China has regained the first world position in supercomputing that it lost in 2018 with LineShinea system installed in the Shenzhen National Supercomputing Center that has displaced The Captainthe American supercomputer located in LivermoreCalifornia, from the top of the ranking TOP500. The jump is relevant not only for the power achieved, but because of the way China has achieved it. LineShine does not base its power on GPU o Graphics processing units, the component that has become the centerpiece of modern high-performance computing and also much of the current race for artificial intelligence.
According to the performance data published in the TOP500 ranking, LineShine achieves 2,198 exaflopswhich is equivalent to more than two trillion operations per second. That figure places it more than 20% above El Capitanwhich held first place since 2024. The advance, however, has a high energy cost. The Chinese system consumes about 42.2 megawattsmuch more than the 29.7 megawatts of the American supercomputer, so its superiority in raw power is not accompanied by greater efficiency.
45,000 CPUs with 304 cores each
LineShine does not use GPU as calculation acceleratorsunlike many current supercomputers. Its design is based on a huge number of CPUs connected by its own high-speed, low-latency network called LingQi. Specifically, it uses about 45,000 LX2 processorseach with 304 cores and a frequency of 1.55GHz. The system works on Kylin OSan operating system based on Linux widely used in Chinese scientific and government environments, and is supported by the platform LingKun.
On paper, this represents a less conventional design to compete at the top of the TOP500, where graphics accelerators have prevailed due to their ability to execute large volumes of operations in parallel with very high efficiency.
China has resorted to this route out of technological necessity and industrial policy. US trade restrictions have limited Chinese companies and institutions’ access to advanced chips for yearsaccelerators for artificial intelligence, high-performance computing platforms, specialized software and certain components necessary to build cutting-edge systems. The measures began during the first term of donald trumpcontinued during the presidency of Joe Biden and have hardened again with the current Trump Administration through tariffs and additional controls on GPUs and advanced chips.
The result is that China has not been able to depend on Nvidia and other American vendors to build their large computing systems. Most powerful GPUs are subject to export controls because they can be used to train advanced artificial intelligence models and to build supercomputing systems with possible military, surveillance and strategic technological development uses. Without full access to them, Beijing has had to look for an alternative route based on more general processors, manufactured and assembled within its own technological ecosystem.
Just because LineShine does not use GPUs does not mean that they are no longer important. On the contrary, its absence explains the extent to which US restrictions have conditioned the design of the system. GPUs are still better suited for many AI and parallel computing tasks, but China has shown that it can still compete in raw power if it compensates for that lack with an architecture of tens of thousands of CPUs, a very fast internal network and its own software ecosystem..