Latest / Quickly Quantum / August 19, 2026 — IBM's Cooling Milestone: Progress or Roadmap Theater?
Transcript
- Today on Quickly Quantum: IBM says it just connected two modular cryogenic refrigerators and cooled them together to nearly the coldest working temperature quantum hardware needs — but does that actually retire the hardest risk on the road to a fault-tolerant quantum computer, or is it just another checkpoint on a very long list? Before that, in the headlines: a widely-engaged X thread calls the entire quantum sector a scam, Cornell engineers a cheaper way to make the metal at the heart of your qubits, and physicists in China just stretched quantum entanglement across four hundred and twenty kilometers of fiber — more than four times the old record. Welcome back to Quickly Quantum, your daily brief on the quantum frontier. It's Wednesday, August 19, 2026. Let's get into it. A fight's been simmering across quantum-Twitter all week, and both sides are dug in. A widely-engaged post from @ParadisLabs on X argued — quote — 'the more I learn about Quantum, the more of a scam the entire sector seems,' end quote. The post points back to comments the thread attributes to IonQ CEO Niccolò De Masi, paraphrasing him as predicting that quantum processors will eventually replace graphics processors the way GPUs overtook CPUs — that's the thread's characterization of what he said, not a verified quote, so take it as sentiment rather than settled fact. The stronger pushback was about the money, not the qubit count: pre-profit companies posting triple-digit stock gains looks like a bubble dynamic, not a technology signal. One comparison in the thread that stuck with me: quantum investing to the NFT craze. And Nvidia's own Jensen Huang has flip-flopped between calling quantum fifteen to thirty years away and calling it an inflection point, depending on the week — even the chipmaker quantum's supposedly about to dethrone can't settle on a timeline, which tells you something about where the hype-to-reality gap actually sits. Valla Fatemi's team at Cornell just cut a manufacturing temperature in half, and that's a bigger deal than it sounds. Superconducting quantum chips — the kind IBM and Google build — love tantalum, because it resists corrosion and loses very little energy, which helps qubits stay coherent, meaning they hold their quantum state longer before noise wrecks the calculation. The catch: depositing tantalum onto a chip normally requires heating it past four hundred degrees Celsius, too hot for a lot of standard semiconductor foundries' equipment. Fatemi's team swapped the usual sputtering gas — argon — for krypton, heavier atoms that knock tantalum atoms loose more efficiently, and cut the deposition temperature down to two hundred degrees, while producing films with higher electronic conductivity. It's published in Nature Materials, and to be clear, this isn't a better qubit yet — it's a fabrication result that still needs validation inside real foundries before it translates into actual chip performance. But if it holds up, it widens the pool of factories that could build tantalum-based quantum chips without buying specialized high-temperature tools — the kind of unglamorous manufacturing fix that quietly decides whether this industry scales or stays a boutique. Here's one that actually made me sit up: physicists in China just entangled two quantum memories across four hundred and twenty kilometers of optical fiber — more than four times the previous record. The team, led by Xi-Yu Luo at the University of Science and Technology of China in Hefei, published the work in Physical Review Letters. Quick gloss: entanglement links two particles so measuring one tells you something about the other regardless of distance, and a quantum memory is a device that can actually store that entangled state rather than just passing it through. Distance matters here because photons get absorbed by glass fiber at a steady rate per kilometer, so their odds of surviving a long trip fall off exponentially — that's why sending entangled photons directly tops out, in theory, around three hundred twenty kilometers. Luo's team got around that by having each end hold a cloud of ultra-cold atoms that emits a single photon carrying a trace of its quantum state, converted to telecom wavelengths so it loses far less light along the way — only that one photon has to survive the trip to a meeting point in the middle. Past three hundred twenty kilometers, this memory-based approach actually beat the best possible direct-transmission method, for the first time ever shown experimentally. That's the real headline: quantum memories just proved they can outperform brute-force transmission over distance. It's still just two nodes in a lab, and stringing together multiple repeater links at this range is the next hard problem. But it's a genuine step toward a quantum internet that spans countries, not just city blocks. Shifting from physics to Washington: Senators Chris Coons and Mike Rounds introduced the bipartisan Quantum-GUARD Act earlier this month, aimed squarely at the electric grid. The bill would direct FERC — the federal agency setting grid reliability rules — to actually weigh quantum cybersecurity risk in those standards, and it tasks the Department