China Tops Supercomputing Rankings with LineShine
China has once again claimed the title of the world’s most powerful supercomputer following the arrival of LineShine at the National Supercomputing Centre in Shenzhen. This impressive system, introduced in the 67th ranking of the TOP500, boasts a performance of 2.198 exaflop/s on the High Performance Linpack (HPL) benchmark.
LineShine is noteworthy as it is the first CPU-only supercomputer to exceed two exaflop/s in double-precision performance, pushing aside the previous leader, El Capitan. This return to the top marks China’s first entry into the TOP500 list since 2023, and the results were shared at the ISC 2026 conference in Hamburg, Germany.
The resurgence in supercomputing illustrates a broader competition between China and the United States in the field of advanced computing. While traditional supercomputing continues to be relevant for scientific research and national security, increasing focus is also being placed on AI-optimized computing architectures which are often not included in traditional rankings.
LineShine was developed by the Shenzhen Cloud Computing Center and features a unique Chinese processor architecture called LingKun, with around 13.79 million processing cores from 304-core LX2 processors running at 1.55 GHz. Utilizing the proprietary LingQi interconnect, LineShine operates on the Kylin OS and achieves approximately 80% of its theoretical peak performance of 2.736 exaflop/s, consuming around 42.2 megawatts of power.
A New Era for Exascale Supercomputers
With the advent of LineShine, the number of supercomputers reaching the exascale threshold has risen from four to five. For the first time, these exascale systems are now found across Asia, North America, and the European Union.
Currently, El Capitan holds the second position at 1.809 exaflop/s, followed by Frontier at 1.353 exaflop/s and Aurora at 1.012 exaflop/s. Germany’s JUPITER Booster completes the top five with exactly 1.000 exaflop/s.
The latest rankings also welcomed a new standout: Eni’s HPC7 system in Italy, which debuted in sixth place with 571.5 petaflop/s. Microsoft’s cloud-based Eagle system ranks seventh, followed by Eni’s HPC6, Japan’s Fugaku, and Switzerland’s Alps.
In terms of architecture, the latest list highlights a diverse range of technologies among the leading supercomputers, now including custom Chinese processors and various systems powered by AMD, Intel, and NVIDIA. Despite this diversity, HPE remains the leading system integrator, supplying six of the top ten systems.
The Divide Between AI and Supercomputing
Experts warn that just because a supercomputer ranks higher on the TOP500 list doesn’t mean it excels in AI computing. Many AI infrastructures are built by large cloud companies that often do not participate in traditional supercomputing benchmarks.
Tech giants like Microsoft, Amazon, and Google are heavily investing in specialized AI computing clusters focused on large-scale model training instead of the scientific simulations counted by HPL benchmarks.
Jimmy Goodrich from the University of California highlights that many of these modern AI systems would likely outperform traditional supercomputers if they were assessed. This difference is evident when comparing rankings like HPL-MxP, which focuses on mixed-precision performance relevant to AI workloads.
In this context, while LineShine leads the HPL rankings, it falls to fourth in HPL-MxP with 7.92 exaflop/s, behind El Capitan, Aurora, and Frontier.
Geopolitical Implications
The resurgence of China in the supercomputing landscape comes at a time when the country had paused its submissions due to U.S. export controls impacting advanced semiconductor technologies. The return of Chinese systems to the top may signify a demonstration of their advancements in domestic semiconductor capabilities despite ongoing restrictions.
In related news, energy efficiency appears stable in the recent Green500 ranking, with France’s KAIROS maintaining first place with 73.28 gigaflops per watt, followed by systems like ROMEO-2025 and Germany’s Levante GPU extension, both utilizing NVIDIA technology.
