Writer: Xiangshan Finance and Economics

Baidu's third-quarter performance has been released, and it also announced that it will release a new version of its large model early next year. There are indeed many good news in this financial report, and several data growths are worth noting.

In the third quarter, Baidu's Wenxin large model was called 1.5 billion times, Baidu Search's AI function covered nearly 70% of monthly active users, Baidu Wenku's AI function had over 50 million monthly active users, with a year-on-year growth of over 300%.

Frankly, this growth actually exceeded my expectations a bit.

After all, 1.5 billion calls and 70% user function coverage can be considered a milestone. After all, currently, Google's search AI function coverage is less than 50%.

For domestic large model enterprises, this data is also quite encouraging.

There have been occasional doubts about AI large models in the market in the past two years, but I believe that Baidu's third-quarter report data is sufficient to dispel many people's doubts about this industry.

The number of calls to Baidu's Wenxin model and the active growth of AI user terminals also sufficiently demonstrate that Baidu's AI strategy is like the sun at 6 or 7 in the morning, bursting with strong vitality.

The End of the Old Era and the Beginning of the New Era

Corporate financial reports are an "information mine," and changes in numbers reflect the most vivid business information of the company at present.

Baidu's Q3 financial report is no exception.

After reading the third-quarter report and the financial report conference call, what is most memorable is the growth of several indicators of the large model business: Wenxin large model was called 1.5 billion times. Baidu Wenku's AI function MAU increased by 300%. Baidu Search's AI function covers 70% of users.

The sharp increase in the number of calls to Baidu's large model means that there is already an explosive trend on the application side.

These data also reflect a fact: After proposing a comprehensive shift to an "AI-centric strategy" six months ago, Baidu's large model business line has not only landed steadily but also grown rapidly. At the same time, these data also fully demonstrate that Baidu's adherence to its core AI strategy and application-driven approach is yielding initial results due to the correctness of its approach.

"In the long run, an AI-focused strategy will enable us to seize significant growth opportunities. Our long-term strategic focus is to enhance AI capabilities, which is the cornerstone driving our transformation towards AI," said Robin Li at the Q3 financial report conference call.

On the user side, the data growth indirectly reflects that products with large model capabilities are gradually becoming a "rigid demand" on the user side.

This can be cross-verified from the user experience.

In actual life perception, AI capabilities have become "ubiquitous." For example, during a recent business trip to Shanghai and Beijing, I stayed at several different hotels and found that all of them had service robots. More and more hotel brands are using large language models for services.

It's not just hotels; in fact, AI capabilities have become "standard configurations" for products such as mobile phones and cars in the past two years. And this "standard configuration" not only occurs in the consumer market but also includes productivity scenarios such as meetings and offices.

In fact, in the short term, apart from the consumer market, the growth of AI in productivity scenarios is more valuable because AI, as a rigid demand attribute of an efficiency tool, makes this scenario the closest to commercialization and the easiest to monetize.

Baidu's Q3 financial report data can also confirm this.

Third-quarter data showed a 300% increase in MAU for Baidu Wenku's AI function, with a significant increase in user volume. Moreover, there were 200,000 people queuing up for appointments on the first day of the free canvas's debut. It seems that there are still many rigid demands in the productivity market that have not been fully released, and subsequent MAU growth is promising.

In terms of industrial applications, Baidu Cloud's growth is also impressive. In the third quarter, intelligent cloud revenue reached 4.9 billion yuan, a year-on-year increase of 11%, and the proportion of AI-related revenue continued to increase to over 11%.

Cloud computing is a technology that complements AI services. The growth of Baidu Cloud also means that Baidu's second growth curve is already stable, and the B-end commercialization centered on the AI large model is steadily advancing.

Seeing this, I think if we put the puzzle pieces of growth together and view Baidu's large model business in the third quarter from a more macro perspective, perhaps we can gain more inspiration.

Growth indicates that for Baidu itself, the strategy with artificial intelligence at its core has yielded phased results.

First, internally, Baidu's long-term AI strategy has been continuously validated.

From the initial establishment of the AI strategy to the current good growth momentum of the large model, it has actually been quite challenging, but fortunately, there have been good results. With the correct results, the previous strategic investments have become meaningful.

Second, externally, the growth performance of Baidu's large model applications in Q3 has reassured the entire industry. A new era of large model applications is opening up.

The AI large model is a groundbreaking new technology. The growth of the AI large model industry is a long-cycle growth. After all, for any enterprise, transforming a new technology into a practical application that can solve problems is not simple.

