Baidu unveils next-gen AI chips

Baidu introduced a new generation of artificial intelligence processors and large-scale computing systems at its annual Baidu World conference on Thursday, marking another step in China’s efforts to strengthen its supply of locally developed high-performance technology. 

The company announced two semiconductors designed to provide Chinese firms with reliable and cost-efficient computing power. The M100 chip, aimed at inference tasks, is expected to enter the market in early 2026. The M300, which can support both model training and inference, is scheduled for release in early 2027. Training involves building an AI system using extensive datasets, while inference enables the system to generate predictions and process user requests.

Baidu has been working on its own chip designs since 2011. At this year’s event, it also introduced two large interconnected computing systems known as supernodes. These products link multiple processors to compensate for the limitations of any single chip and create a unified high-capacity computing environment.

The first system, Tianchi 256, will combine 256 of Baidu’s P800 chips and is expected to become available in the first half of next year. A larger version using 512 chips is planned for the second half of the year. Similar approaches have been developed by other Chinese technology companies, including Huawei, which launched its CloudMatrix 384 system based on 384 Ascend 910C chips and said earlier this year that more advanced versions would follow.

Baidu additionally presented an updated edition of its Ernie large language model, which the company said shows improved performance in text, image, and video analysis.

As Kazinform News Agency reported earlier, OpenAI recently announced the rollout of its latest AI models, GPT-5.1 Instant and GPT-5.1 Thinking, marking a significant upgrade in both performance and interaction quality for its flagship ChatGPT platform.