Baidu's Kunlun M100 launches on schedule; flagship M300 due next year, new N-architecture chips by 2028

Baidu on Thursday (November 12) laid out the latest roadmap for its Kunlun chips at the opening ceremony of Baidu World 2026: the M100 has launched on schedule, the flagship M300 will be released next year, and multiple chips based on an all-new N architecture will follow by 2028.

M100 targets inference and value; M300 is the flagship upgrade

According to on-site reports, the M100 is positioned for exceptional cost-effectiveness and large-scale inference workloads, while the M300, as the flagship, will bring comprehensive upgrades in computing power, memory and interconnect, serving both training and inference for very large multimodal models. The M100 and M300 are developed by Baidu's Kunlunxin chip unit and form the core of Baidu's in-house AI chip line.

Robin Li: compute demand is shifting from training to inference

Baidu founder Robin Li said at the event that AI computing demand used to be driven mainly by training, but has now shifted to a balance between training and inference, as the large-scale adoption of AI agents drives an explosive growth in inference workloads. He said Baidu would keep investing in technology for the long term.

The M100 and M300 are not new announcements: both chips were unveiled at Baidu World 2025 last year, when the M100 was planned for early 2026 and the M300 for 2027. This year's event confirmed the M100 has arrived on schedule, with the M300 still due next year. Progress in home-grown compute chips is closely watched as the United States restricts exports of advanced AI chips to China.