Latest / Quickly Quantum / July 31, 2026 — IBM's Quantum Advantage Papers Face the Trust Test
Transcript
- Today on Quickly Quantum: IBM says it just cleared quantum computing's real bar — not just speed, but verifiable trust — and the proof involves a supercomputer that burned over half a million CPU-core hours and still couldn't agree with itself. Before that, in the headlines: a silicon chip that runs its own error correction from inside the fridge, a photon handshake linking two totally different kinds of quantum hardware, and a well-timed reminder that full-fledged quantum computers might always be five years away. Welcome back to Quickly Quantum, your daily brief on the quantum frontier. It's Friday, July 31st, 2026. And with IBM having just posted the biggest single-day stock drop in company history this month, today's story is IBM's attempt at a comeback. Let's get into it. First up: HRL Laboratories. They just published in Nature a silicon quantum processor that controls itself — meaning the chip that reads and corrects qubit errors sits right there in the cryostat, the deep-freeze chamber holding the qubits, instead of down the hall in racks of room-temperature electronics. That's a real fix for what's been quietly strangling every qubit technology's scale-up plans: wiring. HRL's device packs a 54-quantum-dot array, configurable into up to 18 exchange-only qubits, built on 200-millimeter isotopically enriched silicon-germanium wafers, and it talks to its control chip through a 296-channel superconducting ribbon cable — essentially a flat highway of wires thin enough to survive the cold without leaking heat onto the qubits. The payoff: control errors ten times lower than HRL's own earlier demonstrations, each operation finished in under a microsecond, and a roughly fivefold drop in errors as they scaled up their error-correcting code, which is exactly the trend you want if this is going to grow bigger. HRL's CEO, Rob Vasquez, put the philosophy simply: quote, 'the technologies that enabled conventional computing weren't just the highest-performing — they were the ones that could be manufactured cheaply and at scale,' unquote. On X, quantum researcher @RussellQuantum put it this way: quote, 'qubit counts were always the distraction. Control overhead was what actually constrained scaling. HRL just moved that hardware cryogenic, placed directly alongside the qubits,' unquote. Here's the twist — HRL is in the process of being acquired by IBM, folding this cryogenic control IP straight into Big Blue's roadmap. And yes, eighteen qubits is still a prototype, with that tenfold gain measured against HRL's own past work, not a head-to-head with rivals. Timed almost perfectly against all that, New Scientist ran a piece today on basically the industry's oldest inside joke: full-fledged quantum computers might always be five years away. The premise is simple — every time the field clears a benchmark, the goalposts just move to define the next one, so 'five years' never actually arrives, it just resets. New Scientist frames it as a genuine tension, not just snark: the milestones are accumulating, year over year, but each one gets waved off as not quite the real thing by the time the next headline lands. We'll flag the usual caveat on this one — it's a single-source item for us here today, so we haven't independently confirmed how New Scientist is characterizing the field's history, but the underlying tension is one anyone can check for themselves. And honestly, that's not a bad discipline for us to hold ourselves to either — every 'first' this show reports on deserves the same question: first, compared to what, exactly? IBM's CEO put an actual date on quantum's payday this week. Arvind Krishna told CNBC's Mad Money he expects a measurable hit to IBM's top and bottom line by 2028 or 2029, and by the end of the 2030s, in his words, quote, 'this is a trillion dollars of value,' unquote. He tied that confidence straight to today's advantage papers, saying quantum computers can already do things, quote, 'better, faster, cheaper' than classical machines can, at least in specific cases. Krishna also used the appearance to play defense: IBM took its largest single-day stock decline on record after a profit warning about delayed customer deals, and he told investors those deals are getting pushed back, not killed — about 40% of the postponed contracts, he said, already closed within three to four weeks. So sit with that for a second: an incumbent under real earnings pressure just handed investors a very specific, very long-dated promise, in the same week its research arm needed a headline to point to. Read the timeline as the pitch that it is. Over in Korea, researchers at Pusan National University and UNIST just published something important for the plumbing of a future quantum internet. They got two different kinds of quantum light sources — a warm cesium atomic vapor cloud and a semiconductor quantum dot — to produce photons indistinguishable enough to interfere with each other, using something called the Hong-Ou-Mandel effect: basically, two photons hitting a beam splitter and bunching together if they're truly identical. That's harder than it sounds. Quantum dots are great at firing off single photons on demand but can't store them, while atomic vapor is a natural