Nvidia and researchers at Germany’s Jülich Supercomputing Centre have fully simulated a 50-qubit universal quantum computer on classical hardware. The partners describe it as the first achievement of its kind.
The run took place on JUPITER, Europe’s first exascale supercomputer. The previous high-water mark was a 48-qubit simulation, set in 2019 by Jülich researchers using Japan’s K computer.
Two qubits, one very large leap
The milestone was announced in a press release dated November 11, 2025. Every qubit added to a simulation doubles the amount of information the classical machine has to track, so small gains in qubit count demand outsized jumps in memory and compute.
For this run, that meant handling a state vector of more than 2 quadrillion complex amplitudes. JUPITER is built on Nvidia GH200 Grace Hopper Superchips, which pair a CPU and a GPU with hybrid memory access, allowing processors to share a larger pool of memory than a GPU could reach on its own.
The nodes are tied together with Nvidia’s Quantum-X800 InfiniBand networking.
The software did a lot of the heavy lifting
The Jülich Universal Quantum Computer Simulator, now upgraded to a version called JUQCS-50, received several key software improvements.
The first was enhanced CPU offloading, which lets the system push parts of the workload onto CPU memory when GPU memory runs short. The second was an eightfold compression of memory requirements. The third upgrade was smarter management of network traffic between nodes. Taken together, the changes delivered a speedup of approximately 16.6 times compared with the previous 48-qubit simulations.
The project was supported by the JUPITER Research and Early Access Program (JUREAP) and is integrated within JSC’s JUNIQ quantum user facility.
Why simulate a quantum computer at all
Physical quantum hardware is extremely sensitive to its environment, and errors creep in constantly. A classical simulation behaves like a perfect quantum computer with zero interference, making it useful for benchmarking quantum algorithms and testing error correction and fault-tolerance methods.
What this means for the quantum race
For Nvidia, the result reinforces its positioning as central to quantum computing workflows without building a quantum computer itself. The GH200’s hybrid memory design was a key ingredient, offering a concrete showcase for how memory architecture determines how far a quantum simulation can scale.
For Europe, the run is an early proof point for JUPITER, which was inaugurated in September 2025. A record-setting scientific result within roughly two months of launch demonstrates the system’s capability as part of the exascale computing tier.
Because memory demands double with each added qubit, brute-force simulation cannot keep pace indefinitely. The 2019 record took approximately six years to fall, and each new record requires either dramatically more hardware or cleverer compression.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.