The top 10 systems in the global supercomputer performance rankings have held their positions for six months, while new entries hit a record low. As high-performance GPUs concentrate in large AI data centres rather than traditional supercomputers, the centre of gravity of global computing infrastructure is shifting, an analysis said.
On Nov. 17 local time, Taiwan outlet IT Home reported that TOP500’s November list showed 27 new systems, or 5.4 percent of the total. It was the smallest number since rankings began in 1993.
China’s LineShine at the National Supercomputing Center in Shenzhen ranked first, posting 2.198 exaflops on the High Performance Linpack benchmark. It was followed by the United States’ El Capitan at 1.809 exaflops, Frontier at 1.353 exaflops and Aurora at 1.012 exaflops, and Germany’s Jupiter Booster at 1.000 exaflops.
While rankings stagnated, overall computing performance improved. Performance of the 500th system rose 7.3 percent from June to 2.857 petaflops, and the number of systems above 100 petaflops increased to 27 from 25. TOP500 said the increase reflected more expansion and performance upgrades of existing equipment than replacement with new systems.
In particular, the direction of GPU investment is changing. Epoch AI said the world’s AI computing capacity has surpassed the equivalent of 15 million Nvidia H100 GPUs. As of the fourth quarter of 2025, the AI computing capacity held by the top five U.S. hyperscalers was estimated at about 71 percent of the global total.
Traditional supercomputers focus on scientific computing such as climate and physics simulations, while AI factories are used for large-scale model training and inference, and for operating AI services. They prioritise token throughput, inference efficiency and operating costs over traditional computing performance, so the scale of actual GPU investment may not be fully reflected in TOP500 rankings.
The highest-ranked new entry in the latest list was Hangang from the Korea Institute of Science and Technology Information (KISTI), which placed 12th with performance of 342.6 petaflops.