Latest / Quickly Quantum / Quantum's TSMC Problem: IBM's $1B Foundry, No Customers — September 17, 2026
Transcript
- A billion dollars from the government, matched dollar-for-dollar by IBM — and the question nobody's answered yet: will a single other quantum company actually trust IBM to build its chips? That's our main story today: Anderon, IBM's quantum foundry spinout, just finalized a billion-dollar government award, matched by IBM, to scale a wafer fab in Albany, and whether it becomes the industry's neutral supplier or stays IBM's own private plant is genuinely up in the air. Before we get there, in the headlines: an AI model that writes quantum circuits on its own and cuts synthesis time way down, Riverlane setting up shop in Maryland, IQM landing its first machine in South America, and a Stuttgart lab making an atom hold its shape for eleven milliseconds — a record, and a strange one. Welcome back to Quickly Quantum, your daily brief on the quantum frontier. It's Thursday, September 17, 2026. Let's get into it. First up: teaching an AI to write quantum circuits instead of hunting for them by trial and error. IonQ — which builds its machines from trapped ions, individual charged atoms held in place by lasers and electromagnetic fields — teamed up with Oak Ridge National Lab, NVIDIA, and the University of Tennessee Knoxville on a generative model called DQAOA-GPT that spits out optimized circuits directly instead of tuning parameters step by step. The results, presented at IEEE Quantum Week and awarded best paper: on a hundred-variable optimization problem, the old trial-and-error approach saw its runtime balloon from thirty-four seconds at four qubits to over eleven minutes at twelve qubits. DQAOA-GPT held flat at about twenty-eight seconds no matter the qubit count, and it roughly doubled solution quality along the way. Now, before you get too excited — this is a simulation result on optimization benchmarks, run on NVIDIA hardware, not a change to any real chip's qubit count or error rate. But does it generalize beyond this narrow slice of optimization problems? That's the open question, and if the answer's yes, it's a real software-side unlock for hybrid quantum-classical work. Next: quantum error correction — the software and control layer that catches and fixes the mistakes quantum chips inevitably make — just got a new U.S. address. Riverlane is opening its American headquarters in Maryland's Discovery District, right next to the University of Maryland campus, adding to its existing Boston presence. CEO Steve Brierley called it a way to put the company, quote, "at the heart of an exceptional ecosystem of researchers, industry leaders and government partners." The move comes with a research partnership with UMD — student fellowships, mentorship, workforce development — and lands Riverlane near neighbors like Microsoft and IonQ, who already call Maryland home. This is a press release, not a technical milestone — no headcount, no investment figure, no timeline attached. But real estate follows money, and Maryland's quietly become one of the places companies want a flag planted. IQM's on a roll this year. The company just signed a deal to put its first machine in South America — an IQM Spark system heading to Brazil's Eldorado Research Institute in Campinas, near São Paulo, arriving in the first quarter of 2027. The agreement also throws in remote access to IQM's fifty-four-qubit cloud system back in Europe, so Brazilian researchers can prototype locally and scale up when a problem outgrows the on-site hardware. IQM CEO Jan Goetz said, quote, "Brazil has the talent, the research base and the ambition to build a real quantum ecosystem. What's been missing is direct access to real quantum computers." It's the latest stop in a busy year for the company — an order backlog of one hundred two million euros as of August, twenty-six systems sold, seventeen delivered worldwide. Now here's a genuinely fun one. Researchers at the University of Stuttgart just set three world records at once for circular Rydberg atoms — an oversized version of an atom where the outer electron gets forced into a stable circular orbit, one of the candidate building blocks for quantum simulators, machines that model quantum systems and are considered stepping stones to full quantum computers. Researchers at the University of Stuttgart's 5th Institute of Physics — headed by professor Tilman Pfau — got the circular state to hold stable for eleven milliseconds — more than twenty times longer than comparable states get in free space — with the electron orbiting at about 1.1 micrometers across, roughly ten thousand times bigger than an ordinary atom's orbit. They also trapped the atoms with a laser for one hundred thirty-three milliseconds, another record. And here's the part that matters for cost: they did it at room temperature, no liquid helium cooling required, by shielding the atoms from stray microwave radiation with conductive plates. It's published in Nature Communications, and it's a fundamental physics result, not a working qubit — nobody's built a multi-qubit device out of this yet. And a smaller one to close out the headlines: Italian quantum computing firm Planckian was picked as one of thirteen companies for the EuroHPC Quantum Grand Challenge, Europe's push to fund a spread of domestic quantum hardware bets. The award: three hundred thousand euros to help scale its architecture. Is three hundred thousand euros enough to matter on its own? Probably not — that's a rounding error next to the billion-dollar figures coming up next, and that's sort of the point. This is early-stage seed money, one of thirteen small wagers Europe's placing to see which quantum approaches actually pan out. It's part of the