At the ISC supercomputing conference in Hamburg, researchers at Germany’s Forschungszentrum Jülich demonstrated how JUPITER, Europe’s first exascale supercomputer, is accelerating frontier scientific discovery. Powered by NVIDIA Grace Hopper Superchips and Quantum-X800 InfiniBand networking, the system has transitioned from a hardware milestone to a critical production asset across multiple disciplines. Thomas Lippert, director of the Jülich Supercomputing Centre, emphasized that JUPITER positions Europe at the forefront of exascale computing by enabling complex simulations that were previously unattainable. In neuroscience, the CytoNet foundation model showcases exascale capabilities in mapping the human brain. Led by Katrin Amunts and Christian Schiffer at the Institute of Neuroscience and Medicine, the project trained on 6.5 petabytes of cellular-scale imaging data from 21 post-mortem brains. Executing across 4,096 Grace Hopper chips in under five days, the model correlates individual neuron structures with broader brain organization. Researchers are now integrating multimodal reasoning and NVIDIA Nemotron models to develop AI agents capable of autonomously designing and interrogating neuroscientific experiments. Climate modeling has simultaneously reached new resolution thresholds. The ICON Earth system model, developed by a consortium including ETH Zurich, the German Climate Computing Centre, and the Max Planck Institute for Meteorology, recently won the Gordon Bell Prize for Climate Modelling at SC25. Running on 20,480 Grace Hopper processors on JUPITER, ICON simulated 146 days of real-world climate conditions in just 24 hours of compute time. By modeling the coupled atmosphere, ocean, land, and full carbon cycle at a uniform 1-kilometer resolution, the system captures fine-scale phenomena like ocean eddies and phytoplankton blooms directly from physical laws rather than approximations, offering unprecedented insight into biosphere-atmosphere interactions. Beyond environmental and biological sciences, JUPITER is reshaping telecommunications infrastructure. Announced in March, a strategic partnership between Ericsson and Forschungszentrum Jülich leverages the supercomputer as a training engine for next-generation wireless networks. The initiative focuses on developing brain-inspired, energy-efficient AI architectures for 5G evolution and 6G deployment, targeting low-power inference at the network edge and modular supercomputing designs. In quantum computing research, JUPITER set a world record by fully simulating a universal 50-qubit quantum processor. Collaborating with the NVIDIA Application Lab, researchers utilized the coherent CPU-GPU memory architecture of the GH200 chips to manage vast quantum state data. By allowing data to spill seamlessly between processor and GPU memory with minimal latency, the system surpassed the previous 48-qubit simulation benchmark. The resulting JUQCS-50 simulator will be accessible through the JUNIQ user facility under Kristel Michielsen’s leadership, serving as a critical testbed for stress-testing algorithms ahead of commercial quantum hardware deployment. Collectively, these initiatives underscore exascale computing’s evolution from experimental benchmark to essential research infrastructure. By enabling high-resolution simulations, autonomous AI agents, next-gen network design, and advanced quantum algorithm testing, JUPITER validates the NVIDIA Grace Hopper platform as a foundational tool for tackling Europe’s most complex scientific challenges.