Alibaba and Baidu turn to in-house AI chips as U.S. curbs tighten on Nvidia

Alibaba has begun using custom chips for smaller AI models, while Baidu is piloting its Kunlun P800 for Ernie model training, signaling China's growing shift toward domestic AI hardware amid reliance on Nvidia

Chinese tech giants Alibaba and Baidu have begun integrating their self-developed AI chips into their machine learning workflows, marking a strategic shift away from heavy reliance on Nvidia, according to sources familiar with the matter.

Alibaba reportedly started using its custom processors for training smaller AI models earlier this year. Meanwhile, Baidu is said to be piloting its latest Kunlun P800 chip in the development of newer versions of its Ernie large language model. While neither company has entirely moved away from Nvidia’s hardware, the increasing adoption of domestic alternatives highlights a growing trend in China's push for tech self-sufficiency.

Shift driven by U.S. sanctions and national strategy

The shift comes amid tightened U.S. export controls that restrict access to high-performance AI chips for Chinese companies. As Washington continues to limit the shipment of advanced processors such as Nvidia’s H100 and Blackwell series, firms in China are ramping up efforts to develop competitive home-grown solutions.

Although Nvidia’s H20 chip — the most advanced model currently approved for sale in China — remains more powerful than many local options, internal users at Alibaba suggest that the performance gap is closing. Several company engineers report that Alibaba’s AI chips now rival the H20 in training efficiency for certain tasks.

In response to the evolving competition, an Nvidia spokesperson said the company will continue focusing on earning the trust of developers worldwide.

Despite growing domestic capabilities, both Alibaba and Baidu are expected to continue using Nvidia’s technology for their most advanced AI systems, balancing innovation with access to proven solutions.