10x Means Less Than What It Sounds Like
There’s a number making the rounds in the quantum-computing press, and it tells you almost everything about the state of the industry
Jay Gambetta, Director of IBM Research, posted this number on LinkedIn last week, endorsing a BCG article that urges CEOs to “shape where quantum creates value.” His standout data point: IBM Quantum users have “utilized over 10 times more qubits in 2024 than 2022 for industry relevant applications and the systems are being pushed harder than ever.” He frames it as the field finally asking the harder question — not “can we run a circuit?” but “does this circuit do something useful?”
Great question. Bold of him to ask it in public before checking the answer.
10x. Let me say plainly what that number is, because it is doing the heavy lifting here, and it is lifting absolutely nothing, with excellent posture.
10x is a measure of how much compute people bought and ran. Definitely not a measure of anything that compute produced.
The metric measures demand, not advantage
“Users programmed 10x more qubits” is a usage statistic. It belongs to the same family as “ChatGPT processed 10x more tokens” or “AWS ran 10x more instance-hours.” Or “Americans ate 10x more crab rangoon.” Those numbers tell you a product got more popular. They tell you nothing about whether the product did something no other product could do.
Crab rangoon may be delicious. It has not solved protein folding.
For quantum computing, that distinction is the entire ballgame. The whole premise — the trillion-dollar premise — is that a quantum computer can solve a commercially valuable problem that a classical computer cannot solve in reasonable time. That claim is called quantum advantage. And here is the thing the qubit-utilization number is quietly built to make you forget:
No quantum computer has yet demonstrated a commercially useful advantage over classical hardware. Not one. Not at IBM, not anywhere.
Friends, I wish it was different, but this is just the score. And no amount of redefining “advantage” in more convenient ways to the industry can change this. It’s 0–0, late in the second half, and someone is selling commemorative jerseys. 10x of them.
The headline “supremacy” result from 2019 was contrived sampling tasks with no application, the computational equivalent of being the world’s fastest at a sport no one plays — and they were subsequently matched or beaten by improved classical algorithms running on ordinary machines. IBM’s own 2023 “quantum utility” experiment showed a 127-qubit device producing results beyond exact classical simulation — a real and careful result — but approximate classical methods then caught up within months, which is the quantum-computing equivalent of setting a land-speed record and then watching a guy on a bicycle pull up beside you at the next light.
That is the actual frontier: not “quantum did something useful that classical can’t,” but “quantum did something that classical can still mostly chase down.” And note that the advantage threshold keeps rising partly because classical isn’t done — and IBM’s own device team is the proof.
So when Gambetta writes that qubits got “utilized... for industry relevant applications,” it is conceding the game while changing the subject. More qubits got programmed on problems drawn from industry domains. None of them beat a GPU at a job anyone would pay to have done. Even the strongest-sounding enterprise claim — a bank reporting it “outperformed models we run in production” — turns out to be a comparison against its own undisclosed incumbents, not against the best classical method, and let’s face it, beating your own production models is easy if your production models are bad (not saying they are, but maybe show them and we can assess.)
The GPU, for what it’s worth, did not issue a press release about all this. The GPU is humble like that.
His own next sentence
Here is the part that should end the conversation. Immediately after the 10x headline, Gambetta writes: “the honest reality is that enterprise adoption still has a way to go with production scale solutions that leverage quantum.”
Read those two sentences together, because he put them together. Like a man toasting the bride and then mentioning the divorce in the same breath. Sentence one: users ran 10x more qubits on industry-relevant applications. Sentence two: there are no production-scale solutions that leverage quantum. Both are true. They are only in tension if you let the first one imply value was delivered — which is exactly the implication the post is producing and the second sentence quietly retracts.
That’s the whole rhetorical machine in one post. Lead with a usage metric that sounds like an achievement, then admit a paragraph later that the achievement — a production-scale solution that beats classical — does not exist. He calls the gap “the central challenge of this decade,” which is the most generous possible name for “the part where it doesn’t work.” Here an alternative framing: it’s the thing that has to be true for any of the trillion-dollar numbers to mean anything, and right now it’s a promissory note written on a napkin.
