LineShine, a Chinese built system running on roughly 45,000 domestic LX2 CPUs linked by a custom network rather than GPUs, edged out El Capitan on the TOP500 benchmark to give China the top ranking for the first time since 2018.
China's LineShine supercomputer has taken the number-one spot on the TOP500, the twice-yearly ranking of the world's most powerful publicly disclosed machines, and the path there runs through US chip export controls rather than around them.
LineShine, operated by the National Supercomputing Center in Shenzhen, pushed the US-built El Capitan out of the top slot on the June 2026 list, the first time a Chinese system has held the ranking since the Sunway TaihuLight era ended in 2018. The system crossed 2,000 exaflops on HPL, a benchmark that measures how fast a machine can solve a dense system of linear equations. One exaflop equals a quintillion (a billion billion) floating-point operations per second, and crossing two thousand of them is the kind of milestone the supercomputing community treats as a meaningful marker of raw computational scale. LineShine landed roughly 20 percent above El Capitan, the Lawrence Livermore machine that had held the top spot since 2024.
The architectural surprise is what makes LineShine more than a scoreboard update. Where most modern supercomputers lean on GPUs, the specialized chips that have become the workhorses of AI training and many scientific workloads, LineShine uses none. Instead it ties together roughly 45,000 LX2 processors, each carrying 304 cores running at 1.55 GHz, over a custom Chinese-built high-speed interconnect called LingQi. The LX2 is a Chinese-domestic general-purpose CPU, not a GPU, and the design choice is conspicuous given that the Trump administration has spent the past two years restricting China's access to advanced accelerators, notably NVIDIA's high-end chips. The Verge and the trade press frame the result as a deliberate "build around the controls" strategy: scale up general-purpose CPU silicon and wire it together with a network fast enough to keep the cores fed with data.
The trade-off shows up in the power bill. LineShine draws roughly 42.2 megawatts, compared with about 29.7 megawatts for El Capitan, which means the Chinese machine delivers fewer computations per joule even as it tops the raw-speed list. That gap is the reason Nature's coverage and the trade press both caution that "fastest" on TOP500 is not the same as "best." The Green500, a separate ranking that measures performance per watt, tells a different story, and the workloads that actually drive modern scientific and AI research, such as HPCG, Graph500, and MLPerf, stress a machine in ways HPL does not.
China's win is real but narrow. The TOP500 is one benchmark, and on it the United States still holds three of the top five slots. AMD-powered systems, including El Capitan, account for four of the top ten per StorageReview's June 2026 read. LineShine's claim to global leadership is also a claim the Chinese government is happy to amplify: the state-affiliated Global Times framed the result as a direct message to Washington, language that should be read as Beijing's editorial line rather than as independent confirmation of intent.
The more durable question is whether the architectural bet pays off beyond HPL. NSCC Shenzhen has positioned LineShine as a domestic, exascale-class "global showcase" system, and the next test will be whether CPU-scale-plus-custom-interconnect holds up on the benchmarks that matter for AI training and large-scale simulation. If it does, US export controls will have produced something their architects did not intend: an alternative supercomputing stack that does not need the chips Washington was trying to keep out of Chinese hands.