Europe's First Exascale Supercomputer JUPITER Goes Live With 24,000 NVIDIA Chips
by
Aaron Leong
—
Monday, September 08, 2025, 12:02 PM EDT
Europe has officially entered the elite global supercomputing race with the inauguration of JUPITER, its first exascale supercomputer. Housed at the Jülich Supercomputing Centre, JUPITER (Joint Undertaking Pioneer for Innovative and Transformative Exascale Research) is a collaborative investment by the European Union and Germany which hasn't come cheap—think $587 million (€500 million). This freshly-minted tool is hoped to close the technology gap with the United States and China, particularly in the fields of AI, climate science, and fundamental research.
Now, an exascale-class machine is capable of performing at least one quintillion (a billion billion, or 1018) calculations per second. To put that in perspective, if every person on Earth were to perform one calculation per second, it would take more than four years to match what JUPITER can do in a single second.
Ranked as Europe's most powerful supercomputer and fourth in the world, JUPITER's raw power will enable researchers to tackle massive calculations. For one, it will be able to run climate and weather models at an unprecedented kilometer-scale resolution, which can enable far more precise forecasts of extreme weather events like heatwaves, heavy storms, and floods. JUPITER is also expected to greatly help in scientific discovery in fields such as neuroscience, where it can simulate human brain processes to advance research on diseases like Alzheimer's, and in renewable energy, by optimizing the design and efficiency of technologies like wind turbines and hydrogen systems.
Without a doubt, one of JUPITER's key roles is as an AI Gigafactory. It will be accessible to startups and researchers across Europe, providing the massive computing power needed to train and deploy generative AI models and large language models (LLMs).
NVIDIA's GH200 Grace Hopper Superchip
In terms of specs, JUPITER's primary engine, the Booster module, is equipped with
approximately 24,000 NVIDIA GH200 Grace Hopper Superchips, optimized for
highly parallel workloads. NVIDIA rates performance at over one exaFLOP/s in double precision for scientific simulations and more than 40 exaFLOP/s for AI applications.
Power draw is 11 megawatts, which is the equivalent of a small factory, but thanks to water cooling and reuse of the waste heat it generates to
warm buildings on the Jülich campus, the supercomputer has also earned the title of the most energy-efficient of the world's top five fastest systems on the Green500 list.