Why the number goes up anyway
The uncomfortable mechanical truth is this: the 10x can rise for reasons that have nothing to do with progress toward quantum advantage:
More users. IBM put more machines on the cloud and onboarded more people. Pure marketing success. More users running circuits means more “qubits programmed,” definitionally. If I open a gym and 10x more people walk in, that is not 10x more fitness. That’s a crowd.
Bigger chips available. When you offer 127- and 156-qubit devices instead of smaller ones, the same experiment “uses more qubits” without being more useful. Order a bigger pizza, eat the same three slices, technically you “utilized” more pizza. Congratulations on your utility-scale dinner.
More experimentation, not more value. The BCG data Gambetta is endorsing shows spend shifting toward algorithm and software development — from 21% to 40% of total quantum spend — because the algorithms don’t exist yet. People are programming more qubits precisely because they’re still searching for something worth doing. That’s a sign of an unsolved problem, the sound of a very expensive room full of very smart people going “hmm.”
In other words, the metric rises fastest exactly when the field is spending the most money looking for a use case it hasn’t found. It is a measure of search effort dressed up as a measure of discovery — the treasure map sold as the treasure.
The “industry relevant” sleight of hand
The magic comes in two words: industry relevant. Read them carefully. Industry relevant means the circuits are drawn from domains where, if there were an advantage, it would matter. It does not mean there is one. Running a molecular-simulation circuit is “industry relevant”; it is not the same as simulating a molecule better than classical chemistry codes already do. The phrase smuggles the conclusion into the noun.
This is the move repeated throughout the genre. Take a real engineering fact (chips got bigger, error correction crossed real thresholds for small surface codes in 2024–25, usage grew) and let the reader’s imagination supply the missing clause: ...therefore commercial value is arriving. The fact is real. The clause is the vibe.
What would actually count
None of this means quantum computing is a fraud. The error-correction demos of the last two years are legitimate physics, and some view a fault-tolerant machine doing useful chemistry by the early 2030s as a defensible bet. But it is a bet, contingent on algorithms we do not have yet, running on hardware we do not have yet, which is a lot of “do not have yet” for something described as “getting real.”
And notice which use case the BCG article leads with: Shor’s algorithm breaking RSA. It's the one genuinely-quantum task with a clear, undisputed payoff — and the payoff? Not value to your company but “your encryption breaks.” So more work and more expenses to your company. The article concedes this itself, calling it “more a security risk than an immediate economic opportunity.” The single most-cited reason quantum computing is supposed to matter is a threat, not a product. Everything on the revenue side of the ledger — the chemistry, the optimization, the materials — is still the speculative part.
When the flagship application is the one you’re defending against rather than buying, “getting real” is doing a lot of work.
The scoreboard is short, and “qubits programmed” is not on it:
A single end-to-end commercial problem where a quantum machine beats the best classical method on time, energy, or accuracy, physical qua computational resources (and not just contingent physical resources) — and stays beaten after the classical side wakes up and responds. (The classical side always wakes up. It is annoyingly well-rested.)
A reproducible economic advantage, not a sampling stunt and not a benchmark that quietly excludes the classical algorithms that already match it. “We won, as long as you don’t count the competitors” is a category of victory mostly available to toddlers.
A result that doesn’t dissolve into “quantum-inspired” — where the quantum research’s main product turns out to be a better classical algorithm running on a laptop. (Note that BCG’s own rosy market figures explicitly “include quantum-inspired solutions.” They’re padding the quantum prize with classical compute.)
Until one of those arrives, “10x more qubits” deserves to be read as what it is: a sales figure. And there is a good reason: IBM’s own roadmap puts real machines at 2029. Before that, there’s no revenue to show — only activity. So “10x more qubits” is what you put on the slide while you wait. It tells you IBM sold more quantum compute to people who used it to do — well, something, presumably nothing groundbreaking, or we’d have heard about it at a volume that makes this number look like a whisper.
And it’s being recited as if it told you quantum compute is worth buying.
So far it’s definitely worth selling.