New exascale supercomputer, Jupiter, is now live and is the fastest in Europe and the fourth in the world. Funded by the European Commission and Germany, it’s designed for researchers in Europe working on domains like AI, weather and climate modelling, and biomedical research. 30 selected projects are already in line to use Jupiter. (https://www.nature.com/articles/d41586-025-02981-1)
Meanwhile, Switzerland has just launched its own open-source pure-text based LLM, Apertus, developed by EPFL, ETH Zurich, and CSCS as an alternative to models like ChatGPT and DeepSeek. Presented as ‘public infrastructure’ similar to public utilities and highways, the model is designed to be completely open and was trained on public data. (Also shared in the newest SHADE newsletter: https://www.engadget.com/ai/switzerland-launches-its-own-open-source-ai-model-133051578.html)
One thing these two projects have in common? Advertising that they run on 100% renewable energy - Jupiter paying to only use renewables from Germany’s grid, and Apertus on Switzerland’s, making both models, on paper, carbon-neutral. But is that really enough to erase potential negative environmental impacts? Given the scale of the supercomputer for example, the Nature article cites that at full capacity, Jupiter’s power demand would compete with about 11,000 homes. The same argument on opportunity cost is raised for Apertus in this LinkedIn post - the full picture of their environmental impacts is more complicated, and without judging their scientific merit, it would have been good to see sustainability be more seriously (and transparently) considered: https://www.linkedin.com/pulse/apertus-isnt-yet-win-you-think-maxime-grenu-ncswe/