Latest / Quickly Quantum / 104 Qubits Simulated, and the Hard Part Still Runs on a Supercomputer — October 20, 2026
Transcript
- NVIDIA says its researchers have simulated a 104-qubit fault-tolerant quantum processor on JUPITER, the largest supercomputer in Europe, built at Jülich Supercomputing Centre. The striking part is the claim that the simulated approach uses 90% fewer physical qubits than previous methods; the complication is that this is a simulation, not a quantum computer demonstrating the result in hardware. What does that tell us about the distance between a promising design and a useful machine? Today we’ll separate the computational achievement from the hardware still to be built. Welcome back to Quickly Quantum, your daily brief on the quantum frontier. It's Tuesday, October 20, 2026. Let's get into it. Quantinuum says its Helios quantum computer reached 99.9995% logical memory fidelity using a color code, equivalent to five errors per million operations, according to The Quantum Insider. Logical memory fidelity measures how well error correction protects stored quantum information; this result is ten times lower in error rate than Quantinuum’s own world record achieved last month. The company says logical errors fell exponentially as it increased code distance, the measure used to understand how many physical qubits are needed to protect the information, and each tested increase reduced remaining errors by a factor of 12. That is a concrete gain in error suppression, though this is a memory result, not a claim about every operation a useful computer must perform. The Quantum Insider reported that on Helios, Helix used 10 physical qubits per logical qubit, compared with up to 37 for the color code, while the color code delivered lower logical error rates. Quantinuum says the result puts its accuracy performance within the target range for Apollo, its planned universal fault-tolerant quantum computer expected to be available by 2029. And how far does that move the security question people often attach to quantum progress? Security commentator Marin Ivezic wrote on X, “Quantinuum announced a quantum error-correction record on Monday. It is a great result, but in my view it does not bring a CRQC / Q-Day any closer. Despite what some comments and media proclaim.” I think that distinction is persuasive: a better logical memory result matters to the engineering path, while the post itself points out that it doesn’t establish a near-term cryptographically relevant machine. The evidence here is a real error-correction milestone, and I’m glad to see it measured on commercial hardware. On the public-investment front, the UK Department for Business, Innovation, Science and Trade and UK Research and Innovation announced a £78 million ($103.8 million USD) package, according to Quantum Computing Report. It’s aimed at commercializing quantum sensing, building quantum-networking supply chains and developing workforce skills. The allocation includes £50 million ($66.53 million USD) in direct innovation grants for commercial sensing and positioning applications, £20 million ($26.62 million USD) for supply-chain component development in quantum networking, and an £8 million ($10.65 million USD) expansion of the QuEST Apprentices Programme. That apprenticeship expansion is intended to train 150 specialist quantum and engineering technicians by 2030, the report says. Separately, EPSRC opened a £17 million ($22.6 million USD) funding opportunity for integrating AI and high-performance computing with quantum computing; awards are up to £3.75 million ($5 million USD). For you, the notable feature is the mix: the package backs near-term sensing and networking work as well as skills, while the separate opportunity targets quantum-classical integration. The report also says the £78 million combines funding streams; I’d treat the headline as a package, not as one single pot of new cash. Meanwhile, another government-backed bet is extending a company’s research runway: Xanadu has secured up to USD $10 million in additional DARPA funding, Quantum Zeitgeist reports. The contract extends support from DARPA’s Quantum Benchmarking Initiative for one year, allowing the company to accelerate its research and development roadmap. Xanadu’s photonic approach uses light for quantum processing, and the article says its Aurora quantum computer demonstrates networkability, which the company believes is a foundation for future fault-tolerant systems. The article also describes the architecture as operating at room temperature. Those are company-positioned advantages; the funding itself doesn’t demonstrate that a fault-tolerant machine is ready. Dr. Christian Weedbrook, Xanadu’s founder and chief executive, said continued participation in the initiative reinforces the company’s goal of developing quantum computing solutions with practical value across industries. I think the most useful signal here is the additional year of R&D support: DARPA is continuing to evaluate the approach, while the technical outcome remains ahead of the funding announcement. Turning from an extension to a reported program milestone, The Quantum Insider says Alice & Bob has shared Phase 1 results from France’s PROQCIMA program. We covered the company’s GPU decoding work earlier this month; this update is about hardware demonstrators and measured building-block performance. Alice & Bob says it assembled and operated three independent demonstrators, two on 6-qubit chips and one on a 12-qubit chip. The tests included noise-biased CNOT gates, which are controlled operations designed to preserve the cat qubits’ built-in protection against bit flips; repeated cycles of a three data qubit, distance-3 repetition code; and a non-Clifford gate, an ingredient needed for universal fault-tolerant computing. The Quantum Insider reported that its measured logical error rate was not yet below threshold, the point at which error correction begins to reduce errors as the code scales. The company also built a digital twin—a computer model based on measured processor performance—and says its simulations identify consistently longer qubit coherence across the chip as a priority. Phase 1 is a meaningful hardware step with further work ahead on error rates. And on the company-formation side, Quantum Computing Report says Beijing-based startup FlipFlop Tuoyuan has emerged from stealth with hundreds of millions of RMB (USD $14M+) in early-stage financing to commercialize neutral-atom quantum processing units. Its stated medium-term target is a 100-logical-qubit processor, using a neutral-atom Rydberg tweezer-array architecture. The distinction between that target and a demonstrated system matters: the company is setting an ambition, while the