memory but a lousier source, so marrying the two — without extra filtering or frequency tricks that add loss — is exactly the kind of hybrid link a modular quantum network needs. Cooling the quantum dot to about 12.5 kelvin tuned its emission to 917.48 nanometers, matching the cesium photon closely enough to hit a spectral overlap of 0.88, and they measured interference visibility of 0.65. It's a lab-scale demo, and scaling from two sources talking on an optical bench to an actual quantum network is a real engineering leap — but it's a legitimate building block for connecting quantum computers that don't speak the same native language. Rounding out today's headlines: France's monthly national quantum update landed, and the headline is homework done well but maybe not fast enough. The French Court of Auditors reviewed the country's National Quantum Strategy and concluded it's been well managed — but warned the international field is moving faster than France's roadmap, with foreign acquisition pressure already picking off domestic companies. On the brighter side, Pasqal kept expanding — new partnerships in South Korea, a Canadian photonic packaging push, and its €50 million Q-PLANET pilot line for neutral-atom chips — while quantum firm Quobly signed a $5 million deal with SEALSQ to build post-quantum security into its silicon platforms. President Macron even named quantum a priority for deeper cooperation with Germany. We'll flag this one as single-source reporting from The Quantum Insider, so treat the roundup as their read on the month rather than settled fact. Every country's writing the same memo right now — 'we're doing fine, but everyone else is speeding up' — and it's worth remembering that IBM, the company at the center of our main story today, is the incumbent that memo is usually written about. Our main story today, and I'm calling it 'the trust bar,' because that's literally the bar IBM says it just cleared: not just quantum computers doing something classical supercomputers can't touch, but proving, with math you can check, that the answer they spit out is actually right. Regular listeners caught the headline version of this yesterday — IBM claiming a new era of quantum advantage. Today the actual substance landed: three separate peer-reviewed papers, real numbers, and open circuits published on something called the Quantum Advantage Tracker, meaning anyone — any lab, any skeptic — can go pull the exact circuits IBM ran and try to poke holes in them. Quick refresher for anyone just joining: quantum advantage is the moment a quantum computer solves a problem faster or better than the best classical supercomputers can manage. We've heard that claim before, plenty of times, and it's mostly fallen apart under scrutiny, because here's the catch nobody advertises: once a quantum computer gets big enough to actually be interesting, you can't calculate the classical answer to check it against — that's the whole point of quantum advantage. So how do you know the quantum computer isn't just generating fancy-looking noise? That's the verification gap, and it's the actual bar this round of results is aimed at. IBM built this on a coalition — the University of Chicago, Israeli firm Qedma, Helsinki-and-Milan-based Algorithmiq, Japan's RIKEN research institute, and simulation company BlueQubit — and ran three demonstrations. The University of Chicago one used 70 logical qubits — qubits built from many physical qubits stitched together with error-correcting code, so they're more reliable but harder to build — running a sampling problem seeded with hard-to-simulate gates, wrapped in something called a spacetime code that let the quantum computer calculate its own mathematically rigorous confidence bound on the answer, without needing a classical check at all. It finished in about fifteen minutes, with error rates ten times lower than the raw hardware underneath it. The second demonstration is the one with the number that actually stopped me: Qedma paired its error-mitigation software with IBM's Heron chip to simulate a 74-qubit magnetic system flipping back and forth over time. To check the result, Japan's RIKEN threw its Fugaku supercomputer — one of the most powerful machines on Earth — at the same problem, burning over five hundred thousand CPU-core hours. And in the hardest regime, the different classical simulation methods disagreed with each other and couldn't produce a consistent answer, while the error-mitigated quantum result kept resolving clean, steady oscillations the whole way through. They even cross-checked pieces of it on a completely different kind of hardware, Quantinuum's trapped-ion machine, and got agreement there too. The third demonstration, with Algorithmiq, ran 56 qubits tracking how information spreads through disordered materials — the kind of physics relevant to batteries and chemical catalysts — again pushing into territory where independent classical methods reportedly couldn't keep up. IBM's research director, Jay Gambetta, put the framing plainly: quote, 'We're in the quantum advantage era... Scientists can trust it. Now it's moving from people benchmarking the systems to using these for science to look at applications,' unquote. And the timing matters — this lands the same month IBM posted its biggest single-day stock drop