of Energy with building a testing ground for utilities migrating to post-quantum cryptography, the new encryption standards designed to resist a future quantum computer's code-breaking power. It also orders a study of where the bulk power system is vulnerable, across both its information technology and its physical operational technology. Senator Coons put it plainly, saying quantum computing brings economic opportunity but also, quote, 'tremendous cybersecurity risks,' end quote. This builds on a June executive order pushing quantum-readiness across government, now extending that logic specifically to critical infrastructure. But this is a proposal, not law — the bill was introduced and referred to committee on August sixth, and passage and funding timelines are anyone's guess. Still, it's a real signal that the harvest-now-decrypt-later threat — hackers stealing encrypted data today to crack open once quantum computers are strong enough — has bipartisan attention pointed at the grid specifically. Sixty-five point three percent — that's how much NTT DOCOMO, the Japanese telecom giant, says it slashed its daily peak location-registration signaling, according to Quantum Computing Report. The company has put a second commercial quantum application live on D-Wave's hybrid annealing platform — quantum hardware paired with classical computing to solve optimization problems, not general-purpose quantum computing. This one manages how mobile towers hand off devices moving between coverage zones, and DOCOMO says it also trimmed paging signal load by seven percent. The benchmark behind those numbers spanned three hundred thirty-three base stations, and the solver reportedly finished the whole multi-objective optimization in about five minutes. DOCOMO's network overall serves more than ninety-three million mobile subscriptions across Japan, and the company says these optimizations are now folded into its day-to-day network planning. We haven't independently confirmed these figures — they come from the vendor and customer pairing, without third-party benchmarking against the best classical solvers, so treat the percentages as DOCOMO's own reported result rather than settled fact. Still, this is D-Wave's whole pitch: near-term value from optimization, not from gate-model quantum computing chasing textbook algorithms. A live telecom network actually using this daily is a stronger data point than most of what passes for 'production' in this industry. Last quick hit: OTI Lumionics and Samsung's Advanced Institute of Technology say they've published peer-reviewed results in the Journal of the American Chemical Society, and the framing here needs scrutiny. They're calling it a two-hundred-plus logical qubit simulation — a logical qubit being an error-corrected, idealized qubit — but they say it ran entirely on classical hardware: a single thirty-two-core server with eight hundred gigabytes of memory, no quantum computer involved anywhere. It's single-source, from the companies' own release, and it's genuinely useful if the numbers hold up — quantum-inspired algorithms finding real chemistry value on ordinary classical machines. But watch the headline number: two hundred logical qubits describes simulated scale on classical hardware — nobody's built a two-hundred-qubit quantum computer here. Think of it as a target for future quantum hardware to beat, once that hardware actually exists. Our main story today, and here's the tension I want you to hold onto for the next few minutes: I'm calling this the Checkpoint Question. IBM says it just cleared a real engineering hurdle on the way to building the first fault-tolerant quantum computer — but does that milestone actually retire the hardest risk on its roadmap, or does it just move the goalposts to the next hard thing? Here's what happened. IBM connected its first two modular cryogenic systems — refrigerators built to cool quantum chips down near absolute zero, where electrical resistance and thermal noise stop scrambling the fragile quantum states qubits depend on. Combined, the two modules stand more than eight feet tall and wide. IBM says it cooled them jointly down to 4 Kelvin — a temperature where a lot of quantum noise starts to quiet down — in under five days, then continued cooling further to below 15 millikelvin, thousandths of a degree above absolute zero, which is roughly where superconducting qubits actually need to operate. Each module, according to IBM, also offers up to twelve times more wiring space than the company's current widely-used systems — and that matters more than it sounds, because more wiring means more room to route the control signals a much bigger chip full of qubits is going to need. IBM's roadmap has one number attached to it that's been public for a while now: 2029, the year the company has committed to shipping Quantum Starling, its first fault-tolerant machine. Fault-tolerant means the system can detect and correct its own errors faster than they pile up — the thing that would actually let quantum computers run useful, long calculations instead of today's short, noisy demonstrations. IBM's roadmap shifted toward what's called qLDPC codes — quantum low-density parity-check codes, an error-correction scheme designed to protect qubits using far fewer physical qubits