Baidu is integrating the AI stack from chips to cloud services, facing new opportunities as U.S. export restrictions on Nvidia H200 GPUs ease. Ten Chinese companies, including Baidu, received approval for imports through 2026.
At the Create 2026 conference, Baidu unveiled its roadmap to independence from Western chipmakers, launching the M100 chip and planning the M300 for 2027. Analysts predict a sixfold increase in chip sales to nearly $1.1 billion by 2026, with the Kunlun unit valued at $28 billion.
Baidu aims to enhance both AI models and hardware. Its latest frontier model, ERNIE 5.1, reported a 94% drop in pre-training costs, utilizing a Kunlunxin cluster.
The new focus comes as China’s AI landscape shifted due to DeepSeek's low-cost model, disrupting the belief that advanced AI mandates high-end hardware. Baidu capitalizes on this by optimizing both software and hardware, while selectively accessing Nvidia technology.
A potential separate IPO for the Kunlun unit could propel Baidu's growth, presenting a major player in the semiconductor market, while the efficiency of ERNIE 5.1 could redefine AI financial expectations. However, U.S. export policy remains a crucial and unpredictable factor in Baidu's strategic planning.