Baidu (Minutes): AI search commercialization has begun early testing

The following are the minutes of the Q2 2025 earnings call for$Baidu(BIDU.US) compiled by Dolphin Research. For financial report interpretation, please refer to 《Baidu's 'Breaking the Cauldrons and Sinking the Boats': Can It Lead to a Rebirth?》

I. Review of Core Financial Information:

1. Capital Expenditure and Cash Flow:

During the reporting period, the company's free cash flow was negative RMB 4.7 billion, excluding iQIYI's free cash flow, which was also negative, mainly due to increased investment in AI business. As of June 30, 2025, Baidu held cash, cash equivalents, restricted cash, and short-term investments totaling RMB 124.2 billion, with a net cash position of RMB 155.1 billion.

2. Personnel Adjustment Plan: During the reporting period, the company's operating expenses decreased by 4% year-on-year, primarily due to reduced personnel-related expenses. As of June 30, 2025, Baidu's core business had approximately 31,000 employees.

3. Guidance:

The company has not yet reached the stage of large-scale commercialization, but has initiated cautious small-scale commercialization tests during the reporting period, achieving satisfactory early results, showing potential in previously difficult-to-commercialize queries. Looking ahead, Baidu will continue to accelerate the development of foundational models and strategically focus on areas with practical application value to maintain core competitiveness. The company is confident in further accelerating global expansion and capturing significant value in multiple global markets.

II. Detailed Content of the Earnings Call

2.1, Executive Statements of Core Information:

1. Business Progress

1) Core AI Technology:

- The company optimized its self-developed four-layer AI infrastructure, significantly improving inference concurrency and resource utilization through the large-scale stable deployment of the Prefill Decode separation architecture, and greatly reducing inference costs.

- In terms of the model library, a series of models, including the newly open-sourced ERNIE 4.5 series, were added, enhancing the flexibility of enterprise use cases.

- The ERNIE model achieved new breakthroughs in digital human technology applications, with realism and capability levels in specific scenarios even surpassing human performance. For example, a dual digital human live broadcast driven by ERNIE 4.5 turbo generated tens of millions of GMV within 7 hours. Digital human technology has been widely applied in industries such as healthcare, legal services, education, and automotive.

- The development of agent functions further progressed, enabling multi-turn dialogues with users, connecting to relevant service providers, and facilitating end-to-end task completion, expanding search capabilities from information provision to actual action.

2) Search and Mobile Ecosystem:

- The transformation of Baidu Search by AI continues to accelerate. As of the end of June, AI-generated content covered over 50% of mobile search result pages, and by July, this proportion further increased to 64%.

- This AI transformation covered over 90% of Baidu App's monthly active users in July.

- Over 60% of search pages start with rich media elements such as images or videos.

- User metrics continue to improve, with Baidu App's monthly active users (MAU) reaching 735 million in June, a year-on-year increase of 5%. In the second quarter, the average daily usage time per user increased by 4% year-on-year.

- The daily generation of AIGC videos has reached millions since May.

3) Customer Service Products:

- The company leverages AI capabilities to engage in multi-turn dialogues with users through agents, connecting service providers and facilitating task completion across multiple verticals.

- Digital humans, as an innovative monetization avenue, are increasingly applied in various fields such as healthcare, legal services, education, and automotive, enhancing conversion efficiency through real-time interaction and 24/7 availability.

4) AI Cloud Business:

- Revenue Performance: AI cloud business revenue reached RMB 6.5 billion, achieving a year-on-year growth of 27%. The growth of the AI cloud business is the main driver for Baidu Core's non-online marketing revenue to exceed RMB 10 billion for the first time, with a year-on-year increase of 34%.

- Technical Advantages and Infrastructure: With a unique four-layer AI architecture, the company continues to optimize end-to-end capabilities, achieving over 90% utilization in key clusters for critical tasks, solidifying its position as a top AI cloud service provider in China.

- Product and Tool Development: The platform offers a comprehensive model library, including mainstream foundational models like the ERNIE 4.5 series. The toolchain is the most comprehensive, covering industry-leading reinforcement learning and post-training tools. The internally widely adopted AI coding assistant CodeMate contributed over 45% of the company's new code generation in July, significantly enhancing engineering productivity. The company also launched true no-code capabilities, with approximately 200,000 applications created on the platform by July.

- Customer Demand: The customer portfolio continues to improve, with existing customers deepening cooperation and increasing spending, while mid-sized enterprise customers show strong growth momentum. The company has established multiple strategic partnerships with China's leading lifestyle platforms and top gaming companies. In the embodied AI industry, partnerships have been established with 20 companies.

5) Mobile Ecosystem:

- Online Marketing Revenue: Baidu Core's online marketing revenue was RMB 16.2 billion, a year-on-year decrease of 15%.

