Overview
- NVIDIA said more than 80 new science systems were announced in the past year, totaling roughly 4,500 exaflops of reported AI performance across global research centers.
- Japan’s RIKEN will deploy two GB200 NVL4-based supercomputers in spring 2026 — a 1,600‑GPU AI-for-science system and a 540‑GPU quantum-focused system — with CUDA‑X adoption and floating‑point emulation work to accelerate legacy HPC codes.
- France announced Alice Recoque, Europe’s second exascale-class supercomputer, to be built by Eviden with AMD EPYC “Venice” CPUs, Instinct MI430X GPUs and BXI networking, a project backed by EuroHPC, GENCI and CEA with an estimated €554 million budget.
- The U.S. DOE outlined seven NVIDIA-accelerated AI systems for Argonne and Los Alamos, including Solstice at 100,000 Blackwell GPUs and additional systems slated through 2027, while Lawrence Berkeley’s Doudna targets a 2026 launch.
- Europe’s JUPITER at Jülich surpassed 1 exaflop on the FP64 HPL benchmark using 24,000 GH200 Grace Hopper Superchips and is already running high‑resolution global climate simulations.