NVIDIA and the Jülich Supercomputing Centre ran a 104-qubit simulation of fault-tolerant quantum computing on JUPITER, Europe’s first exascale system, using 8,192 GH200 superchips. The run validates magic state cultivation, a protocol requiring 90% fewer physical qubits, and positions GPU supercomputers as the quantum industry’s design platform.
What Is Covered in This Article:
- A 104-qubit magic state cultivation simulation on JUPITER, the largest comprehensive fault-tolerant quantum simulation to date
- 8,192 NVIDIA GH200 superchips tracking roughly 35.2 trillion complex amplitudes under realistic noise
- The 90% physical qubit reduction claim for cultivation versus traditional distillation factories
- Google’s hardware demonstration of the same protocol on Willow, published in Science days earlier
- Competitive positioning against Google, IBM, AMD, and Quantinuum in the race to fault tolerance
The News: NVIDIA announced on October 19, ahead of its GTC Berlin keynote, a record simulation of fault-tolerant quantum computing on JUPITER, Europe’s first exascale supercomputer at the Jülich Supercomputing Centre. The 104-qubit run is the largest comprehensive simulation of a fault-tolerant quantum processor to date, modeling magic state cultivation, an approach to fault tolerance that NVIDIA says requires 90% fewer physical qubits than previous methods. The simulation ran on a system of 8,192 NVIDIA GH200 Grace Hopper Superchips using NVIDIA cuQuantum software and the CUDA-Q platform’s fault-tolerant logical layer. An NVIDIA GB200 NVL72 system at the NVIDIA Accelerated Quantum Research Center served as the development platform, letting researchers scale tests from 4 GPUs to hundreds before moving to JUPITER.
“Useful quantum computing will require fantastically complex, fault-tolerant systems, and understanding how they behave at scale is essential to building them,” said Jensen Huang, Founder and CEO of NVIDIA. “Europe’s investment in JUPITER gives researchers the ability to investigate questions that were previously beyond reach,” said Kristel Michielsen, director of the Jülich Supercomputing Centre at the Forschungszentrum Jülich facility.
NVIDIA’s 104-Qubit JUPITER Simulation Makes GPU Supercomputers the Design Platform for Fault-Tolerant Quantum
Analyst Take: NVIDIA can drive demand for quantum computing infrastructure a decade before quantum computers work. This announcement shows the model at full scale. Every fault-tolerant architecture now in development must be validated in simulation before it is built in hardware, since a wrong bet on an error correction scheme costs years and billions. The 104-qubit JUPITER run demonstrates that only an exascale GPU system can perform that validation at the fidelity designers need, which makes the AI supercomputer the quantum industry’s equivalent of EDA tooling. Quantum vendors and sovereign programs will become GPU customers on the path to their own machines, a revenue stream that pays NVIDIA regardless of which qubit modality wins.
The timing sharpens the story. Google Quantum AI published a hardware demonstration of magic state cultivation on its 105-qubit Willow processor in Science on October 15, reporting a roughly 40x reduction in state preparation errors. Three days later, NVIDIA simulated the same protocol family at comparable qubit scale under realistic noise, including coherent phase errors and three-level leakage that Google identified as its dominant hardware problem. Simulation has caught up to the frontier of quantum hardware and that convergence is a commercial opening for NVIDIA.
Simulations Model Noise That Stabilizer Methods Cannot Touch
The technical substance separates this run from prior records. Jülich’s earlier 50-qubit JUPITER benchmark simulated a universal quantum computer under ideal, noiseless conditions, per the Jülich announcement. The new work doubles the qubit count while adding coherent phase noise and leakage into non-computational qutrit states, effects that standard stabilizer tableau methods cannot represent. The state vector encoded 45 bits, roughly 35.2 trillion complex amplitudes, distributed across the GH200 fleet, with five-qutrit groups packed into 8-bit registers to make leakage tractable. The simulated protocol grows a distance-3 color code into a distance-7 hybrid color and surface code with active flag error detection, the full circuit a real machine would execute. Each additional simulated qubit doubles the memory requirement, so 104 qubits sits near the ceiling of what even exascale state-vector simulation can reach. That exponential wall implies the caveat that this tool verifies protocol components at the ~100-qubit scale. Full fault-tolerant machines of thousands of logical qubits will remain beyond direct simulation permanently.