in company history, after a profit warning blamed on customers delaying software and infrastructure spending. IBM needs this narrative to work. That's the technical core of it — three different problems, three different verification tricks, and in every case a classical method that either couldn't finish or couldn't agree with itself. And all of it comes from IBM's own ecosystem, checking IBM's own homework — which is exactly the kind of setup that deserves real scrutiny. That kind of self-grading is exactly what's burned this industry before — genuinely useful data, published in the open, but also the setup for exactly the kind of claim that hasn't survived scrutiny in the past. Scott Aaronson, the quantum computing theorist, has spent years dismissing comparable bank-and-industry advantage claims with a name that's stuck: qombies — claims that look alive in a press release and then don't survive contact with actual scrutiny. And the scrutiny moves fast: reports surfaced just this month of a laptop solving a different problem that was once assumed to require a quantum computer, which tells you how quickly 'classically intractable' can expire as a claim. Live Science, covering this same announcement, framed IBM's position bluntly: the company has, quote, 'dared others to prove them wrong,' unquote — which is a tell in itself. Nobody dares anybody to disprove a slam dunk. That said, the reaction from people actually working in the field leaned more impressed than dismissive. The crypto-and-markets outlet @beincrypto summed up the headline number on X: quote, 'IBM and the University of Chicago just achieved "trusted quantum advantage," using 70 logical qubits to solve a complex computation in just 15 minutes,' unquote — and that number checks out against IBM's own paper, so no pushback needed there. @QtonicQuantum, who works in quantum cybersecurity, read the verification framework itself as the interesting part, posting on X: quote, 'This one was designed to be attacked, shipped its own counter-tool, and has held for eight months,' unquote — in other words, the self-certifying math held up under people actively trying to break it, which is a genuinely different posture than 'trust us.' So where does that leave us? The honest read is that this round is a real step up from prior advantage claims specifically because IBM published the circuits — the Quantum Advantage Tracker means Aaronson, or anyone else with a grudge and a GPU cluster, can go try to replicate or break these results themselves, rather than taking IBM's word for it. Publishing the actual circuits is a meaningfully higher bar than a press release with a headline number, and that's worth crediting. But peer-reviewed and open to scrutiny is not the same as independently reproduced, and the RIKEN result — half a million CPU-core hours on one of the planet's biggest supercomputers still not agreeing with itself — is simultaneously the most impressive number in this story and the most convenient one. Classical methods disagreeing with each other tells you they struggled with the problem; it doesn't by itself tell you the quantum machine's answer was the right one, and IBM's coalition isn't dwelling on that gap. It's also worth updating something we flagged the last time IBM's quantum bet came up: back then, the read here was that IBM needs quantum to be the answer, and whether the market's actually buying that long-term was still an open question. Today's papers don't close that loop, but they do tilt it — this is the most technically defensible version of IBM's pitch we've seen, precisely because it invites the fight instead of avoiding it. Time for the Hype Check. Here's the case: real peer review, three separate physics problems, open circuits anyone can go probe, and a supercomputer that genuinely couldn't keep up in at least one demonstration — that's more substance than most 'advantage' claims this show has covered. But it's still IBM's own ecosystem grading IBM's own test, the Algorithmiq result cuts off before showing its full classical comparison, and the field's history of over-claiming means independent replication is the only thing that actually settles this. Add it up, and I'm putting this one at a 7. So here's where this goes next: if independent groups actually take IBM up on that dare and the numbers hold, IBM's quantum unit gets to walk into next earnings with something better than a promise — actual science its stock price can lean on — and rivals like Google and Rigetti suddenly have to answer for why they haven't published anything this checkable. If it doesn't hold, if someone finds a crack in the Floquet result or a classical trick nobody tried yet, Scott Aaronson's qombie list gets one entry longer, and every 'advantage' headline for the next year gets read with a lot more side-eye — including IBM's own trillion-dollar promise from earlier in this episode. If today's papers hold up under all that scrutiny, this is the kind of week people look back on — so if you want the update when the replication attempts start rolling in, follow Quickly Quantum wherever you're listening, and if this episode helped you actually understand what 'quantum advantage' even means, send it to the one friend who keeps asking you to explain 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!