JD Cloud Plans a 100,000-GPU Cluster Built Entirely on Domestic Moore Threads Silicon
TechNode·medium signal
TechNode reported September 9 that JD Cloud intends to build a 100,000-GPU intelligent computing cluster using Moore Threads chips rather than Nvidia parts. It is one of the largest announced all-domestic training clusters in China and a direct test of whether Chinese accelerators can hold together at that scale. The plan collides with the same week's news that domestic accelerator prices are up 20-50% on HBM scarcity.