Europe's newest supercomputer just pulled off a quantum feat that no machine on Earth had managed before.

Researchers at the Jülich Supercomputing Centre and NVIDIA announced this week that they have fully simulated a universal quantum computer with 50 qubits for the first time, breaking the previous world record of 48 qubits that the same Jülich team set on Japan's K computer back in 2019. The work was carried out on JUPITER, Europe's first exascale machine, which officially launched at Forschungszentrum Jülich last September.

The jump from 48 to 50 qubits sounds small. It isn't. Every added qubit doubles the memory and computing power required, because quantum systems grow exponentially with size. A standard laptop can handle around 30 qubits. Simulating 50 qubits requires roughly 2 petabytes of memory — about two million gigabytes — and the ability to keep more than 2 quadrillion complex numerical values perfectly in sync across thousands of computing nodes.

"Only the world's largest supercomputers currently offer that much," says Prof. Kristel Michielsen, Director at the Jülich Supercomputing Centre. "This use case illustrates how closely progress in high-performance computing and quantum research are intertwined today."

Why simulating quantum computers matters

Real quantum hardware is still in its early, noisy days. Today's leading quantum processors have more qubits than 50 on paper, but they make errors so often that running long, useful algorithms is still out of reach. Classical simulators step in as a kind of training ground: they let researchers test algorithms, validate experimental findings, and design future quantum systems before the hardware is ready to handle them.

Two algorithms in particular stand to benefit from the new simulator. The Variational Quantum Eigensolver, or VQE, helps researchers model molecules and materials — work that could speed up the search for new drugs, catalysts, and battery chemistries. The Quantum Approximate Optimisation Algorithm, or QAOA, tackles optimization problems that show up everywhere from logistics and finance to artificial intelligence.

Being able to faithfully simulate what a 50-qubit machine would actually do gives scientists a way to test whether these algorithms work as advertised, long before they need to compete for time on scarce, fragile quantum hardware.

How JUPITER did it

The record relied on more than 16,000 NVIDIA GH200 Superchips inside JUPITER. These chips fuse CPUs and GPUs into a single tightly linked unit, which means data that overflows GPU memory can spill into CPU memory without grinding the calculation to a halt.

Engineers at the NVIDIA Application Lab — a joint effort between Jülich and NVIDIA — rebuilt the laboratory's flagship quantum simulator, the Jülich Universal Quantum Computer Simulator (JUQCS). The new version, JUQCS-50, can keep crunching numbers even when data sloshes back and forth between processor types. The team also introduced a byte-encoding compression trick that slashes memory needs by a factor of eight, plus a dynamic optimization system that constantly tunes data exchange across the entire machine.

"With JUQCS-50, we can emulate universal quantum computers with high fidelity and tackle questions that no existing quantum processor can yet solve," says Prof. Hans De Raedt of the Jülich Supercomputing Centre, lead author of the study, released as a preprint.

Open to outside researchers

Crucially, JUQCS-50 will be made available to outside research groups and companies through JUNIQ, the Jülich Unified Infrastructure for Quantum Computing. That turns the record-setting simulator into a shared resource — a benchmark for evaluating future supercomputers and a working tool for any researcher who needs to know what 50 qubits would actually do.

For the wider quantum community, the result is a quiet but significant nudge forward. The frontier just moved by two qubits, and Europe is now the place where the next frontier will get tested first.