Baidu Q4 conference call: In 2024, Baidu's intelligent cloud AI-related revenue will grow by nearly 300%, and the Wenxin large model 4.5 will be open-sourced

On the evening of February 18, during Baidu's Q4 2024 and full-year earnings call, Baidu founder Robin Li revealed that the Wenxin large model 4.5 will be open-sourced, and 4.5 will be the most powerful large model in Baidu's history, "hoping that customers and users can experience this model more conveniently than before." He stated that the decision to open-source the 4.5 series stems from a strong confidence in their technological leadership, and that open-sourcing will further promote the widespread application of the Wenxin large model and expand its influence in more scenarios: "But I want to emphasize that whether open-sourced or closed-source, the foundational model only has real value when it solves real problems on a large scale." The API call volume for the Wenxin large model has grown from 50 million API calls last year to 1.65 billion calls. Currently, Baidu's AI transformation in search has made progress, with 22% of search result pages containing AI-generated content. The forms of search have diversified, now including short videos, live broadcasts, and various other content, enhancing user experience. Through the Wenxin Yiyan intelligent assistant, Baidu helps businesses continue to operate efficiently during holidays, increasing the advertising effectiveness for clients. Baidu expects its advertising business to gradually recover by 2025, especially with revenue growth driven by AI-generated search results. The earnings call disclosed that in 2024, Baidu's intelligent cloud AI-related revenue will grow nearly threefold year-on-year. Driven by AI, Baidu's intelligent cloud business saw strong growth in the fourth quarter, with quarterly revenue increasing by 26% year-on-year. Currently, Baidu's intelligent cloud has the largest implementation scale of large models in the country, providing AI services to over 60% of state-owned enterprises and numerous private enterprises. During the call, Baidu's management revealed that Baidu is becoming the preferred choice for mid-sized clients. In terms of autonomous driving, Baidu pointed out that the Luobo Kuai Pao autonomous driving service performed strongly, providing approximately 1.1 million public rides in Q4 2024, a year-on-year increase of 36%, and reaching a cumulative ride volume of 9 million in January. The company has achieved 100% fully unmanned driving operations. Baidu has obtained a permit for open road testing in Hong Kong and plans to expand its autonomous driving services globally. Additionally, Baidu stated that despite facing pressure, while maintaining AI transformation and strategic investments, it expects core profit margins to gradually improve by 2025. Investments will continue to focus on high-growth areas, including AI cloud, autonomous driving, and foundational models, ensuring efficient resource allocation and improved operational efficiency. Baidu also plans to return value to shareholders through measures such as stock buybacks, with a plan to implement a $5 billion stock repurchase program by December 2025, further enhancing market confidence. The following is the full Q&A: Q: Baidu recently announced the open-sourcing of the Wenxin large model 4.5 series. What is the strategic thinking behind this decision by management? How does Baidu view the current competitive landscape of foundational models? How does it plan to maintain its leading position in this rapidly developing market? A: The generative AI and foundational model market is still in its early stages but is developing at a very fast pace. The success story of DeepSeek will undoubtedly accelerate the popularization of foundational models, as foundational models become more accessible and affordable. I believe we are entering a truly transformative phase. We will see a surge of new AI applications and use cases, which will bring incredible opportunities and push the boundaries of AI technology From DeepSeek, we learned one thing: that open-sourcing the best models can greatly promote people's use or adoption of that model. When a model is open-sourced, people naturally want to try it out out of curiosity, which helps with broader adoption. The Wenxin large model 4.5 will be the best model we have ever created, and we hope that users and customers can try it out more easily than ever before. Therefore, we decided to open-source the Wenxin large model 4.5 series, based on our deep confidence in our technological leadership. This confidence comes from our decade-long investment in research and development, continuous technological innovation, and our unique position as one of the few companies in the world with UM4 AI stack capabilities. The Wenxin large model has already proven its strong market appeal, growing from 50 million API calls last year to 1.65 billion. So through open-sourcing, we believe that more developers and users will recognize the true value of the Wenxin large model, driving broader adoption and expanding its influence in a wider range of scenarios. Similarly, making the Wenxin Yiyan chatbot available for free will allow more users to compare our foundational model with other models under the same conditions, as they are not charged. The last significant upgrade of the Wenxin large model was in October 2023, so it has been a while since then; stay tuned for the upcoming Wenxin large model 4.5. However, I want to emphasize one very important point. Whether open-sourced or closed-source, the true value of the foundational model can only be realized when it effectively solves real-world problems and achieves large-scale applications. We have been committed to an application-driven approach, continuously iterating the Wenxin large model. With this mindset, since its launch, we have successfully transformed our products using the Wenxin large