Baidu (Minutes): AI native marketing is the future second curve
The following is a summary of the $Baidu(BIDU.US) FY25Q3 earnings call minutes organized by Dolphin Research. For earnings interpretation, please refer to " Baidu, Can Telling AI Stories Really Save Itself?"
I. Review of Core Financial Information
1. AI Revenue Contribution Reaches 40%: In the third quarter, revenue directly related to AI reached nearly 10 billion, accounting for 40% of total revenue. Among them:
(1) AI Cloud Infrastructure (including large model APIs, computing power leasing) accounted for half, with a year-on-year growth rate of 33%. Since the AI part has been singled out, this growth rate is not particularly impressive. The significant change was in computing power leasing, with a growth rate of 128% this quarter, accelerating significantly compared to 50% in the previous quarter.
(2) AI Applications reached 2.6 billion, with a year-on-year growth rate of 6%, mainly including Baidu Wenku, Baidu Netdisk, digital employees, etc. Although also under AI+, the growth rate is low, which Dolphin Research estimates is mainly due to the drag of Baidu Netdisk (promotional discounts this quarter) and other miscellaneous items.
(3) AI Native Marketing Services, mainly referring to Agents and digital humans, achieved 2.8 billion in the third quarter, with a growth rate of 262%, indicating a period of strong growth and greater application potential.
2. User Loss and Advertising Pressure Remain Significant: Overall marketing revenue in Q3 declined by 18%, slightly better than guidance. However, excluding the AI part, the implied traditional advertising segment declined by 27% year-on-year. The company mainly attributed this to the impact of AI-generated content, revealing that 70% of mobile search results in the third quarter contained AI-generated content, reaching a stage of penetration.
However, the nearly 30% decline may still be partly due to competition, i.e., traffic migration to peer ecosystems (search within other applications, native AI applications), such as Tencent's advertising growth rate of 21% in the third quarter, which mentioned the significant contribution of WeChat Search.
From user numbers, the problem is also evident, with a net decrease of 27 million MAU for Baidu App in the third quarter, although slightly better than the same period last year, the positive effect of improved product experience through AI search integration since the beginning of the year is weakening, indirectly indicating some signs of traffic migration.
3. Apollo Go is "Running Fast" Internationally: In other income, besides the cloud, smart driving is a highlight. In Q3, Apollo Go's total order volume reached 3.1 million, with a growth rate of 212%, significantly accelerating compared to Q2. Besides penetration in 22 cities in mainland China, Apollo Go has been advancing in international markets such as Hong Kong, Dubai, and Switzerland since the beginning of the year.
4. Core Profit Margin Pressure Released: Baidu's core business operating profit, i.e., gross profit minus three operating expenses, fell to 1.2 billion in Q3, with a profit margin of only 5%, far below normal levels. This is related to several factors: 1) Business structure changes. AI business revenue has increased, but the profit margin is very low, thus dragging down the overall level. 2) Q3 marketing expenses increased significantly, possibly related to the promotion of cloud and AI applications, including the establishment of sales teams and promotional expenses.
Additionally, a one-time provision of 16.2 billion was made this quarter, mainly for the impairment of some traditional equipment. This provision does not affect cash flow but will optimize future costs, representing an early release of pressure.
5. Slight Expansion in Capital Expenditure: Baidu's Capex is significantly lower than peers, partly due to years of AI investment accumulation and partly because self-developed chips are cheaper than external procurement. However, the cost demand for AI is significantly higher, and Baidu's Capex has already restarted expansion since last quarter.
In the third quarter, Baidu's capital expenditure was 3.4 billion, with a year-on-year growth rate of 106%, which can be linked to the aforementioned traditional equipment elimination. As old equipment is phased out, new server-related expenditures also need to increase. Additionally, considering Baidu's Kunlun chip business, the expansion in capital expenditure may not be for external GPU procurement but possibly related to supporting facilities for Kunlun chips.