- Large Language Model Agent: Revenue from agents for advertisers increased by 50% quarter-on-quarter, contributing 13% to Baidu Core's online marketing revenue. Revenue generated by digital humans increased by 55% quarter-on-quarter, contributing 3% to Baidu Core's online marketing revenue.

- Future Outlook: Although the AI transformation is still in its early stages, the company has received positive feedback from small-scale commercialization tests and is committed to creating long-term value for users, customers, and shareholders through AI innovation.

6) Intelligent Driving:

- Technological Progress: Apollo Go (Robotaxi) provided over 2.2 million fully autonomous ride services to the public in the second quarter.

- Operational Scale: This figure represents a year-on-year increase of 148%, marking the company's strongest quarterly growth in two years. As of August, the cumulative number of public ride services exceeded 14 million. As of June, Apollo Go's global footprint covered 16 cities, with the fleet accumulating over 200 million kilometers of autonomous driving mileage.

- Future Expansion: The company has made significant progress in international markets, establishing strategic partnerships with Uber and Lyft, and plans to deploy thousands of fully autonomous vehicles in multiple markets across Asia, the Middle East, and Europe. In Hong Kong, the testing range has been expanded to the Southern District. In Dubai and Abu Dhabi, open road testing in designated areas began in August. Apollo Go is a leader in the global right-hand drive robotaxi market. The company is also accelerating the implementation of a light-asset business model domestically.

2.2, Q&A Analyst Questions and Answers

Q: What is the current market landscape for rapidly iterating AI models? How does Baidu strategically position ERNIE, and how does it align with Baidu's broader business strategy?

A: The current market sees extremely rapid model iteration, with more powerful new models being released almost weekly, showcasing deeper logic and creativity. The landscape of foundational models is becoming increasingly diverse, with different models having advantages in specific tasks, rather than a 'one-size-fits-all' approach. Baidu is clearly positioned as application-driven innovation, focusing on strategic areas that can have a meaningful impact and maintain leadership since the launch of ERNIE two years ago. For example, in the AI search transformation, we direct model capabilities towards generating and selecting multimodal search results; cloud customers also use our search API for RAG (Retrieval-Augmented Generation); hyper-realistic digital human technology performs excellently in live e-commerce scenarios, even surpassing real humans.

Q: Can management share the plans for ERNIE in the second half of this year? What are the focus areas for the new version ERNIE 5.0?

A: Looking ahead to the second half of this year, we will continue to accelerate the development of ERNIE. We are currently working on developing the next flagship version of ERNIE, which is expected to have significant improvements in key capabilities and plans to launch as soon as it is ready. Meanwhile, we will continue to release iterations and updates for existing models. In the future, we will closely monitor industry developments to ensure our technology roadmap captures the most promising market opportunities.

Q: How did the AI-driven search upgrade progress in the second quarter? Can you share the latest metrics on how user behavior is changing with the new experience?

A: In the second quarter, we continued to accelerate the AI search transformation. Baidu is a leader in utilizing AI to transform search globally, actively revolutionizing the search experience by replacing traditional links with intelligent AI answers and multimodal content. This creates a more efficient and intuitive user experience. Users exposed to AI search exhibit higher unique visitor numbers (UV) and retention rates, indicating that the new generation of search experience significantly enhances user satisfaction.

Q: How do you view the ultimate form of AI search, including product form, user coverage, and commercial potential?

A: The ultimate form of AI search remains an open question, but our strategy is very clear: fundamentally reconstruct search. First, we no longer just index information but provide intelligent AI-generated answers presented with relevant multimodal content. High-quality AIGC content directly enriches search results, broadening the content accessible to users. Second, search will evolve from finding information to completing tasks and connecting real-world services, such as personalizing pages through MCPs and agents, understanding user context, memory, and preferences to provide customized responses. In the long term, AI search will unlock enormous commercial possibilities and upside potential.

Q: Can you provide a breakdown of AI Cloud revenue composition and profitability? What is the ratio of subscription-based revenue to project-based revenue, and how will it evolve in the future?

A: In the second quarter, AI Cloud revenue grew by 27% year-on-year, reaching RMB 6.5 billion. Enterprise cloud is the main growth driver, with its growth rate consistently surpassing the overall AI Cloud business. Within enterprise cloud, subscription-based revenue accounts for more than half and continues to grow steadily, mainly due to over 50% year-on-year growth in subscription-based AI infrastructure. Project-based revenue is related to customer deployments and fluctuates between quarters. We are prudently reviewing the project portfolio to gradually reduce its proportion for higher revenue stability.

Q: What is the short-term and long-term profitability outlook for Baidu's AI Cloud business?

A: In terms of profitability, Baidu AI Cloud achieved year-on-year growth in non-GAAP operating profit and maintained a healthy profit margin. This is due to a healthier revenue structure, with a tilt towards higher-value products. Although profit margins may fluctuate between quarters due to the dynamic market environment, we believe there is significant potential for improvement in the long term as business scale expands and product mix optimizes.

