China-based Alibaba Group (NYSE:BABA) and Baidu (NASDAQ:BIDU) are now utilizing processors designed internally to train some of their artificial intelligence models, partly displacing Nvidia (NASDAQ:NVDA) chips, according to The Information.
A combination of U.S. export controls and the Chinese government pressuring companies to use domestically produced processors has led to the partial switch, according to the report, citing people familiar with the matter. Alibaba has started training smaller models with its Zhenwu processing unit, while Baidu has experimented with its Kunlun P800 chip.
However, Alibaba and Baidu are both still heavily reliant on Nvidia’s processors for training their most advanced models. The sources said the most advanced Zhenwu chip performs moderately better than Nvidia’s A100. The A100 was released in 2020. Baidu’s Kunlun P800 has primarily been used in post-training, or the refining stage, of AI models.
Alibaba has reportedly worked with Semiconductor Manufacturing International Corporation, a partially state-owned foundry in China, to produce its Zhenwu chips.
“The competition has undeniably arrived,” an Nvidia spokesperson said to The Information. “Customers will choose the best technology stack for running the world’s most popular commercial applications and open-source models. We’ll continue to work to earn the trust and support of mainstream developers everywhere.”
Nvidia has reduced its China-based revenue by more than half over the past year. In fiscal year 2025, China and Hong Kong accounted for about 13% of total sales, but by the second quarter of fiscal 2026, Nvidia’s most recent earnings report, the figure was reduced to 6%. Even with the decline in sales to China, Nvidia’s year-over-year revenue surged 56% in the second quarter.