It took 140 years from Nikola Tesla's invention of the induction motor in 1883 to the present, where electric vehicles are everywhere on the streets. Over these 100-plus years, from laboratories to factories and then to various products, ranging from small toys to large airplanes and cars, motor technology has become more efficient, smaller in size, and more versatile.

In this process, electromagnetic induction-related technology products have become "better to use, easier to use, and more useful" and have become one of the key basic technologies of modern science and technology.

Today's AI large model is the same.

One of the important reasons why Baidu's long-term AI strategy has been successful is that, while adhering to the long-term strategy, the development of Baidu's large model has always adhered to the "first principle" of application-driven. In other words, Baidu's AI strategy started from applications from the beginning.

In an interview, when asked about what is the first principle in Baidu's strategic choices, Robin Li also clearly stated that it is "application-driven."

"This is very different from the practices of many American companies. Many American companies have always dreamed of creating disruptive technologies and envisioned what they will achieve in a certain number of years. If an application can be produced, it seems to be a side product," said Robin. "I prefer to come into contact with scenarios and applications as soon as possible to see what problems arise in the process. The most common problems encountered by everyone are the ones we prioritize solving."

Whether it's electrical technology or information technology, so-called "disruptive" technologies are often those that can be widely used, improve efficiency, and bring about commercial changes. The prerequisite is that they must first be made user-friendly.

The same is true for large models.

The three things Baidu has been doing these years are: making large models "better to use, easier to use, and more useful."

To make large models "better to use," on the one hand, it is necessary to continuously invest in researching and developing basic large models and continuously iterate on large model capabilities.

At the financial report conference call, Baidu stated that it will continue to iterate on the capabilities of the Wenxin large model and optimize model efficiency. In addition, Robin Li also revealed that Baidu will release a new version of the Wenxin large model in early 2025 to consolidate its leading position in basic models.

In addition to upgrading models, it is also necessary to solve technical difficulties in applications. For example, the iRAG technology recently released at Baidu World 2024 greatly solves the hallucination problem of text-to-image generation and enhances image productivity.

To make large models easier to use, Baidu released the no-code tool "Miaoda," allowing users to develop applications without writing a single line of code.

In addition, Baidu has further constructed an intelligent agent ecosystem, using intelligent agents as the cornerstone and combining AI functions to reconstruct Baidu Search; Baidu Wenku has been reconstructed into a productivity tool.

In terms of being "more useful," Baidu's construction of a new ecosystem of intelligent agents + search means that a large number of users will form new habits.

After all, users are willing to use large models to form usage habits, the number of users can increase, and the commercialization of large models is expected to take a step further, forming a positive cycle of industry growth.

In the past, the core action for people to obtain information and solve problems online was searching, which is a habit for most people. In the future, it will be using intelligent agents and the capabilities of large models to solve problems.

Baidu has hundreds of millions of users, and the transition in user habits is very smooth, which means that more users may be able to use and make good use of large models in the future. This is of far-reaching significance for the entire AI large model industry.

In my opinion, the significance of inventing intelligent agents in the AI era is no less than that of inventing pinyin input methods. When did the Internet have a large number of Chinese users? It was when pinyin input methods became popular.

The same is true in the AI era.

When the user volume and user habits of large model products truly explode, the application of AI large models can announce that the old era has ended, and the door to a new era has opened.

Reconstructing the Fundamentals: What is the Value of the New Ecosystem of Intelligent Agents + Search?

Regarding the future development of the industry, I have this judgment: After the AI large model opens up a new era, the fundamentals of all large model companies will be reconstructed. Especially for companies like Baidu that are steadily running at the forefront of the industry.

For Baidu, the value of reconstructing the fundamentals lies in two levels:

1. On the business side, the new ecosystem of intelligent agents + search reconstructs Baidu's fundamentals.

First, the entry value will be reconstructed.

At Baidu World 2024, Robin Li stated: "Intelligent agents are the most mainstream form of AI applications and are about to reach their tipping point." He believes that corporate intelligent agents will eventually replace traditional official websites and become the first interface for direct consumer interaction.

In fact, Baidu has already built the corporate intelligent agent for BYD - BYD's official intelligent agent, which has been launched. It was mentioned at the financial report conference call that after the launch of BYD's official website intelligent agent, its sales lead conversion rate increased by 119%.

This growth in sales leads actually indicates one thing: Baidu's intelligent agents + search are the primary entry points for AI information and AI services.

The value of entry points actually speaks for itself. Search in the PC era and apps in the mobile era are both very valuable entry points.

Entry points were centralized in the PC era, concentrated in search engines, and decentralized in the mobile era, scattered across different app platforms such as social media, e-commerce, and content.

In the era of AI large models, entry points are both centralized and decentralized.

