China’s Baidu is rapidly becoming a major player in the country’s AI chip sector, emerging as a challenger to other domestic firms — including Huawei — as both try to fill the void left by export restrictions on foreign GPUs.
Once best known as the leading search engine in China, Baidu has reoriented its strategy toward AI, autonomous driving and cloud infrastructure. Through its majority-owned subsidiary Kunlunxin, the company designs its own AI chips and sells them to third-party data centers, while also offering computing capacity via its cloud services.
Five-Year Roadmap: M100, M300, and Supernodes
At its recent flagship event, Baidu unveiled a roadmap calling for the release of the M100 chip in early 2026, optimized for inference; and the more powerful M300 chip in 2027, targeted at training ultra-large multimodal AI models.
Alongside individual chips, Baidu is building out large-scale infrastructure: supernodes such as “Tianchi 256” (256-chip configurations) and a planned “Tianchi 512” upgrade. Executives say these clusters will deliver major performance gains and enable deployment of trillion-parameter models at scale.
This vertical integration — chips, infrastructure, cloud services, and AI models — reflects Baidu’s ambition to provide a “full stack” AI offering. It already uses a mix of Kunlunxin chips and third-party GPUs to power its own ERNIE AI models.
Domestic Demand Surging Amid Supply Constraints
With U.S.-based GPU makers facing export restrictions — and Chinese firms hesitant to import even lower-end GPUs — demand for domestic AI chips is surging. Analysts at Deutsche Bank and JPMorgan have expressed optimism about Kunlunxin’s prospects, forecasting sharp growth in chip sales by 2026.
At the same time, shortages of AI-grade chips have hit other major Chinese tech companies. For example, Alibaba and Tencent have publicly cited limited chip supply as a constraint on data-center and AI expansion.
Baidu’s chip push — combining in-house hardware design, supernode infrastructure, and cloud offerings — may allow it to fill that supply gap and emerge as a strategic domestic supplier of AI computing power. As one industry analyst put it: success for Kunlunxin could make Baidu a “strategic supplier to the rest of China’s AI industry.”
Meanwhile, Google is reportedly in talks with Meta Platforms to supply its custom AI chips (TPUs) for Meta’s data centers, potentially challenging the market dominance of Nvidia. The potential Meta–Google deal signals a broader industry shift as hyperscalers seek alternatives to Nvidia GPUs and strive for greater control over AI infrastructure.