model while serving enterprise customers. We have successfully applied the Wenxin large model to transform our consumer products, such as Baidu Search and Baidu Wenku. Externally, through the Qianfan platform, we are enhancing the experience of enterprise customers in model and application development. The industry-leading capabilities of the Wenxin large model in instruction following and state-of-the-art retrieval-augmented generation technology have greatly reduced hallucination issues, and it has been widely applied in various scenarios. The comprehensive toolchain of the Qianfan platform also allows our customers to customize any model based on their application scenarios. Looking ahead, we plan to accelerate the iteration of the Wenxin large model based on the potential for performance improvement and cost reduction. We will continue to evolve in areas that have the greatest impact on the real world. We are excited about the new chapter in AI development and look forward to seeing AI technology bring more groundbreaking applications that create social value. Thank you. Q: My question is about the search business. After the recent revisions to the search over the past few quarters, how should we view future adjustments or changes? What is the current proportion of generative AI results, and what is our target proportion? With the deepening integration of AI, can you share more details about user metrics, considering that AI chatbots like DeepSeek and Doubao also provide search functionality? So, how do you view the competitive dynamics between Baidu Search and these AI chatbots? A: Currently, about 22% of all search result pages contain AI-generated content. However, there is still a lot of work going on behind the scenes as we fundamentally transform search, making it more powerful and efficient than ever before. With Ernie, we have expanded search beyond text and links, providing a variety of content formats, including short videos, live broadcasts, notes, and product cards. Different formats can be dynamically combined to create a personalized search experience. As we continue to transform search, we are also developing capabilities that enable deeper personalization, adapting to each user's habits and preferences. Our efforts have already led to better user outcomes. In January of this year, we found that among Baidu's monthly active users, 83% interacted with generative AI content. Even more encouraging, in December, the search query volume per user on the Baidu App increased by 2% year-on-year. We will continue to focus on continuously improving user experience and expanding the role of AI in search, as we see increasing user engagement. On this basis, I would like to share a small case study illustrating how the Wenxin Yiyan intelligent assistant creates value for our advertisers. During this year's Spring Festival holiday, many small and medium-sized enterprises took a week off, while customer demand in certain industries peaked. Our Wenxin Yiyan intelligent assistant effectively filled this gap. For example, a hearing aid company could not operate its customer service team during the Spring Festival due to many employees being on leave. However, through the Wenxin Yiyan intelligent assistant, we helped the company effectively identify and filter highly relevant potential sales leads, allowing them to efficiently follow up and serve customers even with reduced staffing during the holiday. Regarding the issues you mentioned, such as AI chatbots, we believe the AI revolution is still in its early stages. Whether it is AI-native chatbots like DeepSeek or our AI-driven products, they are different ways to explore the potential of AI. As the leader in the Chinese search market, we remain agile with hundreds of millions of users, have a comprehensive market perspective, continuously drive the best innovations, and integrate truly innovative AI capabilities. We will continue to lead the AI transformation and enhance the search experience for our large user base. Furthermore, search is inherently deeply rooted in language and text understanding, which aligns closely with the capabilities of large language models. Therefore, our positioning is to become the leader in AI-driven search transformation. We believe that search is evolving into an integrated platform that goes beyond AI chatbots, providing not only intelligent answers but also guiding users through the entire process, from finding answers to providing in-depth analysis, completing tasks, and ultimately delivering comprehensive services and solutions. While AI chatbots represent an important but early stage in the development of AI applications, there has yet to be a killer application. For us, the key is to maintain a fast and steady pace of AI transformation, focusing on discovering the next-generation search experience that users truly need and want. Q: Could you please comment on the core advertising growth prospects for the first quarter and the full year of 2025? If possible, could you share your macro assumptions in these advertising business outlook forecasts? Do you think the advertising business has bottomed out in the fourth quarter of 2024 and is starting to recover? Lastly, could you share some potential monetization opportunities related to generative AI search?