[Zhipu Acquires Zhongke Jiahe for Billions, Securing Domestic Chip Adaptation Capability] According to monitoring by Beating, Zhihu has spent billions to acquire Zhongke Jiahe. The latter is an AI infrastructure company, with its team originating from the Institute of Computing Technology at the Chinese Academy of Sciences. Its core product acts as a 'translation software' layer between models and various chips, enabling the same model to run more easily on multiple domestic chips. Zhipu is currently facing engineering pressures from large-scale inference. After the launch of GLM-5, some users encountered issues such as garbled text, repetition, and rare characters. Zhipu's investigation revealed that the problems stemmed from cache conflicts under high concurrency. A technical review of the issue specifically thanked Zhongke Jiahe for their support and participation. Following the acquisition, Zhongke Jiahe can help Zhipu reduce redundant adaptation efforts for domestic chips, improve computing power utilization, and lower inference costs. However, this only addresses compilation and chip adaptation capabilities. Issues such as cluster scheduling, network stability, and computing power supply still require Zhipu's continued efforts. [Original Link]