China's LineShine is the world's fastest supercomputer — and it doesn't use a single GPU
China has reclaimed the top spot in global supercomputing for the first time since 2017. The new leader is LineShine, installed at China's National Supercomputing Center in Shenzhen, which scored 2.198 exaflops — that's 2.198 quintillion calculations per second — on the standard Linpack benchmark. That puts it 22% ahead of the previous champion, El Capitan, and makes it the fastest publicly ranked machine on the planet.
The CPU-only design
What makes LineShine unusual is that it achieves this without graphics processors (GPUs), which have powered virtually every recent supercomputing record. Instead, it runs on China's custom LingKun platform, built around 304-core LX2 processors clocked at 1.55 GHz. To hit exascale performance with CPUs alone, the system uses a staggering 13.78 million cores. A proprietary LingQi interconnect links them all together, and the whole machine runs on Kylin OS, China's domestic operating system.
That CPU-only approach has a tradeoff. On the HPL-MxP benchmark — the mixed-precision test most relevant to AI workloads — LineShine drops to fourth place with 7.92 exaflops. El Capitan, which uses AMD MI300A hybrid chips, leads that ranking at 16.7 exaflops. In other words, LineShine is built for raw compute scale, not AI acceleration.
What this means for the US
El Capitan, housed at Lawrence Livermore National Laboratory in California and funded by the National Nuclear Security Administration, slips to second at 1.809 exaflops. It was built by HPE and AMD for nuclear stockpile stewardship — a specific national security workload, not general AI research.
The June 2026 TOP500 official list shows the full top five: LineShine (China, 2.198 EF), El Capitan (US, 1.809 EF), Frontier (US, 1.353 EF), Aurora (US, 1.012 EF), and JUPITER Booster (Germany, 1.000 EF).
LineShine's rise also illustrates the limits of US export controls on advanced chips. China built its way to the top without American GPU technology, relying entirely on domestic chip and architecture development. As a Nature article on LineShine notes, the CPU-scale strategy sidesteps the restrictions entirely.
The bigger picture
TOP500 rankings measure one specific benchmark. Amazon, Microsoft, and Google all run massive proprietary HPC clusters that never appear on the list. Leadership here doesn't map directly to dominance in AI infrastructure — but it does signal that China's domestic chip development has reached a level capable of beating the world's best at exascale compute. The US and Europe will likely respond, but on what timeline remains to be seen.
