SmartAPI has nearly 7 million users, with over 50,000 newly deployed domestic AI chips to ease the computational pressure.

August 10th, according to LatePost report, ZhiPu's MaaS open platform has nearly 7 million registered users, an increase of about 2 million since early July, with the number of enterprise customers reaching 23,000. Its developer product ZCode has exceeded 1 million users in just one month.

The report stated that ZhiPu's ARR (Annual Recurring Revenue) has grown by about 15 times since the beginning of this year. Market sources revealed that ZhiPu's current ARR may reach $2 billion, but this data has not been officially confirmed.

With the rapid growth in model inference demand, ZhiPu is expanding its domestic computing power infrastructure. It is understood that ZhiPu has deployed over 50,000 domestically produced AI computing chips to cope with the growing inference demand. Earlier, there were market rumors that ZhiPu had built a domestically produced AI computing infrastructure with a capacity of 1GW.

ZhiPu's growth is mainly driven by the surge in programming scenario demand. Following the release of the GLM-5 model, the company's ARR has grown rapidly, with the API business now becoming the core source of revenue. ZhiPu is also continuously optimizing inference efficiency, improving computing power utilization through techniques such as KV cache splitting and inference network architecture.

In addition, ZhiPu launched a high-speed version API in July, with a token generation rate of 400 per second, approximately 8 times faster than the regular API. On July 31st, the company reopened the purchase of the Coding Plan and significantly adjusted the prices, with the Lite version's monthly fee increased to 118 yuan.

Industry experts believe that as the demand for AI programming continues to grow, optimizing inference efficiency will become a core competitive advantage for large model companies. Whoever can generate more tokens on the same computing power will have an edge in terms of cost and commercialization.

Currently, Chinese large model companies such as ZhiPu, Kimi, and DeepSeek are facing computing power bottlenecks and model competition pressure. The market expects that several companies will launch a new generation of models in August, further intensifying the competition in AI large models.