At the 2026 World Artificial Intelligence Conference, the Shengteng 950 supernode was unveiled for the first time. The product is based on the Lingqu interconnection protocol and supernode architecture, achieves the largest 1024 card scale in the industry, provides 1 EFLOPS FP8 and 2 EFLOPS FP4 computing power, has 256 TB of global unified memory addressing space, and relies on the TB-level NPU interconnect's ultra-high bandwidth and 3 μs ultra-low RTT latency. According to reports, Shengteng 384 supernodes have been commercially launched. They are used on a large scale in the Internet, operators, finance, education, medical care, transportation, manufacturing and other industries. It is also the only supernode in China that has trained the SOTA model.

At the 2026 World Artificial Intelligence Conference, the Shengteng 950 supernode was unveiled for the first time. The product is based on the Lingqu interconnection protocol and supernode architecture, achieves the largest 1024 card scale in the industry, provides 1 EFLOPS FP8 and 2 EFLOPS FP4 computing power, has 256 TB of global unified memory addressing space, and relies on the TB-level NPU interconnect's ultra-high bandwidth and 3 μs ultra-low RTT latency. According to reports, Shengteng 384 supernodes have been commercially launched. They are used on a large scale in the Internet, operators, finance, education, medical care, transportation, manufacturing and other industries. It is also the only supernode in China that has trained the SOTA model.