per protected 'logical' qubit than older methods. That shift was itself a big bet: a leaner path to fault tolerance, but one that required new hardware architecture to support it — including cooling systems that can scale to many more qubits without turning into an unmanageable tangle of wires. That's the backdrop for today's announcement. Jay Gambetta, IBM's chief quantum officer, framed the connected modules as a step that will accelerate innovation across hardware, software, and algorithms. And this isn't happening in isolation — IBM says this comes just three weeks after IBM and the University of Chicago claimed a quantum advantage demonstration using seventy logical qubits. So in the space of about a month, IBM has stacked two separate, somewhat independent claims — a computational advantage demo, and now a cryogenic infrastructure milestone — both pointing at the same 2029 target. When we last talked about IBM's quantum bet on this show, back in July, the read was straightforward: IBM needs quantum to be the answer, for its stock story and its broader relevance, and whether the market's actually buying that long-term case was still an open question. Today's announcement doesn't settle that — but it adds a data point, and the question is what kind of data point it actually is. So here's the skeptic's case, and I think it's a fair one: this is infrastructure engineering, not a new physics result. The genuinely unproven parts of IBM's 2029 promise are still sitting untouched on the other side of this announcement. Real-time decoding — reading out error signals and correcting them fast enough to keep up with a running computation — at the scale qLDPC codes require hasn't been demonstrated. Neither has magic-state injection across multiple modules, the mechanism IBM needs to actually perform universal fault-tolerant computation once you've built your logical qubits. I'll also say plainly where I think IBM's public tone runs a little hot: the company has at times sounded like it's already cracked the code, and I don't think that's earned yet. Getting two boxes cold together checks one box on a very long list, and I think that assessment is basically right, though I don't think it makes today's news meaningless. Modular cryogenics is a real bottleneck, and it's easy to underrate if you're not the one building this stuff. Today's quantum systems, from IBM and everyone else, are largely limited by exactly this problem — how many wires you can physically route into a single dilution refrigerator before the wiring itself becomes the constraint, not the chip inside it. A twelve-times increase in wiring capacity per module, if it holds up at scale, is the kind of unglamorous number that determines whether you can even attempt the next order of magnitude in qubit count. Solving that wiring problem matters, because you can't scale to the qubit counts qLDPC codes demand without solving it first — and IBM's own phrasing here, calling this 'a milestone toward' Starling rather than proof Starling ships on schedule, is more careful than a lot of the chatter swirling around it. Here's what I keep coming back to, though: in the space of one month, IBM has put out two separate public claims aimed at the same 2029 target — the seventy-logical-qubit advantage demo it claimed three weeks ago, and now this cooling milestone. Each one, on its own, is a real, checkable engineering step. But two announcements that close together, both pointing at the same deadline, also tells you IBM knows how skittish investors have gotten about that date, and it isn't leaving the narrative to chance. I don't think that's spin dressed up as science — you can't fake cooling two eight-foot cryostats to fifteen millikelvin with a press release. But it does tell you IBM is managing the 2029 narrative as deliberately as it's managing the wiring, and the actual proof point — a working, error-corrected logical qubit computing inside a connected modular system — hasn't happened yet. So does today's milestone strengthen or weaken where we landed on IBM's quantum bet back in July — that the company needs quantum to be the answer, and whether the market's buying it long-term is still open? My take: it strengthens the engineering case a bit, and leaves the market question exactly where it was. IBM is visibly doing real work here. Whether that work lands on time, and whether investors and customers stay patient all the way through 2029, remains unresolved — and this announcement, by design, isn't built to answer that. Time for the Hype Check. I'm putting this one at a five. The engineering is real and independently plausible — jointly cooling two large cryostats to below fifteen millikelvin is a legitimate, verifiable achievement. But framing it as a milestone toward fault-tolerant computing does a lot of work that the actual unproven pieces — decoding at scale, cross-module magic-state distribution — haven't earned yet. Solid five: real progress, wrapped in roadmap marketing. If today's episode helped you separate real engineering from roadmap marketing — the IBM story especially — do me a favor and follow Quickly Quantum wherever you get your podcasts, so tomorrow's episode just shows up without you having to think about 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!