Intelliscan is considering developing an AI Chip in-house to address the rapid growth in GLM model demand

According to Omdia Beating monitoring, China's leading large-scale model company, Zhipei AI, is considering developing a customized inference chip and has started to approach local chip design companies for potential cooperation. Zhipei AI is currently on the U.S. blacklist and is unable to purchase NVIDIA's cutting-edge chips.

With the usage of its flagship open-source model GLM-5.2 on the Vercel developer platform skyrocketing by 27 times in a week, the shortage of computing resources is becoming increasingly severe.

Although Zhipei AI has deployed domestic computing power such as Huawei's and has conducted extensive software adaptation work, in order to overcome supply chain restrictions and reduce long-term cloud-based inference costs, Zhipei AI has chosen to follow the path of Google's TPU and OpenAI's custom chips.

This self-developed chip project is still in the early discussion stage. If it can progress smoothly, it will be manufactured by domestic semiconductor foundries, with the design-to-tape-out process expected to take more than two years.