Chinese Tech Companies Expand Use of Domestic AI Chips

Chinese technology companies are increasing their reliance on domestically produced chips to power artificial intelligence systems. Tencent said production of China-designed graphics processing units would progressively ramp up through the year. Alibaba discussed expanding use of its self-developed semiconductors during an earnings call on Wednesday.

The statements come as U.S. export restrictions have limited access to Nvidia technology for more than a year. Chinese firms have responded by accelerating development and deployment of local alternatives in pursuit of self-sufficiency for AI ambitions.

A range of domestic chip developers including Moore Threads, MetaX and Huawei have increased activity through public listings and new product launches. Alibaba designs its own AI chips and deploys them in data centers supporting its cloud computing business.

An Alibaba executive said on the earnings call that the company's T-Head proprietary GPU chips have achieved scaled mass production. The executive added that in an environment of compute scarcity this capability provides a structural advantage for revenue growth and gross margin improvement.

Alibaba also indicated it could sell servers equipped with its chips to other companies building data centers or partner to construct such facilities. The moves highlight the company's expanding role in China's semiconductor sector. Tencent made its comments one day before Reuters reported that the U.S. had approved several Chinese firms including Alibaba and Tencent to purchase Nvidia H200 chips.

No H200 chips have been produced under the approval so far. It remains unclear whether final U.S. government clearance has been secured. Reports over the past year have indicated varying levels of approval for Nvidia to ship certain chips such as the less powerful H20 to China while Chinese firms were at times encouraged to pursue domestic options.

Chinese companies advance toward systems capable of more complex tasks known as agentic AI they are increasing demand for sophisticated chips. The focus has moved from initial training of AI models to large-scale inference or running those trained models.

Domestic hyperscalers cannot delay deployment while domestic chip supply expands. Industry observers note the possibility of hybrid infrastructure combining Chinese and U.S. chips for inference workloads. Production of local GPUs is expected to increase steadily during the remainder of the year.