Zhipu acquired Zhongke Jiahe for several hundred million yuan, buying domestic chip compatibility capabilities

According to Beating monitoring by DONGCHA, Zhipu has spent several hundred million yuan to acquire Zhongke Jiahe. The latter is an AI infrastructure company, whose team comes from the Institute of Computing Technology of the Chinese Academy of Sciences. Its core product is essentially like adding a layer of “translation software” between models and different chips, making the same model easier to run on a variety of domestic chips.

Zhipu is facing significant engineering pressure for large-scale inference. After GLM-5 went live, some users encountered garbled text, repeated loops, and uncommon characters. Zhipu’s investigation found that the issue was caused by cache conflicts under high concurrency. The related technical review also specifically thanked Zhongke Jiahe for participating in and supporting the work.

After the acquisition, Zhongke Jiahe can help Zhipu reduce repeated adaptation for domestic chips, improve compute utilization, and lower inference costs. However, this only addresses compilation and chip adaptation capabilities. Problems such as cluster scheduling, network stability, and compute supply still need Zhipu to continue solving.