Q: Can you share details of recent AI search commercialization tests? How are these ad formats and business models evolving, and what will the profitability be like?

A: We have begun early tests of AI search commercialization and have seen very encouraging signals. AI search brings an intuitive, integrated AI-native application experience that enhances rather than interrupts the user experience, allowing the vast majority of previously difficult-to-commercialize keywords to now be commercialized under AI search, significantly expanding ad inventory.

We are transitioning from simply generating sales leads to delivering real-world services through AI-native commercial products such as agents or digital humans, connecting users with vertical service providers in healthcare, travel, and education. In the long term, the ability to meet user needs end-to-end allows us to gradually shift from a CPC (cost-per-click) model to a CPS (cost-per-sale) model, which offers a higher commercialization ceiling. In the short term, we expect revenue and profit to face significant pressure, but in the long term, this lays a solid foundation for stronger growth.

Q: How do you assess the sustainability of AI-driven cloud demand, especially in the context of economic weakness and intensified market competition?

A: We observe strong and sustained growth in demand for AI-driven cloud services. The Chinese cloud market is transitioning towards AI-centric computing, with the adoption of general AI and foundational models accelerating. Demand comes not only from early adopters like internet companies but also from a broader range of industries such as utilities, financial services, and the public sector, where interest in AI-driven cloud solutions is rapidly rising.

Technological advancements are driving strong new demand in emerging fields like embodied AI, and we have partnered with 20 of China's most promising embodied AI startups, including four top humanoid robot companies, effectively capturing new opportunities.

Q: With the relaxation of H20 chip restrictions, has management seen any meaningful improvement in supply? How do you view chip restrictions, and will they continue to be a limiting factor for future growth?

A: Regarding the chip issue, our focus is on building flexible AI architectures to maximize GPU utilization and support multiple chips, including domestic ones. This enables us to better serve customers as the supply environment changes. Looking ahead, we believe a self-sufficient supply chain, coupled with an increasingly mature local software stack, will lay a solid foundation for sustainable innovation in China's AI ecosystem. Baidu is well-positioned to lead this transformation.

Q: Considering the recent headwinds in advertising revenue and continued investment in AI, what cost optimization and efficiency improvement plans are in place to help protect margins?

A: We are committed to AI investment, particularly in the AI transformation of search, where significant investments have been made. Given that AI search commercialization is still in its early stages and scale has not yet formed, revenue and profit are under considerable pressure in the short term, with the third quarter expected to be particularly challenging. To mitigate short-term impacts, we will actively drive internal efficiency improvements, including strengthening resource coordination across different business groups and enhancing overall resource utilization efficiency.

Q: How should we view profit trends for 2026 and beyond?

A: While we remain committed to long-term AI investment, we will be very prudent in managing the pace to avoid future margin deterioration or volatility. Looking ahead, we see potential for profit improvement as the core advertising business recovers and stabilizes, and non-advertising businesses expand their revenue share and improve their profitability. We believe that strategic direction and rigorous execution can support a gradual recovery in margins over time. By the end of this year or around that time, we expect to have greater visibility into next year and will provide a long-term clear outlook beyond the current quarter.

Q: How does Apollo Go assess its long-term defensive capability against peers in the face of an increasingly competitive landscape? And what is the roadmap to achieving sustainable profitability?

A: Baidu Apollo Go has established a solid foundation in technology, large-scale continuous investment, and years of rigorous execution, becoming the undisputed global leader in this field. We focus on scaling actual operations and have established global leadership in both left-hand and right-hand drive robotaxi markets.

In the Wuhan market, we are the first company to achieve unit economics (UE) breakeven, and all operations in mainland China are fully autonomous. Our RT6 is the world's first and only mass-produced vehicle designed for L4 autonomous driving, with the lowest unit cost among global L4 vehicles, providing a significant advantage in large-scale commercial operations. Achieving UE breakeven in Wuhan (where taxi fares are over 30% lower than in China's first-tier cities) demonstrates the excellent operation and cost efficiency of the business model.

Q: How does Apollo Go's recent partnerships with Uber and Lyft fit into your global expansion strategy?

A: To accelerate global expansion, we are proactively pursuing global partnerships. We announced a partnership with Uber in July, followed by a partnership with Lyft in August. Partnerships with these world-leading mobility platforms will help us enter and expand into global markets such as the Middle East, Asia, and Europe more quickly. Looking ahead, we expect ride volumes to accelerate, and the global operational fleet size to multiply. With this momentum, we are confident that Apollo Go will continue to lead the market and remain at the forefront of the global driving field.

Risk Disclosure and Statement of this Article:Dolphin Research Disclaimer and General Disclosure