China’s LineShine supercomputer project targets 2 exaFLOPS using only domestic chips

China has announced a major new supercomputer project called LineShine, and the goal is clear: build an exascale system without depending on foreign chips. The system was unveiled at a conference at the National Supercomputing Center in Shenzhen and is expected to target more than 2 exaFLOPS of sustained performance.

That number would put LineShine in the same broad performance class as the fastest systems in the world. For comparison, El Capitan, which sits at the top of the TOP500 list, reaches 1.809 exaFLOPS on the HPL benchmark and uses AMD EPYC processors with AMD Instinct MI300A accelerators.

LineShine is not just about raw speed, but about proving China can build a full exascale system with its own hardware stack

The most important claim around LineShine is not only its planned performance. It is the claim that the machine will use domestically produced technology instead of relying on international vendors. That matters because high-performance computing has become deeply tied to export controls, chip supply, AI development, and national technology independence.

According to reports, LineShine will be built in phases. The pilot system will use 100 Huawei Kunpeng servers with 12,800 cores. The larger industrial complex is expected to include 1,580 blade servers, 92 compute cabinets, 36 network cabinets, and a total of around 47,000 CPUs.

The system is also expected to include a major storage and cooling setup. Reports mention 650PB of storage capacity, 10TB/s of storage bandwidth, 67 liquid-cooled storage cabinets, 428 storage nodes, and more than 3,200 meters of secondary cooling pipes.

The choice to use a CPU-only design is also notable. Many of today’s top supercomputers rely heavily on GPU or accelerator hardware to reach extreme performance. LineShine is being described as a system built around efficient CPUs, high-bandwidth memory, high-speed interconnects, and a domestic software toolchain that includes compilers, debuggers, and tuning tools.

The project is expected to support both AI and traditional high-performance computing work. Reported use cases include large AI models, remote sensing, materials science, bioinformatics, weather research, pharmaceuticals, oil exploration, life sciences, and electromagnetic simulation.

There is still one big unknown: timing. Reports do not give a firm operational date, though Wccftech suggests LineShine could arrive around 2029 or 2030 based on China’s growing computing needs.

That uncertainty matters because planned supercomputers can change before they are completed. Performance targets, hardware choices, and launch timelines can all shift during development. For now, LineShine should be viewed as an ambitious announced project, not a finished machine.

Still, the message is important. China is not only trying to build a faster supercomputer. It is trying to show that it can build the hardware, software, storage, cooling, and system tools needed for exascale computing on its own. If LineShine reaches its target, it would be a major statement in the global race for AI and high-performance computing power.