From the user's perspective, if a large model is easy to use, then the user base will grow larger and larger, just like operating systems in the mobile era, where the winner takes all and the stronger becomes even stronger - this is centralization.

From an experience perspective, different functions of large models correspond to different user needs, so entry points are also decentralized.

Baidu's new ecosystem of intelligent agents + search precisely meets such user needs. This may also be why Baidu has been emphasizing that "intelligent agents are the best form of AI applications."

Second, the value of services will be reconstructed.

In the past, Baidu connected services through "search + mini-programs + hosted pages." In the era of large models, it has become "intelligent agents + search."

According to the Q3 financial report conference call, intelligent agents are an important form of enriching Baidu Search content and experience. The AI functions of search can bring rich content such as AI summaries, images, videos, intelligent agents, and digital humans, enhancing the user experience.

Just as Robin Li demonstrated at Baidu World 2024, not only can AI generate content, but searching for company websites and inquiring about industry advice may also activate intelligent agents. In other words, intelligent agents are a "new medium" to provide users with more services.

For users, the large model capabilities carried by intelligent agents mean a better experience. For merchants, having large model capabilities also allows them to better provide services from various industries to users.

Finally, reconstructing entry points and services with large models also means reconstructing brand business growth models.

Having a new ecosystem of intelligent agents + search means that brand businesses have the opportunity to find a new growth model.

Based on past experience, the emergence of mobile apps, mini-programs, and other applications has changed the way people interact with information, so brand businesses have seen more growth.

This is just like how mobile apps and mini-programs brought business growth opportunities in the past, and intelligent agents can also bring new opportunities. Currently, many enterprises have realized this. After the debut of corporate intelligent agents at Baidu World 2024, 582 enterprises reached cooperation on the first day.

2. Changes in fundamentals mean changes in valuation logic.

On the valuation side, the return of the valuation logic of classical technology companies may become a new trend.

There are roughly two paths for technology giants to change their valuation logic:

One is investment, where giants become investment banks and do not put all their eggs in one basket. Many internet giants do this, so the valuation of many technology companies in the past has used the STOP valuation method.

"Investment banking"-style technology companies are not difficult to do. I think setting up a strategic investment department, having a strategic mindset, and being financially capable can quickly scale up the business.

However, the technology industry has reached this stage, and this path is no longer viable. The valuation of "investment banking"-style technology companies is gradually approaching its upper limit. Therefore, everyone has gradually returned to the path of product development.

The other is the truly difficult "hard technology," which is pursuing a long-term strategy.

The difficulty with this route is the ability to both look up at the stars and keep one's feet on the ground.

This is actually also the valuation logic of "classical" technology companies. Focus on a disruptive technology, apply it, and commercialize it. Technology enterprises with such a mindset have more imagination space at present.

For example, just like Baidu, which has always regarded artificial intelligence as a long-term strategy for 10 to 20 years. Moreover, the most important thing is to be able to produce phased results.

If you choose this path, it requires the enterprise to not miss any new technology that may lead to change, and the core of ultimately succeeding lies in running through the closed loop of technological change, application iteration, and commercial landing.

If we view Baidu from this perspective, we may gain new insights.

Search represents Baidu's old business and is the "foundation." The large model represents the future and the "explosive power" for the next 10 to 20 years. The data growth in Q3 indicates that if this future prospect is clear enough, the explosive power is sufficient, and there can always be phased results, then the large model business can consolidate the foundation of search.

Conversely, the search business, nourished by large model technology, can also continuously provide funding, data, and technical support for the iteration of the large model business.

At this point, the new and old businesses are seamlessly integrated and mutually reinforce each other.

Mutual feeding and promotion of growth between new and old businesses is the ideal business model for "hard technology" enterprises. However, the market valuation logic has not yet fully shifted, and future growth potential may not be fully reflected in corporate valuations.

The transformation brought about by Baidu's AI strategy is not only a change in business and valuation logic but also a change in the iterative evolution of technology enterprises.

If I had to define Baidu's phased growth, I would say it is the "evolutionary power" brought by AI large models.

Especially in the past two years, many enterprises have been in a period of confusion, unsure how to explore new businesses or how to transform to adapt to the new environment.

I think it's a good idea to learn from Baidu at the right time.

As a successful case of transformation, Baidu's transformation is actually quite meaningful for the economic development and transformation of the entire society. In the future, I hope that more innovative Chinese enterprises can follow Baidu's path and find their own successful transformation and upgrading route.

Disclaimer: This article provides comments based on the company's statutory disclosure content and publicly available information, but the author does not guarantee the completeness or timeliness of this information. Additionally, the stock market is risky, and investors should be cautious. This article does not constitute investment advice, and investors must make their own judgments.