6. Increased Shareholder Returns on the Way?: Although it is difficult to predict when the advertising business will improve, the market has always had mixed feelings but also expectations for the large cash distribution on Baidu's books (as of the end of Q3, Baidu's core net cash was 103.5 billion, equivalent to 14.8 billion USD). After all, the current return rate is not high, and the Q3 buyback may be better than nothing, but it was not disclosed in the financial report. According to the previous buyback pace, a 2% shareholder return yield could not support the stock price during its previous rapid decline. In the earnings call, the company revealed that it is discussing a dividend plan and setting a "floor level" for the buyback plan to reduce seasonal fluctuations. If a good return plan with a decent yield can be implemented, it would help boost the stock price in the short term and provide bottom support.
7. Detailed Financial Data Overview
II. Detailed Content of the Earnings Call
2.1 Executive Statements on Core Information
1. AI Cloud Business: Full-Stack Architecture Drives Enterprise Intelligence
a. Growth Driver: Subscription revenue for AI accelerator infrastructure surged 128% year-on-year, being the main driver, indicating a healthier and more sustainable revenue model.
b. Full-Stack Advantage: Provides comprehensive support from infrastructure (high performance, low cost), model layer (ERNIE 5.0 full-modal model) to application layer.
c. Application Cases:
- FM Agent: Self-evolving AI agent applied in complex industries such as transportation, energy, and logistics to optimize operational efficiency.
- Digital Employees: Like "Daniel Wu English Coach," providing 24/7 personalized service, solving scalability challenges.
d. Customers and Ecosystem: Optimized customer structure, with spending from large and medium customers both growing.
- Upgraded MaaS platform to "agent-centric," enabling enterprises to quickly develop AI-native applications.
- Internally using AI coding assistants, with over 50% of new code generated by AI, significantly improving R&D efficiency.
2. Intelligent Driving (Apollo Go): Global Scale Expansion
a. Operational Scale: In October, the weekly fully autonomous driving operational mileage exceeded 250,000, reaching a global industry-leading level.
b. Safety Record: Cumulative autonomous driving mileage exceeded 240 million kilometers, with fully autonomous driving over 140 million kilometers, showing an excellent safety record.
c. Global Expansion: Increased coverage to 22 cities (16 cities last quarter).
- Europe: Entered Switzerland, cooperating with local public transport operators.
- Middle East: Obtained full autonomous commercial operation license in Abu Dhabi; exclusive authorization for open road testing in Dubai.
- Hong Kong: Expanded to Kowloon and Kwun Tong districts, consolidating right-hand drive market leadership.
d. Business Model Innovation: Collaborated with Shenzhou Car Rental, Hello Ride, etc., to achieve cross-city autonomous car rental and large-scale operation with higher capital efficiency.
3. Mobile Ecosystem: AI Reshapes Search and Monetization
a. AI-Driven Search: About 70% of mobile search result pages contain AI-generated content, making content richer and more relevant.
b. User Metrics: Baidu App's monthly active users reached 708 million, with daily user time increasing by 2.3% year-on-year.
c. Technology Output: Exported search capabilities to external partners like Samsung, Xiaomi, and Honor through Baidu AI Search API.
d. AI Native Applications: Baidu Wenku & Baidu Netdisk: Combined monthly active users nearly 300 million, launched a general AI agent platform to complete complex tasks through natural language.
e. Marketing Innovation (Agents & Digital Humans):
- Agents: Precisely identify user intentions through multi-round dialogue, providing high-quality sales leads for advertisers. Daily, 33,000 advertisers use it.
- Digital Humans: Provide low-cost, 24/7 AI live streaming, expanding application range from e-commerce to medical, automotive, legal, etc. The number of live digital humans in September nearly doubled year-on-year.
4. Future Outlook and Strategic Focus: Continue to expand AI-driven revenue streams such as AI Cloud, AI native applications, and autonomous driving. Innovatively promote AI native monetization methods like agents and digital humans to more vertical industries and existing customers.
