quantum computing is no longer over who has the most bits. It is over who can prove the errors stay down.