Following Li Yanhong's absence from a crucial meeting, Baidu garnered market attention once more by releasing a financial report that indicated a decline in revenue.

In 2024, Baidu recorded revenues of RMB 133.1 billion, marking a year-on-year decrease of 1%, with fourth-quarter revenues amounting to RMB 34.12 billion, a 2% year-on-year drop. The overarching trend in Baidu's financial report is that the growth rate of AI fails to offset the decline in advertising revenue.

Profits, however, appear more robust, with net profit attributable to shareholders in 2024 reaching RMB 23.76 billion, an increase of 17% from the RMB 20.3 billion recorded in 2023. Had it not been for a provision of around RMB 1 billion related to the closure of Jiyue Auto in the fourth quarter, net profit would have been even more substantial.

Despite essentially unchanged revenue levels, Baidu's net profit witnessed a significant increase, primarily due to a reduction of approximately RMB 2 billion in R&D expenses, a 9% overall decrease, attributed to lower R&D personnel costs.

While Baidu opts for "cost-cutting" measures such as layoffs and salary reductions, DeepSeek is expanding its recruitment efforts with high salaries to strengthen talent barriers, creating a stark contrast.

After Baidu's financial report was released, its share price fell by 7.5% in U.S. stock markets.

In fact, since November 2024, Baidu's share price has dipped below its net asset value, with a current price-to-book ratio of approximately 0.86, indicating that its market value is just 86% of its net asset value. In Baidu's asset structure, cash holds a significant proportion: as of the end of 2024, Baidu's cash and short-term investments totaled RMB 137.4 billion, most of which are wealth management products, accounting for about 56% of its total market value and roughly one-third of its total assets.

Assets are considered valuable only if they can generate revenue, appreciate in value, and bring future cash flows. Baidu's share price falling below its net asset value reflects investors' concerns about the long-term decline in the advertising business overshadowing the short-term growth of AI.

Whether Baidu will utilize its substantial cash holdings for share repurchases may be a crucial factor affecting its market value in the short term. The long-term investment logic hinges on whether the advertising business can improve and when such improvement will materialize. Furthermore, when will AI truly assume a leading role?

I. Traditional Advertising: The Search Moat Faces Accelerated Collapse

It is no secret that Baidu's dominant position in search is precarious, and the outside world is keenly observing how far it may still fall.

In the fourth quarter of 2024, Baidu's online marketing revenue decreased by 7% year-on-year to RMB 17.9 billion, marking three consecutive quarters of negative growth. Although advertising revenue still accounted for 55% of Baidu's total revenue for the year, the number of active users and user engagement remain unpromising.

In mid-2024, the monthly active users of the Baidu Mobile App reached 703 million, decreasing to 679 million by year-end, with user time continuing to shift towards platforms like Douyin and Xiaohongshu. ERNIE Bot was initially seen as a potential solution to the peak traffic dilemma, but fierce competition from Doubao and Kimi has made this path less straightforward.

Amid peak traffic challenges, improving the efficiency of the advertising system is also crucial for salvaging Baidu's declining advertising business, and the introduction of AI technology is highly anticipated.

In the second half of 2023, the ERNIE Bot large model comprehensively revamped its business and advertising systems, enhancing ad placement accuracy and effectiveness through precise analysis of user behavior and needs. At the time, Li Yanhong stated during the third-quarter financial report conference that it was expected to bring in hundreds of millions of incremental revenue. However, in reality, Baidu's online marketing revenue in the fourth quarter was RMB 19.2 billion, a year-on-year increase but a quarter-on-quarter decrease of 2.5%.

Moreover, AI search itself presents a significant disruption to the traditional search model, as embedding advertising content directly into AI-generated answers is considerably more challenging than pay-per-click ranking. Compared to Baidu's traditional search model, which is often criticized for being overly commercialized, the monetization of AI search is currently relatively restrained.

The performance in 2024 also underscores that AI has not reversed the long-standing decline in Baidu's advertising business.

However, during the financial report conference, Baidu also mentioned that currently, about 22% of search result pages contain AI-generated content, and once the AI search function is fully refined, monetization efforts will continue to advance.

Search is deeply rooted in the understanding of language and text, which perfectly aligns with the functionality of large language models (LLMs). The smarter a company's large model, the more accurately it can comprehend user intent and uncover potential needs.

DeepSeek's popularity stems not only from its low training costs but also from its deep thinking capabilities and enhanced human understanding.

Looking ahead, Baidu envisions future search as an "integrated platform" that not only provides intelligent answers but also guides users through the entire journey from finding answers to in-depth analysis, task completion, and ultimately offering comprehensive services and solutions.

There are numerous opportunities for monetization within this framework, such as embedding more native advertisements within answers, where clicking on the advertisement can directly lead to appointments, orders, location searches, car-hailing, and more.

The process of reconstructing search value involves upgrading information retrieval to decision-making services, necessitating comprehensive innovations in content ecosystems, payment systems, and data closed loops.

In other words, the allocation of future advertising budgets will increasingly depend on the comprehensive capabilities of "AI + ecosystem."

Taking the content ecosystem as an example, regardless of AI's impact, it is evident that more people are using Xiaohongshu as a search engine, largely due to its more authentic, useful, relevant, and interactive content.

