At ISC, JUPITER Shows What Exascale Science Looks Like
15:00 · June 22, 2026 · NVIDIA
JUPITER, Europe’s first exascale supercomputer at Germany’s Forschungszentrum Jülich, runs on NVIDIA Grace Hopper Superchips and NVIDIA Quantum-X800 InfiniBand networking — and it’s had a busy year. As the international supercomputing community gathers at ISC in Hamburg this week, four projects running on JUPITER point to what exascale computing can actually do: map the human […]
Summary
JUPITER, Europe’s first exascale supercomputer installed at Forschungszentrum Jülich, is built on NVIDIA Grace Hopper Superchips and Quantum-X800 InfiniBand networking. At this week’s ISC conference in Hamburg, four production-scale projects illustrate how the system moves exascale resources from capability demonstrations into routine scientific use.
The Jülich Brain Atlas team has trained CytoNet, a foundation model that maps brain microarchitecture at cellular resolution. Using 6.5 petabytes of imaging data from 21 post-mortem brains, the model was trained in under five days on 4,096 Grace Hopper Superchips. It links individual neuron structures to larger organizational patterns and serves as the basis for an upcoming AI agent that can reason over multimodal brain data with language interfaces.
A new configuration of the ICON climate model, developed by a consortium including ETH Zurich, DKRZ, the Max Planck Institute for Meteorology and NVIDIA, won the Gordon Bell Prize for Climate Modelling. Running on 20,480 Grace Hopper Superchips, it performs the first fully coupled simulation of atmosphere, ocean, land, biogeochemistry and the carbon cycle at 1 km global resolution, advancing 146 days of climate in a single day of compute and resolving fine-scale processes such as ocean eddies and phytoplankton dynamics directly from physics.
Ericsson and Jülich have begun joint work on AI models for 5G evolution and 6G networks. JUPITER supplies the training and testing infrastructure for energy-efficient, brain-inspired architectures that target radio-edge inference and modular supercomputing concepts, aiming to reduce the power cost of complex network control functions.
Finally, researchers at the Jülich Supercomputing Centre, working with the NVIDIA Application Lab, achieved the first full simulation of a universal 50-qubit quantum computer. The coherent CPU-GPU memory architecture of the Grace Hopper nodes allowed quantum-state data to spill into CPU memory without significant slowdown, surpassing the prior 48-qubit limit and providing a testbed for quantum algorithms inside the JUNIQ facility.
Together the projects show that exascale hardware now supports end-to-end scientific workflows across neuroscience, climate science, telecommunications and quantum engineering.
Why it matters
While based in Germany, JUPITER represents a monumental leap in European AI and supercomputing infrastructure. This development is highly relevant to the Dutch AI market as it provides the broader EU ecosystem—including Dutch researchers and enterprises—with unprecedented computational power for advanced AI research in climate, neuroscience, and telecommunications.