Cultivation Went From Paper to Hardware to Exascale Validation in Two Years
Magic state cultivation was proposed by Google’s Craig Gidney with Noah Shutty and Cody Jones in a 2024 paper as a replacement for distillation factories, the sprawling circuit blocks that some resource estimates projected would consume up to 95% of a fault-tolerant machine’s capacity. The 90% physical qubit reduction NVIDIA cites is the protocol’s promise yet it remains a design thesis rather than a demonstrated property of any full system. What the JUPITER run adds is error budgeting under realistic noise at a scale hardware cannot yet probe cleanly. Google’s own Willow demonstration showed storage errors running roughly 7x higher than simulation predicted, attributed mainly to leakage, which is precisely the effect the NVIDIA simulation models in detail. GPU simulation and quantum hardware now iterate against each other: hardware exposes the noise, simulation quantifies its impact on protocol design, and the loop shortens the path to machines that work. NVIDIA says this loop can shorten development timelines by years, a claim that asks a quantum vendor to publicly attribute a design decision to it.
The Quantum Design Loop Is Becoming a Contested GPU Market
NVIDIA’s position in quantum is infrastructure, and the competitive field is forming around that same insight. IBM mapped a $10 billion path to its fault-tolerant Starling system by 2029 and partnered with AMD to run error correction decoding on AMD hardware, a direct challenge to NVIDIA’s claim on the classical side of quantum, as Futurum covered in IBM and AMD Join Forces on Quantum-Centric Supercomputing. Google develops its protocols on internal infrastructure and publishes the theory NVIDIA now simulates. Quantinuum reported 99.9995% logical memory fidelity on its Helios system the same week, a reminder that trapped-ion vendors are advancing fault tolerance with far fewer physical qubits than superconducting approaches require. NVIDIA’s counter is breadth: NVQLink couples GPU systems to quantum processors from more than a dozen vendors, CUDA-Q spans simulation and hardware control, and the NVAQC gives it a research seat beside every major qubit modality. Futurum’s June 2026 analysis of the IonQ and IBM foundry investments found $4.6 billion in new capital flowing toward quantum industrialization, per Will a U.S. Quantum Foundry Leverage $4.6 Billion in New Capital to Become the Next TSMC?, and a growing share of that capital must pass through classical simulation and control infrastructure before it reaches a dilution refrigerator. Europe is the showcase. JUPITER is a EuroHPC machine, and this result hands European policymakers evidence that exascale AI infrastructure doubles as sovereign quantum research capacity, an argument that supports the next procurement cycle.
What to Watch:
- Whether a quantum hardware vendor publicly attributes an architecture decision to JUPITER-scale simulation within 2 quarters
- Whether Google’s next Willow-class result closes the 7x gap between measured and simulated storage errors
- Whether IBM and AMD demonstrate error correction decoding at comparable fidelity on non-NVIDIA silicon
- Whether EuroHPC’s next procurement cites quantum simulation as a design requirement
Read the complete announcement on the NVIDIA website.
Sources
- NVIDIA Runs Record Simulation of Fault-Tolerant Quantum Computing at Jülich Supercomputing Centre, NVIDIA
Declaration of generative AI and AI-assisted technologies in the writing process: This content has been generated with the support of artificial intelligence technologies. Due to the fast pace of content creation and the continuous evolution of data and information, The Futurum Group and its analysts strive to ensure the accuracy and factual integrity of the information presented. However, the opinions and interpretations expressed in this content reflect those of the individual author/analyst. The Futurum Group makes no guarantees regarding the completeness, accuracy, or reliability of any information contained herein. Readers are encouraged to verify facts independently and consult relevant sources for further clarification.
Disclosure: Futurum is a research and advisory firm that engages or has engaged in research, analysis, and advisory services with many technology companies, including those mentioned in this article. The author does not hold any equity positions with any company mentioned in this article.
Analysis and opinions expressed herein are specific to the analyst individually and data and other information that might have been provided for validation, not those of Futurum as a whole.
Read the full Futurum Group Disclosure.
Other Insights From Futurum:
NVIDIA DGX Spark 64GB’s Low Price-Point Aims To Accelerate Local AI Inference
Adobe and NVIDIA Speed Agentic Creation, but Trust Is What Enterprises Buy
Cisco’s USFP Brings 400G-per-Lane Design to NVIDIA’s Optical Roadmap
Author Information
Brendan is Research Director, Semiconductors, Supply Chain, and Emerging Tech. He advises clients on strategic initiatives and leads the Futurum Semiconductors Practice. He is an experienced tech industry analyst who has guided tech leaders in identifying market opportunities spanning edge processors, generative AI applications, and hyperscale data centers.
Before joining Futurum, Brendan consulted with global AI leaders and served as a Senior Analyst in Emerging Technology Research at PitchBook. At PitchBook, he developed market intelligence tools for AI, highlighted by one of the industry’s most comprehensive AI semiconductor market landscapes encompassing both public and private companies. He has advised Fortune 100 tech giants, growth-stage innovators, global investors, and leading market research firms. Before PitchBook, he led research teams in tech investment banking and market research.
Brendan is based in Seattle, Washington. He has a Bachelor of Arts Degree from Amherst College.