Enter more global markets with commercial potential while ensuring safety and operational excellence. Relying on full-stack AI capabilities, continue to innovate and seize long-term opportunities to bring greater transformative value to the entire business portfolio.
2.2 Q&A Session
Q: How does ERNIE 5.0 drive application innovation (e.g., digital humans)? What are its future iteration focus and core differentiation advantages?
A: We always adhere to application-driven iterations of the ERNIE model. The ERNIE 5.0 released at the Baidu World Conference has reached world-class levels in cross-modal understanding, creative writing, and instruction following.
Taking digital humans as an example, ERNIE not only endows them with smooth and precise language expression capabilities but also drives real-time synchronization of visual images, actions, and micro-expressions, creating a truly convincing user experience. At the same time, ERNIE also empowers our self-evolving intelligent agent FM Agent, significantly improving efficiency in industries such as manufacturing, energy, and finance.
Although achievements have been made, we see greater room for improvement: we expect digital humans to surpass real human sales performance in live e-commerce; hope FM Agent can handle more complex industrial scenarios; pursue AIGC content to match user interests more accurately than KOL creation. These are the key challenges that existing models have not yet solved.
We firmly believe that the ultimate value of AI will be reflected at the application layer. Although the current industry value is still concentrated at the infrastructure layer, in a healthy ecosystem, the greatest value should come from application products that can create real impact for users, advertisers, and enterprises.
Baidu will continue to focus on making ERNIE the strongest in the most critical scenarios of our business ecosystem and continue to invest in technology fields that can create measurable value. Staying close to applications is the fundamental path for sustainable AI development.
Q: How do you view the slowdown in cloud growth in the third quarter and future demand trends? Can growth re-accelerate next year? What are the core drivers of its sustainability?
A: Looking at the performance throughout the year, our cloud business growth rate continues to surpass the industry level. Although there may be fluctuations between quarters, the overall growth trend is strong, and we are confident in the future growth trajectory. Currently, enterprises are fully integrating AI into operations, bringing broad and strong demand for AI cloud services.
In the AI cloud business, areas closely related to AI workloads have shown the most significant growth. In the third quarter, AI cloud infrastructure revenue reached 4.2 billion, with a year-on-year growth of 33%, and subscription-based AI accelerator revenue surged 128% year-on-year, reflecting both strong AI demand and continuous optimization of the revenue structure.
This growth momentum is driven by our full-stack AI capability layout: at the infrastructure layer, our self-developed AI computing architecture is widely adopted for its excellent performance and cost-effectiveness; at the platform layer, the Qianfan MaaS platform has completed an upgrade centered on agents, significantly enhancing support for complex workflows; at the application layer, solutions like digital employees and FM agents are rapidly expanding, creating long-term value for the cloud business.
Based on the current development trend, we believe that full-stack AI capabilities and the continuous demand for AI cloud services will jointly drive the business to achieve healthy, scalable, and sustainable growth.
Q: What user behavior changes have been observed with the integration of search and chatbots? What competitive strategies will Baidu adopt?
A: Facing the trend of increasingly blurred boundaries between search and AI assistants, we have observed significant changes in user behavior: the conversation volume of the ERNIE assistant within the Baidu App has increased approximately fivefold year-on-year, with users expecting seamless transition from information query to task execution, and a surge in demand for multimodal interaction.
Our competitive strategy focuses on a dual-track layout: on one hand, deeply integrating ERNIE into Baidu Search to provide personalized intelligent Q&A through contextual understanding; on the other hand, exploring cutting-edge interaction forms through independent ERNIE applications. We are integrating innovative technologies such as real-time generative digital humans into the product matrix, promoting the evolution of interaction experience towards multimodal and real-time, continuously building the core advantage of the "search + agent" integrated ecosystem.
Q: How does management view the growth prospects and profitability of new AI businesses (AI Cloud, AI Applications, AI Native Marketing)? How will they accelerate the company's overall revenue growth?
