Latest / Quickly Quantum / July 19, 2026 — The NISQ Trap: Was Quantum Advantage Doomed From the Start?
Transcript
- Today on Quickly Quantum: did eight years of 'quantum advantage' headlines have to fail — every single time — and is the Department of Energy about to walk straight into the same trap, just with a federal budget and a 2028 deadline stamped on it? That's the fight that broke out this week between a philosopher of physics and arguably the most influential quantum complexity theorist alive, and as of this recording, it's still going, in public, on the record. No headlines today — this is a one-story Sunday, so we're giving it the whole show. Welcome back to Quickly Quantum, your daily brief on the quantum frontier. It's Sunday, July 19, 2026. Let's get into it. So here's the shape of the argument, and I want to give you the whole board before we start picking sides. Back in 2018, physicist John Preskill coined a term that's basically run the whole industry ever since: NISQ, which stands for Noisy Intermediate-Scale Quantum. The idea was simple, and honestly kind of humble — we don't have the error correction yet to build a fully fault-tolerant quantum computer, one where errors get suppressed faster than they pile up, so instead, let's use the noisy, uncorrected chips we've actually got and go hunting for some narrow problem where they beat classical computers anyway. Call it quantum advantage. For eight years, that's basically been the business model of the entire field — chip after chip, headline after headline, claiming some version of 'we did something a classical computer can't.' Now, a new preprint from Amit Hagar, a philosopher of physics at Indiana University, argues that this entire strategy was, in his words, built to lose. His paper is literally titled 'The NISQ Trap: Eight Years of Demonstrations the Hardware Was Built to Lose,' and the claim isn't just that the results were disappointing — it's that they were structurally guaranteed to be. His argument, in short: the same conditions that let noisy hardware run a circuit with enough fidelity to even finish — low effective depth, strong mathematical structure, qubits that only talk to their neighbors — are the exact same conditions that let a classical algorithm compress and reproduce that circuit efficiently. Same door, both directions. And this landed at a very specific moment, because in June 2026, two executive orders committed the Department of Energy to deploying what they're calling the world's first fault-tolerant, scientifically relevant quantum computer, by 2028. The DOE's own request for information spells out the target: somewhere between 150 and 250 logical qubits — those are error-corrected qubits, built by bundling many noisy physical qubits together so the bundle behaves more reliably than any one of its parts — running a universal instruction set, executing circuits with at least 105 of what they call 'hard' operations, things like T gates or Toffoli gates, at a logical error rate of 10 to the minus eight per operation. That's the number everyone's now arguing about, because Hagar says this is the same loop, just funded by taxpayers this time. So — does he have a case? Let's walk it. Start with Hagar's own case, because it's more careful than 'quantum computing doesn't work' — that's not what he's saying. His argument is about pattern, not physics. He goes back through more than thirty advantage-class announcements since 2018 and argues that, with a single contested exception, every flagship 'quantum advantage' claim in the NISQ era has, within about eighteen months of its announcement, either been reproduced on a classical computer, been shown to secretly rest on structure a classical algorithm can exploit, or gotten formally closed by a simulability theorem — a proof that a classical machine can, in principle, do the same job efficiently. He backs this with six separate theoretical results from 2024 through 2026 on classical simulability. And here's the part I think is genuinely sharp: Hagar traces the root of this all the way back to 1996, to the original quantum error-correction threshold theorems — the math that first proved fault tolerance was possible if your error rates were low enough. His argument is that NISQ was never a parallel strategy to fault tolerance, it was a retreat from it, an attempt to get useful results before those threshold conditions were actually met. And because the threshold conditions weren't met, he says, the field ended up only able to demonstrate exactly the circuits simple enough for the hardware to survive — and those same circuits, by definition, tend to be simple enough for classical computers to crack too. It's not bad luck striking eight years running. It's the same structural ceiling, showing up every single time. Now, to his credit, Hagar doesn't dress this up as unfalsifiable dogma. He's explicit that the pattern could break — all it takes is one demonstration that actually escapes the current simulability results. But he also flips the burden of proof: after eight years and more than thirty tries, he argues it's now on the defenders of NISQ to produce that demonstration, not on him to keep explaining away the misses. That's a genuinely uncomfortable frame