The company's recent situation

We were invited to attend the 2024 Baidu Yunzhi Conference. At the conference, the company announced new developments in the three major areas of computing power, models, and AI applications, including upgrading the two major AI infrastructures of the Baige AI heterogeneous computing platform 4.0 and the Qianfan Big Model Platform 3.0, and upgrading the three major AI native application products of code assistants, intelligent customer service, and digital people.

reviews

The AI heterogeneous computing platform meets the computing power requirements of the whole process. At this conference, Baidu announced the upgrade of Baige's AI heterogeneous computing platform to version 4.0. Baige 4.0 has the ability to manage the computing power of 10,000 cards or even 100,000 card clusters. It provides ultimate computing power support for all types of enterprises throughout the process of cluster creation, development experiments, model upgrades, and model inference. In terms of performance, effective training time accounts for more than 99.5%, and overall performance is up to 30% higher than the industry average. Furthermore, Baige 4.0 is compatible with mixed training with mainstream AI chips at home and abroad such as Kunlun Core, Shengteng, Haiguang DCU, Nvidia, and Intel. We believe that the underlying computing power platform with high compatibility and high performance forms an important foundation for Baidu Cloud to achieve a closed business loop.

Build an “enterprise-grade” large model platform to lower the threshold for using large models. The company released the Qianfan Big Model Platform 3.0, which provides a complete model refinement tool chain in the model development layer; the model service layer supports calling the Wenxin series and domestic and foreign open source models, and also adds models with small vertical capabilities for enterprises to use; in the application development layer, Qianfan 3.0 upgrades enterprise-level RAG and enterprise-level agents, and enables application development based on dialogue. Currently, more than 0.7 million applications have been developed through the Qianfan platform, and the average daily call volume of large models on the platform has exceeded 0.7 billion. We believe that Qianfan Platform 3.0 is expected to further lower the threshold for using large models and provide every enterprise with equal opportunities for growth.

The three major AI native application products have been fully upgraded. On the infrastructure built by Baige and Qianfan, the company upgraded native application products for the three major fields of digital people, intelligent customer service, and code efficiency improvement. The AI native digital human platform Xiling Digital Human 4.0 can generate 3D digital images and videos in minutes; the AI native intelligent customer service customer service customer service Customer Yue's “self-service resolution rate” exceeded 92%, receiving more than 0.15 billion users and more than 0.5 billion interactions; the full-process AI code efficiency tool Wenxin Express Code launched two new functions: enterprise-level code architecture explanation and enterprise-level code review. Currently, Wenxin Express Code has served more than 0.01 million customers, helping increase R&D efficiency by 20%. We believe that the benchmark application of segmented racetracks is expected to drive the industry to explore the diversity of large model applications.

Profit forecasting and valuation

We maintain revenue and profit expectations, maintain outperforming industry ratings, and maintain target prices of US stocks and Hong Kong stocks at HK$116.0 and HK$112.3, corresponding to the 2024/2025 non-standard price-earnings ratio of 11 times/10 times for US stocks and Hong Kong stocks, respectively. Compared with current US stocks and Hong Kong stocks, there is 23% and 17% upward space compared to current US stocks and Hong Kong stocks, respectively (current US stocks correspond to 9 times/8 times 2024/2025 non-standard price-earnings ratio).

risks

The advertising business is under pressure, active users are declining, and AI transformation falls short of expectations.