Baidu Unveils Kunlun M100/M300 Chips and Ernie 5.0 Model at Baidu World 2025 Conference

Technology giant Baidu Inc. hosted its annual flagship conference, Baidu World 2025, in Beijing on November 13, 2025, where it announced crucial breakthroughs in both artificial intelligence hardware and software. The central highlights were the introduction of the next generation of its proprietary Kunlun chips and the significantly updated large language model (LLM), Ernie 5.0. These developments firmly reinforce the company's strategic push toward achieving technological self-sufficiency.

The new hardware portfolio includes two specialized chips. The Kunlun M100 is optimized specifically for large-scale inference operations and is scheduled for release in early 2026. The second chip, the M300, is engineered for the intensive training and inference demands of ultra-large multimodal models, with its official launch slated for early 2027. By developing these chips in-house, Baidu gains critical control over performance metrics and cost structures, effectively mitigating risks associated with potential restrictions on Western technology exports.

In addition to the chips, Baidu introduced two new supercomputing products designed to boost computational capacity. The Tianchi 256 supernode, which integrates 256 P800 chips, is expected to become available during the first half of 2026. A more robust system, the Tianchi 512, featuring 512 P800 chips, is anticipated to launch in the second half of 2026, further expanding Baidu's infrastructure offerings.

On the software side, the company showcased Ernie 5.0, an advanced large language model that operates with a staggering 2.4 trillion parameters. This model demonstrates native multimodal capability, allowing it to seamlessly understand and generate content across text, image, audio, and video formats. This sheer scale makes Ernie 5.0 twice as large as comparable systems, such as Alibaba’s Qwen3-Max and Moonshot AI’s Kimi K2, both of which possess just over 1 trillion parameters. Baidu Executive Vice President Shen Dou underscored the purpose of these innovations, stating they are intended to provide Chinese enterprises with “powerful, inexpensive, and controllable AI computing power.”

The urgency behind these announcements is rooted in escalating geopolitical tensions, which necessitate that Chinese firms secure domestic alternatives to ensure supply chain security. Baidu has been strategically investing in proprietary chip development since 2011, benefiting from strategic state support that emphasizes the national commitment to building a robust domestic semiconductor base. This proactive self-reliance strategy is expected to open up a significant new revenue stream for Baidu through the potential sale of Kunlun chips to external customers.

The market response was mixed following the announcement on November 13, 2025. Baidu shares (BIDU) closed on the New York Stock Exchange at $121 USD, registering a 2.5% decrease for the day. However, the company’s stock saw a slight rebound, posting a 0.8% gain during morning trading in Hong Kong on Thursday. The methodical, phased deployment of these new products, starting with the M100 in 2026 and concluding with the M300 in 2027, reflects a careful strategy aimed at tightly integrating hardware and software to cement Baidu’s leadership in the global AI landscape.