Zhipu Explores In-House AI Chips Amid GLM Demand Surge

According to an exclusive report by The Information, Chinese AI company Zhipu is exploring the development of in-house AI chips to cope with surging demand for its GLM models and the pressure of US export controls. This move marks a further step in Zhipu's infrastructure layout.

Triggers and Key Details

The immediate trigger for Zhipu's consideration of custom chips is the explosive growth of GLM-5.2. The model's average daily token usage reportedly jumped 27 times in its first week on the Vercel aggregator, making it the fastest-growing model there. To meet this surging inference demand, Zhipu has made initial inquiries with Chinese chip design companies, though no partner has been selected yet.

Strategic Considerations and Context

Following DeepSeek, Zhipu has become the latest Chinese AI lab to explore custom ASICs. Author rohanpaul_ai points out that while custom ASICs sacrifice some flexibility, they can effectively reduce power consumption and per-token costs, making them highly suitable for large-scale inference of fixed models. Under the dual pressure of high computing demand and external export controls, developing proprietary chips has become a vital strategic option for leading Chinese AI companies to secure their computing supply chains.

2026-07-07 ~ 2026-07-08 · 5 related posts