Sweden inaugurated the Arrhenius supercomputer on 8 September 2026 at Linköping University, where it is hosted by NAISS, the National Academic Infrastructure for Supercomputing that runs the country's academic computing estate. HPE delivered the machine, and EuroHPC and the Swedish Research Council co-funded it.1
Its GPU (graphics processing unit) partition runs 382 nodes of four Nvidia Grace Hopper Superchips each, alongside 424 AMD Turin 128-core processors, and posts 66.8 PFLOPS (petaflops, floating-point operations per second) on the High Performance Linpack (HPL) benchmark, placing it 42nd on the June 2026 TOP500 list of the world's fastest computers.2 It entered production in early June 2026, so today's ceremony rather than the machine itself falls inside this window.
Arrhenius belongs to EuroHPC's mid-range tier and sits outside the AI Factory programme, which is why the three machines EuroHPC has put on the record this window do not form one category. Across all three the accelerators come from AMD or Nvidia. No European accelerator appears in the disclosed hardware of LUMI-AI, MeluXina-AI or Arrhenius, and the €4.12 billion AI Gigafactories call now tendering seven further sites has produced nothing yet that would change the pattern.
That reading is a cross-read of first-party disclosures rather than a policy any institution has stated, and the counter-case deserves stating with it. Europe holds real leverage one layer down, in the lithography tools without which none of these accelerators can be fabricated at all. Leverage at the tool layer and dependency at the accelerator layer are not the same industrial position, and only one of them shows up in a EuroHPC procurement.
