Overview
- Japan’s RIKEN will deploy two GB200 NVL4 supercomputers — a 1,600‑GPU AI-for‑science system and a 540‑GPU quantum-focused system — using Quantum‑X800 InfiniBand, with operations targeted for spring 2026.
- TACC’s Horizon is slated for 2026 with 4,000 Blackwell GPUs and Vera CPUs, delivering 300 petaflops FP64 and up to 80 AI exaflops at FP4 for research spanning viruses, astrophysics and seismic modeling.
- The U.S. DOE is partnering with NVIDIA on seven AI supercomputers at Argonne and Los Alamos, including Solstice with 100,000 Blackwell GPUs that could reach about 1,000 AI exaflops for training, with some systems expected online in 2027.
- Europe’s JUPITER at Jülich has achieved exaflop performance on the FP64 HPL benchmark and is already supporting high‑resolution global climate simulations.
- Deployments standardize on NVIDIA’s integrated stack — CUDA‑X software, NVLink Fusion and Quantum‑X800 networking — to enable mixed AI, traditional HPC and quantum‑classical workflows.