NVIDIA has announced that over 80 new scientific computing systems powered by its accelerated computing platform have been launched worldwide in the past year, bringing the global total AI performance to 4,500 exaflops. These systems are driving innovation across quantum physics, digital biology, climate science, materials research, and more. At the SC25 conference in St. Louis, Missouri, NVIDIA highlighted the debut of Horizon, America’s largest academic supercomputer, located at the Texas Advanced Computing Center (TACC). Set to launch in 2026, Horizon will be powered by NVIDIA GB200 NVL4 and NVIDIA Vera CPU servers, connected via NVIDIA Quantum-X800 InfiniBand networking. With 4,000 NVIDIA Blackwell GPUs, it will deliver up to 80 exaflops of AI compute at FP4 precision. Designed for large-scale scientific modeling, Horizon will support research in viral dynamics, galactic data analysis, and long-term seismic simulations. The U.S. Department of Energy (DOE) has partnered with NVIDIA to build seven new AI supercomputers at Argonne National Laboratory and Los Alamos National Laboratory. At Argonne, the largest system, Solstice, will feature 100,000 NVIDIA Blackwell GPUs and deliver up to 1,000 exaflops of AI training compute—over 50% more than the combined AI training capacity of the entire TOP500 list as of June 2025. Other systems at Argonne, including Equinox, Minerva, Janus, and Tara, will support AI inference and workforce development. At Los Alamos, the Mission and Vision systems—built by HPE—will use the NVIDIA Vera Rubin platform and Quantum-X800 InfiniBand networking. Mission will handle classified work for the National Nuclear Security Administration, while Vision will advance open science, including foundation models and agentic AI. Both are expected to be operational by 2027. This follows the earlier announcement of Doudna, a supercomputer for scientific discovery at Lawrence Berkeley National Laboratory, set to launch in 2026. Powered by the NVIDIA Vera Rubin architecture and Quantum-X800 InfiniBand, Doudna will support over 11,000 researchers in fusion energy, drug discovery, and astronomy. In Europe, the Jülich Supercomputing Centre’s JUPITER system has achieved exaflop performance on the HPL benchmark, making it Europe’s first exascale computer. Featuring 24,000 NVIDIA GH200 Grace Hopper Superchips and NVIDIA Quantum-2 InfiniBand, JUPITER enables high-resolution climate modeling at kilometer-scale resolution. Other notable European systems include Blue Lion in Germany, Gefion in Denmark, and Isambard-AI in the U.K., which is used for health data research and multilingual AI development. In Asia, Japan’s RIKEN is deploying NVIDIA GB200 NVL4 systems for AI and quantum computing, and is co-developing FugakuNEXT with Fujitsu and NVIDIA, using NVLink Fusion to integrate FUJITSU-MONAKA-X CPUs. Tokyo University of Technology has built an AI supercomputer with NVIDIA DGX B200 systems capable of 2 exaflops of FP4 performance using under 100 GPUs. AIST’s ABCI-Q, the world’s largest research supercomputer dedicated to quantum computing, features over 2,000 NVIDIA H100 GPUs. South Korea plans to deploy over 50,000 NVIDIA GPUs across sovereign cloud and AI factory infrastructure, with Samsung, SK Group, and Hyundai Motor Group investing in Blackwell-powered AI factories. In Taiwan, Foxconn is building an AI factory supercomputer with 10,000 NVIDIA Blackwell GPUs to support innovation across startups and industries. NVIDIA’s full-stack platform—including GPUs, CPUs, DPUs, networking, and software like CUDA-X and AI Enterprise—provides the unified architecture needed to scale scientific discovery sustainably and at unprecedented speed.