In a decisive return to the pinnacle of high-performance computing, China’s 'Ling Sheng' supercomputer has claimed the top spot on the global TOP500 list at the ISC2026 conference in Hamburg. Clocking in at a staggering 2.19 exaflops, the system is the first in history to maintain sustained performance exceeding two quintillion calculations per second. This achievement marks the first time a Chinese machine has officially led the world since 2017, ending a nine-year hiatus from the top of the rankings.
More significant than the raw speed is the architectural path China has taken to reach it. While dominant American systems like El Capitan and Frontier rely on a heterogeneous mix of CPUs and high-end GPUs from AMD or Nvidia, Ling Sheng utilizes a unique 'all-CPU' architecture. By integrating AI matrix acceleration units directly into its ARMv9-based LX2 processors, the system bypasses the need for the high-end discrete GPUs that have been the primary target of U.S. export restrictions over the last several years.
The system's design reflects a total-stack domestic breakthrough, featuring over 14 million cores and the first Chinese-produced High Bandwidth Memory (HBM). Developed under the leadership of Professor Lu Yutong at the National Supercomputing Center in Shenzhen, Ling Sheng represents a strategic pivot toward 'super-intelligence convergence.' This approach aims to unify traditional scientific simulation with modern artificial intelligence workloads on a single, streamlined hardware platform.
Since the U.S. began tightening sanctions in 2019, China had largely stopped submitting Linpack benchmark results to international auditors, leading to a period of perceived stagnation in the West. Ling Sheng’s sudden emergence not only proves that China has been quietly advancing its exascale capabilities but also suggests that it has successfully developed a viable technological 'Plan B' that operates independently of the Western silicon supply chain. With an energy efficiency of 51 GFlops/W, the machine also sets a new benchmark for green computing at the exascale level.