SemiAnalysis Discloses Inference Run on Over 100k Chinese Accelerators

SemiAnalysis posted that it had officially disclosed for the first time running inference on over 100k Chinese accelerators. The firm attributes the disclosure to its English translated report and states that a GLM-5.3 powered internal infra agent halved the time needed to optimize the infra. The post notes the report's phrasing of a cluster of accelerators but offers no further detail on the claim. The account identifies itself as the official channel for the semiconductor and AI infrastructure research firm.