same instinct behind the much bigger story we've got today out of the United States. Which brings us there. Our main story today: the trust problem sitting underneath one of the largest government-and-private manufacturing bets quantum computing has seen to date. Anderon LLC — a subsidiary of IBM — has finalized a one billion dollar direct funding agreement with the U.S. Department of Commerce under the CHIPS and Science Act. Supported by an additional one billion dollar co-investment from IBM, the capital will expand operations at Anderon's three-hundred-millimeter wafer fabrication facility located at the Albany NanoTech Complex in Albany, New York. Let's back up for a second on what a foundry actually is, because it matters here. In the classical chip world, a foundry is a factory that manufactures chips designed by other companies — think TSMC making processors for Apple or Nvidia without designing any of them itself. That separation between design and manufacturing is what let the semiconductor industry specialize and scale. Quantum computing doesn't have that yet. Every working quantum computer today — IBM's, IonQ's, Google's, Rigetti's — is built by a vertically integrated company that designs its own chip and fabricates it in-house or through a partner it fully controls. Anderon's stated ambition is to become that missing neutral layer for quantum: what the release calls an "open commercial pure-play foundry for the global quantum hardware ecosystem." This deal finalizes a letter of intent IBM and Commerce signed back in May 2026. Right now, the wafers running through the Albany line support superconducting qubit arrays — the chip architecture IBM itself uses — along with high-density input-output signaling, laser-annealed Josephson junction tuning, and cryogenic readout components. Anderon's roadmap says it wants to expand beyond superconducting chips into other qubit types too, including silicon photonics and spin qubits. Now, the money alone makes this one of the largest single government-plus-private bets on quantum hardware manufacturing to date, and it's landing the same week EuroHPC's Quantum Grand Challenge — the same program that just funded Planckian — handed out grants to thirteen firms across Europe. Everyone building quantum hardware, it turns out, is worried about the same bottleneck: getting enough working chips off a production line. So that's the setup. The real test is whether anyone besides IBM actually shows up. Now here's where I want to slow down, because the press release answers one question — how much money — but ducks the one that actually decides whether this works. As of today's reporting, no other qubit vendor has yet said it will actually route production through a foundry owned by a direct competitor. The wafers running through the Albany line right now are IBM's own — superconducting qubit arrays for IBM's own chips. The "pure-play" framing, meaning a foundry that only manufactures for others and doesn't design a competing product of its own, is the goal. It isn't yet the reality. And that's a structural problem TSMC never had. When TSMC opened, it wasn't Apple's chip-design division moonlighting as a foundry — it was founded specifically to not compete with its customers. Anderon is the opposite: it's owned outright by IBM, one of a small handful of companies on Earth with a real, working, competitive quantum computer. Would IonQ, or Rigetti, or Google, hand its most sensitive chip designs to a foundry owned by IBM's quantum division? Nobody's answered that in public yet, and there's an obvious commercial reason they might not want to. There's also a structural detail the release skips that jumped out at me: no public equity-stake arrangement has been disclosed for this deal. Compare that to Intel's CHIPS award last year, where the government took roughly a ten percent equity stake as part of the terms. Whatever the reasoning here, Anderon's money is coming in without that same kind of disclosed government ownership stake — which changes who's actually exposed if this doesn't pan out. Here's what would actually move me: a named customer — a real, non-IBM qubit company on the record saying its chips are coming off this line — or hard yield numbers on working qubits, not just wafers processed. Without either of those, twelve to eighteen months out, "open commercial pure-play foundry" reads as marketing language wrapped around an internal fab with federal money attached. There's a broader thread here too, and it's one I've flagged before with IBM: the company needs quantum to be the answer to where its future growth comes from, and whether the market's actually buying that long-term bet is still very much open. Owning the foundry that might someday supply the whole industry is a hedge against quantum not paying off any other way — if IBM's own chips stumble, it can still make money renting out the fab. That's a reasonable strategy. It's also exactly the kind of strategy a company announces before it's proven. So here's the honest scorecard: real capital, real wafers running through an actual three-hundred-millimeter line in Albany, zero named outside customers, and a trust problem baked into the ownership structure that TSMC was specifically designed to avoid. Time for the Hype Check. That's a 4. If today's foundry story raises the same question for you it raised for me — who actually trusts IBM with their chip designs — follow Quickly Quantum wherever you're listening, so you catch the answer whenever a customer finally shows up. 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! I also host Concrete Compute: a daily briefing on the AI buildout. The datacenters, the megawatts, and who actually pays for them. Find it wherever you get your podcasts.