The shift is happening on classical hardware. Nvidia and researchers at Germany’s Jülich Supercomputing Centre have fully simulated a 50-qubit universal quantum computer, and the partners call it a first. The run took place on JUPITER, Europe’s first exascale supercomputer. The previous high-water mark was 48 qubits, set in 2019 by Jülich researchers on Japan’s K computer.
Why every qubit hurts
Each added qubit doubles the information a classical machine must track. For this run that meant a state vector of more than 2 quadrillion complex amplitudes. The 2019 record took about six years to fall, and the next one will need either far more hardware or smarter compression.
JUPITER is built on Nvidia GH200 Grace Hopper Superchips, which pair a CPU and a GPU with hybrid memory, so processors can share a larger pool than a GPU could reach alone. Nvidia’s Quantum-X800 InfiniBand networking links the nodes. The GH200’s memory design was a key ingredient, and a concrete showcase of how memory architecture sets the limit on quantum simulation.
Software did the heavy lifting
The Jülich Universal Quantum Computer Simulator, now called JUQCS-50, got three upgrades. Better CPU offloading pushes work onto CPU memory when GPU memory runs short. Memory requirements were compressed eightfold. Network traffic between nodes is managed more cleverly. Together they delivered a speedup of about 16.6 times over the earlier 48-qubit simulations. The effort drew on the JUPITER Research and Early Access Program and sits inside Jülich’s JUNIQ quantum user facility.
A classical simulation acts like a perfect quantum computer with zero interference. Real quantum hardware is very sensitive to its surroundings, so a noise-free sandbox is handy for benchmarking algorithms and testing error correction. For Europe, a record within about two months of JUPITER’s September 2025 inauguration is an early proof point. For Nvidia, it reinforces its place in quantum workflows without building a quantum computer.
Next: 104 qubits under noise
Nvidia has since gone further. It simulated a 104-qubit fault-tolerant quantum processor on JUPITER, using 8,192 GPUs, to study magic-state cultivation under realistic hardware noise. Because qubits operate at the limits of physics, controlling them introduces error-inducing noise. Fault-tolerant computing gets around that with magic states, a specific arrangement of qubits that is used up continually. The approach Nvidia validated needs 90 percent fewer qubits than previous methods, according to the company.
An Nvidia GB200 NVL72 system at its Accelerated Quantum Research Center served as the development platform, letting researchers scale tests from four GPUs to hundreds before moving to JUPITER. That cut a weeks-long simulation effort to days.
“Useful quantum computing will require fantastically complex, fault-tolerant systems,” said Nvidia CEO Jensen Huang. Kristel Michielsen, director of the Jülich Supercomputing Centre, said JUPITER lets researchers investigate questions that were previously out of reach. Watch this space: the data from runs like this is meant to predict how future designs perform before anyone builds them, with drug discovery, financial modeling and materials science as the stated targets.
Sources
- Nvidia and Jülich simulate a 50-qubit quantum computer on classical hardware
- Forschungszentrum Jülich
- NVIDIA Runs Simulation of Fault-Tolerant Quantum Computing at Jülich Supercomputing Centre
Photo: Pimvantend (CC BY-SA 4.0)




