From Q2 financial report, the latest Baidu-SW: Leading domestically in large-scale model implementation, accelerating growth in cloud business

As a company that has been deeply involved in the AI industry for over ten years and is the first globally to launch a large model directly comparable to ChatGPT, Baidu-SW (09888) has been continuously advancing the commercialization of large models in a pioneering manner since the launch of "Wenxin Yiyuan". It has not only vigorously promoted the empowerment of its own business with large models, but also actively facilitated the application of large models in a wide range of industries and enterprises. Currently, Baidu is leading the industry in terms of the speed, breadth, and depth of large model implementation, as evidenced in its latest financial report.

On August 22nd, Baidu released its financial report for the second quarter of 2024. The data shows that Baidu's revenue for the reporting period was 33.9 billion yuan, with a core operating profit of 5.6 billion yuan, a year-on-year increase of 23%, exceeding market expectations. The financial report indicates that intelligent cloud continued to accelerate its growth in the second quarter.

Behind the solid performance is the significant vitality displayed by Baidu's entire product line after being restructured and empowered by large models, accelerating the value of the company's entire ecosystem as the commercialization of large models continues to speed up. This means that AI has become one of the core driving forces behind Baidu's performance growth.

TF Securities stated that Baidu is actively promoting the application of large models in different scenarios, fully tapping into the potential of large language models, bringing incremental demand by promoting application needs on the enterprise side. The progress in applications can drive innovation in basic models, complement each other to promote model iteration, and is expected to further develop in the future.

A deep analysis of Baidu's financial report reveals that Baidu, which is accelerating the commercialization of large models, is about to undergo a value leap.

Wenxin large model daily call volume exceeds 600 million, Baidu's large model application implementation is the fastest

As one of the members of the Wenxin large model family, Wenxin Yiyuan has set a new record. Since being allowed to provide services to the public on August 31, 2023, the number of Wenxin Yiyuan users exceeded 100 million in nearly 4 months. It took about 4 months to grow from 100 million to 200 million users, but only 2 and a half months to grow from 200 million to 300 million users, showing a significant acceleration in user growth.

At the same time, the daily call index that best reflects the extent of Wenxin large model application is also showing explosive growth. According to disclosures during the post-financial report conference call, as of now, the daily call volume of the Wenxin large model has exceeded 600 million times. Compared to the 50 million daily average call volume announced in Q4 of 2023, the volume has grown more than 10 times in half a year. In addition, the Wenxin large model processes over 1 trillion tokens of text on a daily basis, ranking first domestically, reflecting that Baidu's implementation of large model applications is the fastest in the country.

Behind the industry-leading commercialization of large models lies the fact that Baidu has done two things right, namely seizing the two key elements of good performance and low cost. With the wave of generative AI sweeping the globe, the demand for enterprises to achieve intelligent transformation through large models to enhance core competitiveness is relatively rigid. However, whether large models can solve pain points in enterprise operations and whether the costs are affordable have become the most important considerations for enterprises when procuring large model services.

In terms of performance, as a pioneer in the domestic large model industry, Baidu has leveraged the unique advantages of its four-layer AI technology stack to form an efficient feedback loop between chips, frameworks, large models, and terminal application scenarios, continuously upgrading and iterating the capabilities of the Wenxin large model After upgrading the Wenxin large model to version 4.0 in October 2023, Wenxin has reached the industry-leading level in understanding, generation, logic, and memory capabilities, and has achieved further significant improvements in general abilities such as code generation, code interpretation, and code optimization, reaching the international leading level, and is no less competitive than GPT-4 in terms of overall performance.

The second-quarter financial report shows that in June of this year, Baidu's large model family introduced a new product, the Wenxin large model 4.0 Turbo, and opened it up to enterprise users. Through core technological innovations such as data and basic model optimization, in collaboration with PaddlePaddle, compared to the Wenxin large model 4.0, the Wenxin large model 4.0 Turbo is faster and more effective, enabling machines to think and act like humans, autonomously complete complex tasks using tools, and continuously learn and evolve in the environment.

As Baidu expands its model portfolio and continues to improve model performance, the "price reduction" expansion strategy has become a "booster" for the accelerated penetration of the Wenxin large model in the market. In May of this year, Baidu announced the free use of the ERNIE-Speed, ERNIE-Lite, and ERNIE-Tiny series of models for customers; in July, Baidu significantly reduced the prices of the flagship models ERNIE 4.0 and ERNIE 3.5, further helping enterprises reduce the costs and trial-and-error costs of large models.

