TLDR: Chinese tech giant Baidu has introduced two new artificial intelligence chips, the M100 and M300, alongside its upgraded ERNIE 5.0 foundational model. These advancements, announced at Baidu World 2025, aim to bolster China’s AI self-sufficiency and enhance capabilities in large-scale inference and multimodal model training.
Beijing, China – November 13, 2025 – Chinese technology powerhouse Baidu today unveiled its latest innovations in artificial intelligence hardware and software at its annual developer and client event, Baidu World 2025. The company introduced two advanced AI chips, the Kunlun M100 and Kunlun M300, marking a significant stride in China’s ambition for technological self-sufficiency amidst escalating global competition in advanced computing capabilities.
The Kunlun M100 chip, designed by Baidu’s chip unit Kunlunxin Technology, is optimized for large-scale inference scenarios, particularly for models utilizing the mixture-of-experts (MoE) technique. This chip is slated for release in early 2026 and aims to significantly improve inference efficiency. Following this, the Kunlun M300, a more powerful training-optimized chip, is planned for a 2027 debut. The M300 is tailored to support the training of super-large multimodal models with trillions of parameters, a growing trend in the AI landscape.
Shen Dou, Baidu’s executive vice-president and president of its cloud unit, emphasized the strategic importance of these new chips. He stated that the M100 and M300 will provide ‘powerful, low-cost, and controllable AI computing power’ to underpin China’s drive for AI self-sufficiency. This initiative is part of a broader national effort to reduce reliance on imported advanced processors, such as those from Nvidia.
Baidu also outlined ambitious plans for its chip infrastructure. The company intends to cluster its M100 chips into a ‘Tianchi256’ stack, expected in the first half of next year, which is projected to deliver over a 50% performance gain in serving AI systems compared to its previous clusters. An upgraded version, ‘Tianchi512,’ integrating 512 chips, is scheduled for the second half of 2026. By 2030, Baidu aims to establish a ‘supernode’ capable of supporting ‘millions’ of chips.
Addressing the broader industry landscape, Robin Li Yanhong, Baidu’s co-founder, chairman, and CEO, commented on the ‘very unhealthy’ structure of the AI sector. He noted that the current system disproportionately rewards developers of chips and foundational AI models, suggesting that models built atop these chips, as well as applications, should generate more value. Li also highlighted the impact of AI on software development within Baidu, revealing that approximately half of the new software code produced by the company is now generated by its AI coding platform, Baidu Comate, with aspirations to reach 80% or 90% in the future.
Also Read:
- Cerebras Systems Unveils ‘Cerebras for Nations’ to Empower Global Sovereign AI Development
- Google Unveils Private AI Compute, Matching Apple’s Secure Cloud AI Approach
In conjunction with the new silicon, Baidu also announced ERNIE 5.0, its latest natively omni-modal foundational model. This advanced model boasts capabilities in multimodal understanding and generation, handling text, images, audio, and video inputs and outputs. The announcements underscore Baidu’s commitment to leading AI innovation and contributing to China’s technological independence.