report describes a newly emerged startup and its financing. Quantum Computing Report says the company is also pursuing an “AI-native” operating approach, using autonomous AI agents for experimental design, real-time data processing and hardware control. I’d file this under a new entrant with a defined architecture and a large stated goal; the target gives us something specific to measure against as the work develops. That's the rest of today's quantum news. Now, the story I want to slow down for. NVIDIA is arguing that large-scale simulation can help researchers make better choices before they build quantum hardware. A quantum computer’s qubits are vulnerable to noise, meaning small errors that can disrupt a calculation; fault-tolerant computing uses error correction to keep those errors from overwhelming the computation. NVIDIA announced that researchers used JUPITER at Jülich Supercomputing Centre for what the company calls a record-breaking simulation. The simulation modeled a 104-qubit fault-tolerant quantum processor, and NVIDIA describes it as the largest comprehensive simulation of such a system to date. It ran on a system of NVIDIA GH200 superchips. The “record” is NVIDIA’s characterization. The specific technique is magic-state cultivation. Magic states are specially prepared quantum states that are continually consumed during fault-tolerant computation; the process is important because producing them repeatedly can be a major resource demand. NVIDIA says its simulation validates an approach that requires 90% fewer qubits than previous methods. That is the company’s approach-level claim, not a statement that the entire future computer uses 90% fewer qubits. Why put a supercomputer to work on this? NVIDIA says memory needs for these simulations can grow exponentially as the number of qubits rises, placing large simulations beyond conventional computing systems. Its researchers developed their methods on an NVIDIA GB200 NVL72 system at the NVIDIA Accelerated Quantum Research Center, scaling tests from four GPUs to hundreds before moving to 8,192 GPUs on JUPITER. NVIDIA says this work reduced a weekslong simulation effort to days. The point is to explore how realistic hardware noise affects a design before building the quantum hardware itself. Earlier research from Jülich simulated 50 qubits under ideal, noiseless conditions; this work simulates magic-state cultivation under realistic hardware noise. That adds a more practical ingredient to the model, because real devices make errors. It still models the processor rather than running the quantum computation on a 104-qubit quantum processor. NVIDIA founder and CEO Jensen Huang said, “Useful quantum computing will require fantastically complex, fault-tolerant systems — and understanding how they behave at scale is essential to building them,” and said the Jülich results show how quantum processors and GPU supercomputers can work together on research that would otherwise be out of reach. Kristel Michielsen, director of the Jülich Supercomputing Centre at Forschungszentrum Jülich, said Europe’s investment in JUPITER gives researchers the ability to investigate questions previously beyond reach, and that working with NVIDIA can help scientists make better-informed design decisions. But what does that prove about the eventual machine? NVIDIA says the simulation gives researchers data and tools to predict how future designs will perform before they are built, reducing reliance on hardware trial and error. The company says this could shorten the path to viable quantum computers by years, and points to drug discovery, financial modeling and materials science as potential application areas. Those are projected benefits of better design work, not demonstrated applications from this simulation. The evidence does support a narrower, useful conclusion: researchers can model a fault-tolerant process under realistic hardware noise at a scale NVIDIA describes as record-setting, and the company says the resulting information can help refine designs. JUPITER’s 8,192 GPUs show the classical computing resources involved. And the 90% figure refers to fewer qubits for the magic-state cultivation approach compared with previous methods; it doesn’t tell us the physical-qubit count of a complete future machine. The central open question is how well the simulated advantage carries over when researchers build and operate hardware. I think the strongest part of this announcement is the focus on realistic noise. A noiseless simulation can describe an idealized system; adding modeled hardware noise makes the exercise more relevant to design choices. That improves the value of the simulation as a research tool, though it doesn’t convert a model into experimental evidence from a quantum processor. I’m less persuaded by the broader claim that this alone shortens the route to useful applications by years. Simulation can help researchers avoid some hardware trial and error, as NVIDIA says, but the announcement doesn’t establish how much time this particular work saves on the path to a functioning fault-tolerant computer. That’s where I’d keep the claim carefully bounded. There’s also a useful division of labor here: GPUs can help researchers study the noise and resource needs of quantum designs, while the quantum hardware still has to show that the designs work in practice. I read this as a substantive engineering tool with a genuinely interesting scale, and a forecast of future utility that remains a forecast. The simulation’s scale is the result; the hardware translation is the test. Time for the Hype Check. 7 out of 10 for substance: the strongest reason to score it lower is that the claimed resource saving and performance insight come from simulation, but the 104-qubit modeled system, realistic-noise focus and 8,192-GPU run make it a substantial research result. NVIDIA says the project demonstrates how simulation will grow increasingly important and complex as quantum computing systems grow in size. Useful quantum computing will require fantastically complex, fault-tolerant systems, as Huang put it, and today’s announcement leaves the hardware proof ahead. If a future JUPITER simulation moves beyond 104 qubits, that would mark a larger modeled system; the announcement gives no date for that next level. If this episode helped you sort a simulation result from a hardware milestone, follow Quickly Quantum wherever you listen. 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! Watching on YouTube? Hit subscribe, and tomorrow's briefing will be waiting for you. I also host Space Stakes: the business of the new space race, every day. What actually flew, what the contract is really worth, and who has customers. Find it wherever you get your podcasts.