Behind Baidu's "price reduction" expansion strategy is the rapid decrease in the cost of the Wenxin large model. During the financial report conference call, Robin Li stated that in the second quarter, Baidu upgraded its open-source deep learning framework PaddlePaddle to version 3.0, significantly improving the compatibility of the PaddlePaddle framework with Baidu's AI infrastructure and the Wenxin large model, which will further reduce model inference costs in the future.

It is the significant cost reduction that has laid a solid foundation for Baidu to achieve large-scale implementation of large models. And due to the clear network effects of large models (each additional user creates new value for other users), the larger the user base, the more training materials and iterations, and the faster the evolution towards "general artificial intelligence," large models will play a significant positive role between scale and performance. This means that the performance improvement of the Wenxin large model, which already has the highest daily call volume in the industry, may also be the fastest.

From a final perspective, based on the significant network effects of large models, there will inevitably be a Matthew effect, where the strong get stronger, and market concentration will accelerate towards leading companies, so in the industrial wave of domestic large models empowering enterprise intelligence transformation, the current Baidu may have already "reserved" the largest market share in the future.

AI-Driven Cloud Services Maintain High Growth

The accelerated commercialization of large models has injected new vitality into the development of Baidu's intelligent cloud services. According to the financial report, intelligent cloud services continued to accelerate growth in the second quarter and will maintain strong growth momentum.

During the reporting period, Baidu's intelligent cloud business achieved continuous rapid growth driven by AI, with revenue increasing by 14% year-on-year to 5.1 billion RMB, of which the revenue brought by AI accounted for 9%, up from 6.9% in the first quarter, and continued to be profitable (Non-GAAP) From an industry perspective, with the advent of the AI-native era, the game rules of the cloud computing market have fundamentally changed. To meet the new market demands, many disruptive changes have occurred in cloud computing services. On one hand, the explosion of demand for GPUs is driven by large model applications, while on the other hand, Model as a Service (MaaS) has become a new foundational service.

In order to lower the threshold for deploying large models, Baidu has launched the Mass platform and Qianfan Large Model Platform, providing a complete set of tools including globally mainstream models like Wenxin Large Model, as well as tools for large model development and application.

According to a conference call, in the second quarter, Baidu continued to upgrade the model toolchain on the Qianfan platform. The model customization tool ModelBuilder has introduced DPO and KTO model training methods, significantly improving model fine-tuning efficiency. Additionally, two new model quantization compression algorithms have been added to reduce model inference costs. The continuous improvement of the model toolchain will help enterprises fine-tune more stable and effective dedicated large models. In terms of AI-native application development, Baidu recently made a comprehensive upgrade to the enterprise-level retrieval enhancement generation (RAG) of AppBuilder in terms of scalability, openness, and security. The continuous upgrades have led to a significant increase in usage. The conference call indicated that over tens of thousands of applications have been created on AppBuilder, covering industries such as online education, e-commerce, and government affairs.

Thanks to Baidu's comprehensive layout around intelligent computing infrastructure, MaaS services, and AI-native application development, the Qianfan platform has now served a cumulative total of 150,000 corporate users, helping users fine-tune 21,000 large models.

Furthermore, the continuous success in bidding for large models not only reflects the market's recognition of Baidu's strength in large models but also contributes to the continuous growth of Baidu's intelligent cloud services. According to statistics, from January to May 2024, nearly 260 domestic large model project bidding announcements have been released, with Baidu leading in the number of successful bids among mainstream large model vendors (10 projects), covering the most industries (8 industries), and having the highest total bid amount (over 64 million RMB). The purchasing units for these projects include the All-China Federation of Trade Unions and the Southern Power Grid, among other top state-owned enterprises. This demonstrates that Baidu's large models are accelerating the transformation and upgrading of traditional industries, speeding up industrial landing, and the development of Baidu Cloud services is expected to accelerate accordingly.

It is worth mentioning that in a recent IDC report on the "China Large Model Platform Market Share, 2023: The First Year of Large Models - The Beginning," Baidu Intelligent Cloud ranked first in the Chinese large model platform market share in 2023, reaching 19.9%. This undoubtedly proves Baidu's solid capabilities in deploying large models.

Under AI reconstruction, the search business landscape is solid, and intelligent entities are expected to accelerate their outbreak

While cloud services continue to experience high growth, Baidu's core business - Baidu Search - has ushered in new development opportunities under the AI reconstruction.

Since the outbreak of large models in 2023, search engine companies worldwide have been exploring the application of large model capabilities to search in order to enhance search capabilities. Apart from Baidu, global tech companies such as Google, Microsoft, Meta, and others are actively entering the AI search field The reason why it has become a "must-see place for the military" is because search is one of the main ways for the public to obtain information, an important channel for human-machine interaction, and a huge traffic entrance with high frequency of use. It can be said that whoever can find an effective way to empower search with AI can reshape the search landscape.