if you've spent a career inside this field, because it turns every future advantage claim into a test the skeptic already predicted would fail — which, if you're the one making the claim, is exactly the kind of frame you want to break, and exactly the kind of frame Scott Aaronson showed up to break. And break it is what Aaronson tried to do, fast, in a post on his blog Shtetl-Optimized titled 'NISQ and quantum supremacy did not fail.' Aaronson is a complexity theorist at UT Austin, and if you've followed this field at all, you know his name carries weight — he's one of the people whose math the whole 'quantum advantage' claim rests on. His core objection is narrow but pointed: he says Hagar is straightforwardly wrong about the sampling-based supremacy experiments specifically — the Random Circuit Sampling and BosonSampling demonstrations. In Aaronson's words, those experiments passed the point about two years ago where, absent a genuine breakthrough in classical algorithms, they quite clearly are beating what can easily be simulated on any existing classical computer. And here's where the argument actually gets technical enough to matter: Aaronson accuses Hagar of overreading the classical de-quantization results — specifically an October 2025 paper that both of them cite, and reach opposite conclusions about. Aaronson's read: yes, there's a classical algorithm that can approximate these sampling experiments in theory, but it still needs time that's exponential in the circuit size — meaning on paper it's a mathematical improvement, but in practice, on an actual computer, it's still hopeless for circuits of any real size. Polynomial versus exponential is the whole ballgame here, and Aaronson's saying Hagar collapsed that distinction. But — and this matters for how you should read this whole fight — Aaronson isn't dismissing Hagar as a crank. He writes, quote, 'so it seems worth stating for the record that I have an extremely different view,' which is measured, almost collegial, for a public rebuttal. And he draws a very deliberate line between Hagar and the field's other major skeptic, noting approvingly that Hagar, quote, 'never suggests that we've learned anything to rule that goal out' — meaning Hagar isn't claiming quantum computing is impossible, just that this particular eight-year strategy for proving it works has been a bust. That distinction is the hinge the whole debate turns on, and it's also exactly where our third voice steps in, because unlike Hagar, this one really is arguing the deeper goal itself might be unreachable. That third voice is Gil Kalai, a mathematician at Hebrew University and Reichman University, and he's been making a version of this argument since long before NISQ was even a word — going back to the early two thousands. Kalai's position is more radical than Hagar's, and it's worth being precise about the difference: Hagar is saying the NISQ strategy for demonstrating advantage was doomed by structure. Kalai is saying the underlying goal — a large-scale, fault-tolerant quantum computer, period — may be fundamentally impossible, because of something called correlated noise. Here's the plain-English version. Quantum error correction works by spreading information across many physical qubits, betting that errors on individual qubits are mostly independent of each other, so you can detect and fix them faster than they accumulate. Kalai's conjecture is that entangled qubits — qubits linked together in the way quantum computers require — might experience correlated errors instead, meaning the noise doesn't stay independent, it moves together in ways current error-correction schemes aren't built to catch. If he's right, that's not a hardware-quality problem you engineer your way out of with better chips. It's a structural limit on the whole approach. Now, Aaronson himself has engaged with Kalai's argument for years, and his summary of it is pretty blunt — he's said it's maybe more accurate to describe Kalai as starting with quantum computation being impossible as his axiom, then working backwards to figure out what kind of correlated noise would have to exist to kill error correction and vindicate that starting point. That's Aaronson characterizing the reasoning as backwards from the conclusion, not forwards from the evidence — about as sharp a methodological jab as you'll hear between two people who clearly respect each other's math. So why does Kalai matter for the Hagar-Aaronson fight specifically? Because if Hagar's eight-year track record really is as one-sided as he claims — every advantage demo falling within eighteen months — that pattern is at least consistent with something like Kalai's correlated-noise story being true, even though Hagar himself never goes there and Aaronson explicitly credits him for not going there. It doesn't prove Kalai right. But it means the softer, structural critique and the harder, foundational critique are sitting right next to each other in the same conversation this week, whether anyone intended that or not. Now, let's bring this back down to the thing with an actual budget attached: the DOE's 2028 target. And here the debate splits into two very different kinds of skepticism — one hopeful, one from inside the tent. The hopeful