A: We have launched an AI native business view to clearly showcase high-growth AI assets. This business system includes three core segments:
AI Cloud Infrastructure occupies an industry-leading position with self-developed architecture and MaaS platform. As AI applications become more widespread, we expect its utilization rate improvement to drive profit margin enhancement.
AI Application Products (such as Baidu Wenku and Netdisk) form high-quality revenue sources through subscription models, and their extensive product portfolio continues to create added value. AI Native Marketing Services (agents and digital humans) have become the second growth curve, with this quarter's revenue share reaching 18% of the core online marketing business, up from 4% a year ago, as customers are willing to pay for tangible efficiency improvements.
These AI businesses form a synergistic effect within the ecosystem, and as penetration continues to increase, we are confident in their revenue and profitability potential, expecting them to jointly drive Baidu into a stronger growth trajectory.
Q: What are Apollo Go's future growth and expansion plans? How does its economic model perform in different markets? What is its long-term profitability outlook?
A: After thirteen years of development, Apollo Go has become a global leader in Robotaxi services, with cumulative order volume exceeding 17 million, and order volume in the Chinese market reaching more than 15 times that of major competitors, all achieving fully autonomous driving operations.
We believe the industry has reached a critical point of explosion. Looking towards 2026 and beyond, we will continue to break through from three dimensions:
1. Operational Scale: Increase capacity in existing cities, expand to new cities, and continuously optimize technology through massive data;
2. Cost Optimization: Promote continuous reduction of per-mile costs through technological innovation and operational efficiency improvement, expecting more cities to achieve single-city profitability by 2026;
3. Model Innovation: With flexible models such as light assets, we are ready to quickly enter new markets.
Currently, we have covered 22 cities worldwide, and as scale effects become apparent, we are confident in achieving global expansion and long-term profitability.
Q: How is the progress of AI search monetization and market feedback? Can AI native marketing services replace traditional advertising? What is the future profitability outlook for the advertising business?
A: Currently, nearly 70% of Baidu's mobile search result pages are equipped with AI-generated content, with multimodal content being our unique advantage. This quarter, we focused on improving content quality, especially rich media forms such as images and videos, significantly enhancing user interaction depth: users exposed to AI search results increased their query volume by 6%, and dwell time continued to extend.
In terms of commercialization, we have made initial progress: the e-commerce module in AI search set a peak GMV of nearly 6 million yuan during the Double Eleven period; agent advertising achieved daily revenue of over 25 million yuan; innovative forms such as digital human live streaming are also being continuously explored. Although short-term optimization of user experience remains a priority, we have already seen significant monetization potential in AI search.
We are clearly aware that this AI transformation, while putting pressure on short-term revenue and profit, is an inevitable choice for building long-term competitiveness. We are laying a solid foundation for future growth by balancing short-term investment with long-term opportunities.
Q: What are the reasons for asset impairment this quarter? What are the capital expenditure plans for next year and the profit margin outlook under AI growth? What is the shareholder return strategy after the current buyback expires?
A: Regarding asset impairment, we proactively impaired existing infrastructure that does not meet current efficiency requirements due to accelerated investment in next-generation AI computing technology, making the asset portfolio healthier and better able to meet future high-end AI computing needs.
In terms of capital expenditure, we continue to increase AI investment (cumulatively over 100 billion yuan since March 2023), but improve efficiency through measures such as optimizing the asset base, enhancing resource utilization, and dynamically allocating capacity. We judge that the third quarter is the low point of profit margin, and expect non-GAAP operating profit and profit margin to gradually improve next year.
In terms of shareholder returns, according to our 2023 stock buyback plan, we have repurchased $2.3 billion worth of stock. We are currently reviewing the renewal of this plan, with the new plan aiming to reduce seasonal fluctuations by providing a clearer buyback mechanism and setting a floor for buyback yield to enhance shareholder returns. Additionally, we are actively exploring opportunities for an initial dividend policy. Overall, we expect stock buybacks and potential dividends to bring considerable yield to investors.
<End Here>
Risk Disclosure and Statement of This Article:Dolphin Research Disclaimer and General Disclosure