In the AI era, answers may be generated directly on social platforms, intelligent assistants, or even office software, with search behavior being deconstructed into countless fragmented interactions. The trend towards fragmented entry points is likely irreversible. If Baidu's content ecosystem is not superior, its situation will not improve merely by integrating DeepSeek.

Certainly, Baidu is also enhancing the search experience, such as expanding search beyond text and links to include diversified content forms like short videos, live streams, notes, and product cards, thereby creating a personalized search experience.

However, WeChat, which boasts higher-quality and in-depth content, has also integrated DeepSeek, with AI search being the first scenario. This undoubtedly poses a significant challenge to Baidu.

In the future, with the deep integration of AI search functions, WeChat's social data, transaction channels, and payment systems can form a seamless integration, potentially realizing "search as a service." Of course, Baidu also has its own ecosystem, including mini-programs and maps, but overall, there is still considerable room for improvement.

II. AI: Repetitive Commercialization Paths

Thanks to robust demand for AI, Baidu Intelligent Cloud achieved a year-on-year increase of 26% in the fourth quarter, a notable highlight in the financial report. The report also revealed that the daily average API calls processed by the ERNIE Bot large model reached 1.65 billion in December 2024, a 32-fold increase from the previous year, with external API calls growing by 178% month-on-month. The monthly active users of Baidu Wenku's AI functions reached 94 million, a year-on-year increase of 216% and a month-on-month increase of 84%.

The momentum is extremely strong. However, there has also been a negative trend. In November 2024, the daily average API calls of the ERNIE Bot large model were 1.5 billion, and in August, they were 600 million, indicating a significant slowdown in growth.

The development of large models profoundly impacts Baidu Intelligent Cloud's revenue.

Shortly before releasing its financial report, Baidu made the decision to open-source its technology. On February 13, Baidu announced that "ERNIE Bot" would be free starting April 1. On February 14, Baidu announced that the ERNIE Bot 4.5 series would be successively launched, with official open sourcing scheduled for June 30.

Prior to this, Li Yanhong had repeatedly and publicly emphasized the advantages of closed-source models. He believed that closed-source models can maintain a leading position and that only closed sourcing can lead to a viable business model, earning money that can be focused on computing power and talent. He even stated, "Open sourcing is actually an IQ tax." "Today, whether it's ChatGPT, ERNIE Bot, or other closed-source models, they are undoubtedly more powerful than open-source models and have lower inference costs."

After being impacted by DeepSeek, Baidu made a complete U-turn.

During the recent financial report conference call, Li Yanhong's viewpoint shifted: "From the success of DeepSeek, we have realized that open-sourcing the most outstanding models can significantly promote their application. When models are open-sourced, people will naturally try them out of curiosity, thereby expanding the model's influence across more scenarios." "Whether the model is open-sourced or closed-sourced, it truly has value only when the foundational model can effectively solve practical problems on a large scale."

DeepSeek has altered the competitive landscape of the AI industry, shifting the focus from parameter scale to algorithmic efficiency and ecosystem vitality. Baidu's shift towards open sourcing is also aimed at fostering a thriving developer ecosystem.

Certainly, Baidu is not the only one whose attitude has changed. Sam Altman, CEO of OpenAI, recently also stated that "open-source strategies need to be re-evaluated."

JPMorgan Chase's perspective on this is that given the rapid development of the generative AI market, which can easily erode ERNIE Bot's leading competitive position in terms of model quality and cost efficiency, it is reasonable for Baidu to shift its strategic focus to maximizing the adoption of its AI products. It also pointed out that the cost involves lowering the priority of short-term monetization.

Open sourcing offers benefits such as increased influence and lowered barriers to entry for developers and enterprise users. Additionally, the free availability of ERNIE Bot can attract C-end users, forming a business model of "open-source traffic + cloud service monetization" in conjunction with Baidu Intelligent Cloud.

However, as JPMorgan Chase noted, Baidu needs to de-prioritize short-term monetization, such as subscription fees for ERNIE Bot and charges under the closed-source model. The competition for ecosystems is also a new battle that has just begun.

III. Final Thoughts

Everyone acknowledges Baidu as the Huangpu Military Academy for AI talent, and everyone is aware that Baidu proposed the slogan "All in AI" very early on. However, when discussing whether Baidu will be a leading player in the AI era or even whether it will remain among the last standing, people often recall the years when Baidu started early but arrived late, standing at the forefront of trends yet being squeezed out: video, group buying, food delivery, post bars, new energy vehicles, etc.

When it comes to AI, cooperation with Apple was intercepted by Tongyi Qianwen.

Why does the highly forward-thinking Baidu always end up as an "unlucky" player of the times?

Compared to the advertising business, which is inherently constrained and can only follow industry trends, and in contrast to the debate between open sourcing and closed sourcing, Baidu may require an even more thorough self-revolution.

With the current situation of "continuing decline in advertising share, new AI businesses not yet sufficient to fill the gap, old and new growth drivers in the process of transitioning, and a price-to-book ratio of only 0.86," the time left for Li Yanhong may be even more unforgiving than the figures in the financial report.