China's return to the TOP500 summit after nine years matters for geopolitics and HPC procurement, but carries a critical caveat for AI practitioners: LineShine is entirely CPU-based with no GPU accelerators. At ISC'26 in Hamburg, the 67th TOP500 list debuted with LineShine, installed at the National Supercomputing Centre in Shenzhen, at #1 with 2.198 Exaflops sustained (Rmax) -- more than 20% ahead of former leader El Capitan -- using roughly 13.8 million cores of the indigenous LX2 processor. It also tops the HPCG benchmark at 22.004 Petaflops, confirming it is not a LINPACK-only system. Meanwhile NVIDIA continues to dominate AI-oriented HPC: 81% of TOP500 systems and all of the Green500's top 8 run on NVIDIA hardware. The benchmark that matters for AI training throughput tells a different story than the one China just won.
China claims the world’s fastest supercomputer