Jupiter has been officially powered on at the Jülich Supercomputing Center and debuts as the fastest machine in Europe, ranked number four on the Top500 list. The modular exascale system delivers a peak of nearly one ExaFLOP, measured as 930 PetaFLOPS of FP64 performance. The build centers on NVIDIA Grace Hopper technology, with the booster contributing approximately 793 PetaFLOPs of FP64 capacity.
Hardware counts emphasize scale and density. Each compute node carries four GH200 accelerators, yielding a total fleet of 23,752 accelerators across 5,938 nodes, mounted in 125 server racks. The GH200-class accelerators are highlighted for energy efficiency, with published figures near 60.5 GigaFLOPS per watt. The accelerator farm draws on the order of double-digit megawatts of power. Jupiter is primarily designed for Artificial Intelligence workloads and, in that context, delivers over 90 ExaFLOPS in the FP8 format preferred for many AI tasks.
Reported build costs are estimated near €500 million. European planners expect a multibillion-euro expansion will be required to close capacity gaps, citing figures around €60 billion for deployments through 2030. The plan envisions roughly 13 specialized Artificial Intelligence data centers and gigafactory-scale facilities, with Jülich listed as a candidate for expansion. Backers frame the investment as a matter of digital sovereignty to keep critical infrastructure close to European research and industry rather than depending on foreign cloud providers. Operationally, Jupiter will support climate modelling, materials simulation and biological research, and it is complemented by an inference module called Jarvis for fast model serving, positioning the machine as both a capability and a signal to catalyze further regional investment.