read comes from the roadmap side — IBM, QuEra, IonQ, Microsoft, the companies actually responding to the DOE's request for information. Their argument is basically that 2028 isn't a repeat of the NISQ loop, it's the field's actual exit from it, because for the first time the target is a hard, auditable fault-tolerance milestone — logical qubit counts, not a cherry-picked sampling task. And it's not a made-up number: the DOE's target of logical qubits 'numbering in the low hundreds' lines up with where the current record holders already are. QuEra claims 96 logical qubits, Quantinuum claims 94, and the low-hundreds threshold already sits on the 2028 roadmaps at QuEra, IonQ, and IBM. So from that chair, this looks less like federal money chasing a mirage and more like federal money aligning with where the frontier was already heading. But then there's the skepticism from people inside the process itself — and this is the one that stopped me. Jay Sau, a physicist at the University of Maryland, has pointed out that hitting the logical-qubit count doesn't necessarily buy you anything scientifically, because those logical qubits, even at 150 to 250 of them, won't be perfect. They'll be somewhat more stable than the noisy physical qubits underneath them, but still noisy — and how much more stable remains, in his words, a very open question. Sau's line on this is about as direct as it gets: he says, quote, 'I cannot think of a question that a quantum computer can help with in 2028, unless there is a serious breakthrough in device quality.' That's not Hagar's structural-loop argument, and it's not Kalai's correlated-noise argument either — it's a much narrower, much more concrete worry from someone presumably rooting for this program to succeed: that you can hit every number on the DOE's request for information and still not have a machine that does anything nobody's classical computer could already do. Which is, if you squint, a slightly different flavor of the exact same trap Hagar's whole paper is named after. So where do I come down on this? Here's my honest read: I think Hagar's pattern is real and underappreciated, and I think Aaronson's rebuttal is right on the narrow technical point he's actually making. Those aren't contradictory. Hagar's claim about the broad universe of advantage announcements — the ones built on optimization problems, on simulation tasks with hidden classical structure — holds up pretty well against eight years of retractions and reproductions. But Aaronson's specific defense of Random Circuit Sampling and BosonSampling is also, on the numbers he cites, correct — polynomial-time-in-theory is not the same as beatable-in-practice, and collapsing that distinction the way Hagar's framing sometimes does is a real weakness in the paper. So the honest scorecard is: Hagar wins on the broad pattern, Aaronson wins on the hardest specific case, and neither of them has actually resolved anything about Kalai's deeper claim, which is a separate and, frankly, scarier question that this fight doesn't settle either way. What actually worries me is the DOE piece, because that's where the abstract fight turns into an accountability question with a calendar on it. If Sau's right that logical qubits in the low hundreds can still be too noisy to answer any real scientific question, then hitting every number on that request for information — the 150 to 250 logical qubits, the universal gate set, the 105 hard operations at 10 to the minus eight error rate — could still land the DOE exactly where NISQ landed: a technically impressive machine with no actual advantage attached to it. That's not a failure of physics. That's a failure of what the goalposts were measuring in the first place, and it's the one part of this whole debate where Hagar, Aaronson, and Sau are all circling the same worry from different directions. Time for the Hype Check. Here's the case: this is a real, technical, still-unresolved fight between credible people, over a genuinely important pattern in the field's own track record, landing right as federal quantum spending commits to a hard 2028 deadline that could either break the loop or just rebrand it — that's substantive, not manufactured drama, though nobody on any side has proposed a clean, pre-registered test that would actually settle who's right. My number: an 8 out of 10 on substance. And that's really the open question I'm sitting with going into next week: is there any experiment either side would agree in advance counts as the tiebreaker? Because right now, both camps are citing the same October 2025 paper and walking away with opposite conclusions — and until that changes, this loop, whatever you want to call it, isn't closing anytime soon. If an episode like this is the kind of deep dive you want more of, following the show costs you nothing, and it's exactly how more people find arguments like this one before they show up everywhere else. And if you've got a take on Hagar versus Aaronson, I'd genuinely like to hear it. This has been Quickly Quantum, an AI-voiced podcast, created and built by a real human using today's cutting-edge technology. Nothing you heard on this show is financial advice. I'm Brian Lampert, and I'll catch you all tomorrow — take care!