As a leader in the domestic search field, Baidu has been deeply cultivating the search industry for many years and is also the earliest company in China to explore intelligent search. After the launch of Wenxin Yiyant, Baidu quickly completed the reconstruction of its search business with AI large models. The new search product has three main features: extreme satisfaction, recommendation stimulation, and multi-round interaction, which not only improves the quality of search, but also expands the application scope of search and broadens the boundaries of the product.

Currently, the "AI content" of Baidu search has significantly increased. According to the conference call, during this reporting period, the proportion of search results generated by AI at Baidu was 18%, a significant increase from 11% in the first quarter.

Moreover, the restructured AI search has been widely recognized by the market and users. On August 8th, one of the highest-rated AI application lists in China, the "AI Product List," released the latest data. Baidu's search "AI Smart Answers" topped the domestic overall list with absolute advantages in its first selection. This means that Baidu's AI-empowered search has consolidated its leading position in the search field, and the company's foundation remains stable.

In addition, the release of intelligent agents can be seen as an important "lever" for Baidu to improve its search business. The main purpose of Baidu's intelligent agents is to accelerate the release of the potential of large models and allow intelligent agents to quickly integrate into various application scenarios. From the basic characteristics of intelligent agents, this approach has a high probability of success. This is not only because intelligent agents significantly improve the search experience for C-end users and enhance the advertising efficiency for B-end customers, effectively meeting market demand, but also because the threshold for creating intelligent agents is low enough that even without programming skills, a decent intelligent agent can be created. This simplicity and ease of use lay a solid foundation for the rapid popularization of intelligent agents.

Previously, Li Yanhong stated that basic models need applications to demonstrate their value, and in the development direction of AI applications, he is most optimistic about intelligent agents. Intelligent agents are an almost universally applicable application based on large models, representing the future trend of the AI era. Currently, intelligent agents are in the stage of eruption, but due to the relatively small base, the public's perception is not yet very strong. Once intelligent agents are up and running, they will become a Super APP.

Baidu has seized the opportunity of the eruption of intelligent agents and is expected to accelerate their development. Baidu's search has a daily demand of tens of billions of times, based on the three major advantages of a huge user base, super traffic, and intelligent recommendations, Baidu search has become the best scenario for the landing and distribution of intelligent agents. The financial report shows that search is becoming the largest entrance for the distribution of intelligent agents. Intelligent agents built on the Wenxin intelligent agent platform AgentBuilder can be distributed within the Baidu search ecosystem to enrich content ecology and improve customer efficiency.

The financial report conference call shows that the daily average distribution of Baidu intelligent agents in July has exceeded 8 million times, which is twice that of May, indicating that Baidu intelligent agents are gradually becoming ecological. In addition, Baidu Wenku is one of the most successful products in large-scale model reconstruction. Baidu Wenku's market share in the intelligent PPT field has reached 80%, leading by a significant margin. Out of every 5 intelligent PPT presentations in the market, 4 are from Baidu Wenku.

In the latest "AI Product Ranking" list released on July 7th, Baidu Wenku's AI features once again topped the domestic overall list, achieving a "three-peat". Currently, Baidu Wenku has accumulated over 180 million AI users, with AI feature usage exceeding 2.2 billion times.

Summary

Looking through the financial reports, it is evident that after over a year of layout, the full-scale reconstruction of Baidu's business by large-scale models has begun to release clear value. AI has become the core driving force behind Baidu's performance growth. Whether it is the rapid growth of intelligent cloud services or the continued stability of the search landscape, they are closely related to the empowerment and reconstruction brought by large-scale models.

Moreover, the explosive growth of the daily call volume index of Wenxin's large-scale model indicates the continuous acceleration of the landing speed of Baidu's large-scale models. With the continuous improvement of Baidu's large-scale model performance, it will accelerate the empowerment of Baidu's entire business. The flywheel effect is already accelerating within Baidu's ecosystem, which will inevitably drive the revaluation of Baidu's intrinsic value.

There are many brokerage institutions optimistic about Baidu's future development. Tianfeng Securities, for example, expects Baidu's revenue for 2024 to 2026 to be 138.6/152.9/170 billion yuan; Non-GAAP operating profit to be 28/31.9/35.4 billion yuan. Based on SOTP valuation, Tianfeng Securities maintains a "buy" rating on Baidu with a target price of HKD 140. Additionally, GF Securities maintains a "buy" rating on Baidu with a target price of up to HKD 165.

Taking a longer-term perspective, as a leader in the domestic AI field, Baidu is an unavoidable target for AI industry investment. It already has the potential to become a giant in the AI era, and its value will eventually be reflected in its stock price as the business continues to advance and develop. Therefore, now may be a good time to strategically invest in Baidu for the long term