** A: Yes, we have seen various types of policies introduced in the past few months, such as monetary easing, fiscal stimulus, and trade policies. While we believe these measures will ultimately benefit the economy and will take some time to take effect, given that our advertising business is highly correlated with small and medium-sized enterprises, which are particularly vulnerable to the macroeconomic environment, coupled with the still challenging competitive landscape, we have not seen significant improvement in the fourth quarter. Despite facing short-term pressures, our transformation work combining search with generative models is still progressing steadily. We believe these AI transformations will continue to enhance user experience and create new opportunities. This quarter, we further deepened the AI transformation of search and saw some encouraging improvements in user metrics, as mentioned in our financial report. We believe this will drive revenue growth and open up new monetization opportunities for us in the future. Additionally, we have not yet begun large-scale monetization of generative AI search results, with approximately 22% of queries currently involving AI-generated content. Once our AI search capabilities are sufficiently refined, we will advance the monetization of higher-quality user products. Therefore, we see growth opportunities ahead. Based on these factors, we believe our advertising business is close to its low point, and the application business will gradually improve, with expectations that performance in the first half of the year will be better than in the fourth quarter, and further improvement in the second half compared to the first half. Q: My question is about AI cloud. Given the strong growth of your AI cloud business this quarter, can we expect it to continue on a positive trajectory in 2025? What are the main drivers behind this momentum, in terms of both revenue and profitability? Additionally, can management share insights on the outlook for the AI cloud market in 2025? How should we view enterprise cloud demand? What incremental opportunities does AI bring to this segment? A: Our AI cloud revenue growth accelerated in the fourth quarter, with a year-on-year increase of 26%, contributing to an annual revenue growth of 17%. Notably, revenue related to generative AI grew nearly threefold year-on-year in 2024. This growth is driven by the increasing demand for Wenxin Yiyan and our AI infrastructure, as well as strong market recognition of our technological leadership. As a result, we have successfully attracted a diverse customer base and established a robust opportunity pipeline. In terms of AI infrastructure, we have established partnerships across multiple industries, including internet, automotive, smart devices, manufacturing, energy, finance, utilities, and numerous AIGC startups. Our customer base continues to grow healthily, with strong growth among top-tier and mid-sized clients, demonstrating our expanding market share in the Chinese cloud market. Our Qianfan large model platform offers comprehensive foundational models with industry-leading trustworthiness and performance. We have launched a series of Wenxin large models ranging from flagship to lightweight versions, designed to meet various needs. In addition to Wenxin large models, we also provide a range of high-quality third-party foundational models, including DeepMind's V3 and R1 models With our full-stack AI capabilities and end-to-end optimization, we can ensure that any model hosted on our platform achieves optimal performance and stability while maintaining highly competitive pricing. In addition, we provide a complete set of tools for fine-tuning models and building native AI applications, helping clients fine-tune models and build AI-native applications, enabling them to easily develop solutions that meet their specific needs. Regarding the AI cloud market, we believe it will experience rapid growth in 2024 and 2025. On one hand, during the recent Spring Festival holiday, large language models became a widely discussed topic. This not only further raised public awareness of foundational models but also encouraged more people to think deeply about how to leverage these models to enhance their businesses. On the other hand, we believe the performance of foundational models will continue to improve while costs will steadily decline, further lowering the barriers to using these models. Therefore, we believe more enterprises will integrate foundational models into various aspects of their business operations, from R&D to production to support, driving explosive growth in API calls. We also anticipate an increase in spending on Baidu's AI cloud services, as we have observed a clear trend in the past: customers using foundational models through the API calls of the Qianfan large model platform also tend to increase their spending on our AI cloud services. In terms of profits, due to our focus on high-value opportunities aligned with strategic priorities, we have driven sustainable strong growth, and our operating profit margin under non-GAAP for the fourth quarter of 2024 continues to expand year-on-year. Looking ahead to 2025, we are confident that the revenue growth of AI cloud will maintain strong momentum while continuing to achieve positive operating profits. Q: I would like to ask about our profit margin. Given the short-term pressures facing our core advertising business and the continuous increase in cloud business revenue, how should we view the core profit margin levels for the first quarter and the full year of 2025? Is there further optimization space for improving operational efficiency? Additionally, are there any annual investment plans we need to be aware of? A: Despite facing some pressures in the short term, we remain optimistic about our strategic BI investments, believing that these investments will yield more sustainable impacts. Therefore, we will continue to focus on enhancing the resilience of our business operations and see positive progress across various business areas. First, for our online marketing business, we expect advertising revenue to gradually improve, primarily benefiting from our plans to monetize AI-generated search results on a large scale while also being prepared to seize growth opportunities when the macro economy improves. Second, regarding our AI cloud business, it has already demonstrated strong growth, and as our competitive advantages in the market continue to strengthen, profit margins are also improving. We are confident in