NVIDIA Drives Global Surge in AI Supercomputing at SC25

Details

• NVIDIA announced at SC25 that 80 new scientific computing systems, collectively providing 4,500 exaflops of AI performance, have launched worldwide in the past year.
• The Texas Advanced Computing Center’s Horizon system, launching in 2026, features 4,000 Blackwell GPUs and delivers 300 petaflops to support research in astrophysics, materials science, disease modeling, and seismic analysis.
• The U.S. Department of Energy has invested in seven new AI supercomputers, including Solstice at Argonne National Laboratory, which will be powered by 100,000 Blackwell GPUs and is projected to reach 1,000 exaflops of AI training capability.
• Europe’s efforts include Germany's JUPITER system surpassing one exaflop, as well as major deployments like Blue Lion, Gefion, and Isambard-AI; Asian countries such as Japan, South Korea, and Taiwan are also ramping up AI supercomputing investments.
• Accelerated computing, now dominated by NVIDIA GPUs and the CUDA-X ecosystem, has overtaken CPU-based architectures, marking a new era where GPU-powered systems are the standard for high-performance scientific research.

Impact

NVIDIA’s sweeping expansion signals the end of CPU dominance: GPU-accelerated systems now power 80% of new deployments, dropping CPU-only market share to below 15% in six years. This new scale enables breakthroughs in fields from climate modeling to quantum research and underscores intensifying global competition among supercomputing powers through 2027.