According to AI Technology Review, Zhipu AI has recently completed its acquisition of ZhongKeJiaHe, an AI heterogeneous computing software infrastructure company, for over 100 million RMB. The deal aims to strengthen Zhipu's capabilities in large language model engineering and compiler optimization to address inference bottlenecks amid surging user demand.
ZhongKeJiaHe's core technology stems from the Chinese Academy of Sciences' compiler lab. Its virtual instruction set architecture unifies fragmented domestic chips into a cohesive computing cluster, and its SigInfer inference engine reportedly reduces model latency by up to 74x. The acquisition will improve Zhipu's per-token inference costs and support its planned custom AI chip development, following recent pressure from GLM-5.2 model adoption and U.S. sanctions.