China’s Baidu announces two AI processors, new version of its Ernie model

Baidu has unveiled two new proprietary semiconductors for artificial intelligence (AI), along with a major update to its Ernie large language model (LLM). The company emphasised that the new products are designed to provide Chinese firms with powerful, low-cost, and domestically controlled computing power, directly addressing tensions that restrict access to advanced AI chips produced by companies like Nvidia.The announcements were made at Baidu’s annual Baidu World technology conference on Thursday (November 13).

New AI chips for training and inference

According to a report by news agency Reuters, Baidu, which has been developing proprietary chips since 2011, announced two distinct processors scheduled for release over the next two years:M100: An inference-focused chip set to launch in early 2026. Inference involves using existing AI models to make predictions and process user requests.M300: A more powerful chip capable of handling both training and inference, slated for release in early 2027. Training is the process of building AI models by teaching them patterns from vast datasets.Baidu also announced two new "supernode" products. These systems use advanced networking to link multiple chips, aiming to compensate for limitations in the performance of individual processors. Tianchi 256, which will comprise 256 of Baidu's P800 chips, will be available in the first half of next year.

A more powerful version utilising 512 of the P800 chips is scheduled for launch in the second half of next year.Huawei has also deployed a similar product called CloudMatrix 384. That system, comprising 384 of its Ascend 910C chips, is considered by some industry observers to be more powerful than Nvidia's GB200 NVL72, one of the US chipmaker’s most advanced system-level products.

Ernie AI model gets multimedia upgrade

Alongside the hardware push, Baidu unveiled a new version of its Ernie LLM. The company stated that the updated model significantly improves its capabilities, excelling not only at traditional text processing but also at advanced image and video analysis.