maintaining this strong momentum. Third, for our entire autonomous driving business, especially robotaxi, we remain focused on narrowing losses by improving operational efficiency and enhancing user experience Regarding the investment and optimization issues you mentioned, in 2025, we will continue to optimally allocate resources in high-growth opportunity areas while maintaining consistency with our long-term strategy. Our investments will continue to focus on projects that not only have significant future opportunities but also strong return potential, including further enhancing Ernie's capabilities, deepening the AI transformation of search, strengthening AI cloud services, and expanding the advancement of strategic areas such as autonomous driving. As these plans progress, we will remain committed to improving operational efficiency and creating synergies across business groups to maximize the impact of these strategic investments. Q: My question is about the RoboTaxi business of Luobo Kuaipao. Can management share the progress of the RoboTaxi business as we enter 2025? What are our goals in terms of fleet size, unit economics, and commercial value? Is the industry approaching a turning point? What is the view on the competitive landscape? We have noticed that competitors in China's RoboTaxi sector are leveraging ecosystem strategies with automakers and ride-hailing platforms. What is Baidu's competitive and scaling strategy? A: We have invested in advanced driving technology for over a decade, and through Luobo Kuaipao, we have turned our grand vision into reality, making ourselves one of the world's leading autonomous driving technology companies. We operate in one of the most challenging environments globally; as you know, China's traffic conditions are among the most complex in the world due to its large population, diverse road conditions, dynamic traffic scenarios, and complex urban layouts. The ride-hailing prices in China are about one-seventh of those in the United States, so our successful operations demonstrate our technological and operational capabilities. We designed the RT6 from the ground up, which is our sixth-generation autonomous taxi and the most cost-effective taxi vehicle globally. With these advantages, we have successfully validated our business model and laid a solid foundation for further expansion and globalization. In the fourth quarter, Luobo Kuaipao provided approximately 1.1 million nationwide rides, a year-on-year increase of 36%. By January, the cumulative number of public rides exceeded 9 million. Additionally, as I mentioned earlier, we have achieved 100% fully driverless operation in China, meaning there are no safety drivers in our vehicles anymore. This sets a new industry benchmark and further solidifies our leadership position. As I mentioned before, last November, Luobo Kuaipao received approval to begin open testing in Hong Kong. This is an important step, as Hong Kong is our first right-hand drive and left-hand traffic market, proving our ability to adapt autonomous driving technology to different traffic systems and opening doors for entry into other markets with similar driving systems. This year will be a key year for our expansion, and we expect fleet size and ride volume to grow faster than ever before. In the process of advancing, we are also actively seeking collaboration opportunities and have identified several potential partners, including ride service providers, local taxi companies, third-party fleet operators, and other potential partners. This light-asset model will enable us to scale efficiently while maintaining flexibility, leveraging a diverse group of partners to serve more people As I mentioned in the last earnings call, since the market is still in its infancy, competition actually helps accelerate market development and promotes a more favorable innovative regulatory environment. In this growing market, our mission remains clear: to provide more users with a safer, more convenient, and more comfortable travel experience. Q: Looking ahead to 2025, what will be Baidu's strategic focus in terms of business investment? Specifically, how will resources be allocated among search, autonomous driving, cloud services, and foundational models? Additionally, what is management's outlook on capital expenditures and shareholder returns for the full year of 2025? A: When evaluating capital allocation options for 2025, we will guide our decisions based on the following strategic considerations: we will continue to invest in advancing our AI capabilities as a long-term strategic priority. Based on this, we will further deepen our AI transformation, particularly in the area of search products. In terms of AI cloud, our goal is to drive enterprise customers' adoption of our AI infrastructure, PaddlePaddle deep learning platform, and Ernie model to meet their growing demand for AI products, solutions, and AI cloud services, thereby driving market growth. Therefore, we will focus on expanding domestic operations, exploring innovative operational models, and increasing our international market share. At the same time, we will ensure strict investment return management and effective cost control for capital expenditures. In strengthening technological leadership and accelerating the adoption of AI products and AI cloud services across various industries, we will continue to advance these priorities to drive our decisions. Regarding shareholder returns, we have repurchased over $1 billion in shares since the beginning of 2024, a figure significantly higher than the total repurchase amount for the entire year of 2023. Overall, we have repurchased $1.7 billion in shares and plan to implement a $5 billion repurchase program by December 2025. Looking ahead, we plan to accelerate the repurchase program as part of our ongoing commitment to return value to shareholders and thank them for their long-term trust and support