A CPU-only machine built on domestic chips just became the world's fastest supercomputer, and the premise behind Washington's chip export controls looks weaker because of it.
The 67th TOP500 list, announced at the ISC 2026 conference in Hamburg on June 23, placed LineShine at number one with a High Performance Linpack score of 2,198.40 petaflops per second. That is more than 20 percent ahead of El Capitan, the Lawrence Livermore National Laboratory system that previously held the top spot. LineShine is installed at the National Supercomputing Centre in Shenzhen and was built by the Shenzhen Cloud Computing Center. It had never appeared on a TOP500 list before. It debuted directly at the top.
LineShine runs on a custom platform called LingKun, built around 304-core LX2 processors clocked at 1.55 GHz. The system connects more than 13.7 million of those cores through a proprietary interconnect called LingQi and runs Kylin OS, a Chinese-developed operating system. It draws 42,220 kilowatts of power.
Here is the part that matters most for the geopolitics of chips: LineShine has no GPU accelerators at all. Every other system in the top five relies heavily on graphics processors to boost performance. El Capitan pairs AMD EPYC CPUs with AMD Instinct MI300A accelerators. Frontier uses AMD MI250X GPUs. Aurora at Argonne National Laboratory runs on Intel Xeon CPUs paired with Intel Data Center GPU Max chips. Even JUPITER, the new European entrant at Jülich that became the continent's first exascale system, runs on NVIDIA Grace Hopper Superchips. LineShine did it with CPUs alone.
The gap between LineShine and the American systems it displaced is not subtle. El Capitan scored 1,809 petaflops, about 400 behind. Frontier managed 1,353. Aurora came in at 1,012. LineShine also took the number one spot on the HPCG benchmark at 22 petaflops, which measures real-world application performance rather than raw floating-point speed. On the HPL-MxP mixed-precision benchmark, it placed fourth at 7.92 exaflops, a result the TOP500 organizers described as consistent with a CPU-only architecture.
The June 2026 list marks the first time all five top systems have crossed the exascale threshold, which means each can perform more than a quintillion calculations per second. The global exascale club now has five members, and for the first time since Sunway TaihuLight led the rankings in 2017, one of them is Chinese.
Washington's export controls on advanced chips were designed primarily to cut off China's access to the hardware that powers AI training, the high-end GPUs from NVIDIA and AMD that fuel large language models and other compute-hungry AI workloads. Those restrictions have had real effects. Chinese firms have scrambled to stockpile older chips, develop domestic alternatives like Huawei's Ascend series, and find workarounds through intermediaries in third countries. The controls have unquestionably complicated China's AI ambitions.
But LineShine is not an AI training system. It is a traditional supercomputer built for scientific computing: climate modeling, molecular dynamics, materials simulation, the kind of work that governments have used high-performance machines for since the 1990s. That distinction matters. The export controls targeted a specific category of chip, the advanced GPU accelerator, because that is what AI workloads demand. They did not and could not prevent China from designing its own general-purpose processors and networking them into a world-class scientific computing platform. The LingKun LX2 is not competing with NVIDIA's H100 for AI training workloads. It is solving physics equations. And on that front, it just beat every machine the United States has built.
There is a broader lesson here about the limits of trying to constrain a technology ecosystem through hardware restrictions alone. China has invested heavily in domestic chip design for years, partly in response to earlier rounds of sanctions targeting companies like Huawei and SMIC. LineShine is the most visible result so far: a system that demonstrates world-class capability built entirely on hardware that falls outside the scope of US export controls, either because the processors are domestically designed or because they use architectures that were never restricted in the first place.
For the broader US-China technology competition, the implications are uneven. Washington may have succeeded in slowing China's access to cutting-edge AI training hardware. But it has not prevented China from building the most powerful supercomputer on the planet. Those are two different battles, and winning one does not mean winning the other.
Also read: A single bad block just froze Base twice in 48 hours and the sequencer problem is not going away • Ford spent billions learning that AI cannot replace engineers who know where the bodies are buried • Cloudflare cut 20% of its workforce while growing its engineering team, and Matthew Prince says every company will do the same
