China has retaken the top spot on the TOP500 supercomputer ranking for the first time since 2017, with a system built entirely from domestic processors that achieves world-leading performance without a single GPU accelerator.
LineShine, installed at the National Supercomputing Centre in Shenzhen, recorded 2.198 exaflops on the High Performance Linpack (HPL) benchmark, roughly 80% of its 2.736 exaflops theoretical peak. It displaced El Capitan, the US system at Lawrence Livermore National Laboratory, which held the number one position at 1.809 exaflops.
The achievement is a direct riposte to US export controls. Since 2022, Washington has progressively tightened restrictions on selling advanced GPUs and semiconductor manufacturing equipment to China, aiming to stall Chinese progress in high-performance computing and artificial intelligence. LineShine demonstrates that those restrictions have not stopped China from building world-class HPC infrastructure, and that the country has done so by going a different architectural route.
An all-CPU architecture
LineShine runs on 13.79 million cores across LX2 processors, each packing 304 cores on two compute dies based on the ARMv9 architecture. Each processor integrates 32 gigabytes of on-package HBM memory and 256 gigabytes of DDR5 RAM. The CPUs run at 1.55 GHz and support FP64, FP32, FP16, and INT8 data types through SVE and SME vector units, delivering up to 60.3 teraflops per chip in double precision.
The system uses a homegrown interconnect called LingQi, arranged in a dual-plane multi-rail fat-tree topology providing 1.6 terabits per second of bandwidth per node. Total storage reaches 650 petabytes across 428 storage nodes, with 10 terabytes per second of storage bandwidth.
The entire machine draws approximately 42.2 megawatts of power, yielding an efficiency of 52.07 gigaflops per watt, competitive with GPU-accelerated systems at similar scale.
Architectural diversity at the top
The June 2026 TOP500 list is unusually diverse at the top. Five systems now exceed the exascale threshold: LineShine (2.198 exaflops, China), El Capitan (1.809 exaflops, US AMD MI300A-based), Frontier (1.353 exaflops, US), Aurora (1.012 exaflops, US), and JUPITER Booster (1.000 exaflops, Germany). This marks the first time exascale-class machines span Asia, North America, and Europe simultaneously.
Jack Dongarra of TOP500, speaking to the New York Times, called LineShine an “impressive system” that “developed a system that does not rely on GPUs and surpassed us.”
China has not submitted an entry to the TOP500 since 2019, when US sanctions first made it strategically risky to disclose national HPC capabilities. LineShine was unveiled in April 2026 and formally ranked at the ISC 2026 conference in Hamburg, Germany, on June 23.
Sources: China Defies US Restrictions and Builds the World’s Fastest Supercomputer (Wired, June 28, 2026); LineShine Debuts at No. 1 (TOP500, June 23, 2026); China’s LineShine regains world top spot (DigitalToday, June 24, 2026); China builds exascale supercomputer without GPUs (Jon Peddie Research, May 4, 2026)