Baidu (Minutes): AI search needs to restructure monetization model, short-term revenue may be under pressure

Below is the$Baidu(BIDU.US) FY25Q1 earnings call minutes. For earnings analysis, please refer to《Baidu: DeepSeek as the "Soul Ferry"? Survival Still Depends on Itself》

1. Key Financial Highlights

1. DeepSeek with the lowest price in the industry, driving explosive growth in smart cloud: Baidu accidentally became the first platform to fully capitalize on the DeepSeek boom. After DeepSeek gained popularity, Baidu not only quickly integrated it but also launched a major promotion, offering the lowest-priced computing power in the industry, directly boosting smart cloud growth from 28% last quarter to 42% this quarter.

2. User growth "a long drought ends with sweet rain": Besides cloud, mobile Baidu's traffic also benefited from DeepSeek. In mid-February, the full version of DeepSeek was integrated into Baidu search, activating dormant users, with mobile Baidu's MAU increasing by 45 million sequentially, reaching a record high of 724 million.

3. Advertising performance underwhelming, relying on search monetization to fill the gap: Advertising revenue declined 6% YoY, in line with guidance. However, given that the macroeconomic environment in Q1 was not as bad as expected, some institutions had slightly upgraded their expectations for Baidu's ad performance last month. This led to a disappointment in ad revenue from an expectation gap perspective.

This further indicates that the new traffic in Q1 was primarily driven by AI search, with traditional ads unable to directly capitalize. Baidu plans to gradually roll out AI search monetization in Q2, and investors should pay attention to management's implementation plan and milestones.

4. Improving economic model in innovation businesses: Gross margin declined 7 ppts YoY, mainly due to changes in business mix. However, a rough breakdown shows that other businesses' gross margins continued to improve, with the improvement likely tied to the adoption of RT6 for Apollo Go and accelerated cloud business.

5. AI-driven internal efficiency continues: While Baidu may not have reaped external benefits from AI yet, internal efficiency improvements have been ongoing for nearly a year. In Q1, apart from increased sales and marketing expenses, R&D expenses continued to decline, reflecting AI's role in improving operational efficiency—a trend that can be observed in other R&D-heavy internet platforms.

6. Resumption of capex expansion: The current expansion should be viewed positively. Baidu's capex has lagged behind other internet giants, and last year, to avoid dragging down profits and cash flow, it even actively reduced spending. This was also related to Baidu's ongoing stockpiling of computing chips and in-house chip development. However, in Q1, possibly due to concerns about chip restrictions, capex rebounded by 40%.

7. Share buybacks continue to increase—is Baidu finally loosening up?: Q1 free cash flow was negative, likely due to short-term debt repayment, YY acquisition payments, and increased capex. Net cash (excluding debt) stood at RMB 133.7 billion ($18.6 billion). Baidu repurchased $445 million in core net cash in Q1, continuing the sequential increase, bringing cumulative buybacks under the 2023 plan (3-year $5 billion) to $2.1 billion. The buyback pace remains slow unless the plan was never intended to be fully executed.

However, given the long quiet period in Q1, if buybacks normalize to over $500 million per quarter for the rest of the year, annual buybacks could nearly double to $2 billion. At the current market cap of $30.7 billion, the yield would be 6.5%.

2. Detailed Earnings Call Content

2.1 Key Management Remarks

1. Foundation models: ERNIE 4.5 and ERNIE X1 were released in March, with upgraded ERNIE 4.5 Turbo and ERNIE X1 Turbo launched in April, offering enhanced performance at better prices. Plans to open-source the ERNIE 4.5 series on June 30. The unique four-layer AI architecture enables end-to-end optimization, reducing costs and providing cost-effective AI solutions.

2. Qianfan platform: Expanded model library to include mainstream models, offering inference and multimodal models with leading cost efficiency. Enhanced toolchain with new data builder, upgraded model builder (supporting custom inference model development), expanded fine-tuning capabilities (supporting multimodal models), and introduced one-click distillation to lower AI application barriers.

3. Baidu search: Accelerated AI transformation, establishing a mature and scalable product framework in Q1. In April, ~35% of mobile search results pages contained AI-generated content (vs. 22% in January). Prioritized multimodal content display, with content volume growth (e.g., AI-generated digital human videos grew over 30x in months), improving user experience, search intent fulfillment, and retention. Baidu App MAU reached 724 million in March, up 7% YoY.

4. Smart digital humans: Integrated smart agents with digital humans, widely deployed in the mobile ecosystem. Launched upgraded ultra-realistic interactive digital humans, to be mass-produced soon.

5. Autonomous driving (Apollo Go)

a. Achieved 100% fully driverless operations in mainland China, validating the business model. Entered Dubai and Abu Dhabi in Q1, began open-road testing in Dubai in May, with Abu Dhabi testing upcoming and expanded test zones in Hong Kong with passenger testing permits.

b. Globally deployed over 1,000 fully driverless vehicles. Provided ~1.4 million rides to the public in Q1, up 75% YoY, with cumulative rides exceeding 11 million as of May. Exploring light-asset models, including a strategic partnership with CAR Inc.

6. Business partnerships and customer expansion

a. AI cloud: Deepened collaboration with existing clients and expanded customer base, partnering with China Merchants Group, top domestic e-commerce firms, etc., with a strong pipeline. Expanded into automotive and embodied AI, forming a strategic partnership with Beijing Humanoid Robot Innovation Center.

b. Mobile ecosystem ads: Smart agents improved ad monetization efficiency, with over 29,000 advertisers daily using smart agents in March. AI Agent-related revenue grew 30x YoY in Q1, accounting for 9% of Baidu core online marketing revenue, covering multiple industries.

2.2 Q&A

Q: Regarding AI model iteration and open-source strategy, can you share updates on Baidu's 2025 AI strategy, roadmap, and plans for ERNIE 5.0? Will inference costs further decline?

A: Strategy: Committed to "application-driven innovation," focusing on practical model applications and integrating foundation models with product needs (e.g., AI transformation in mobile ecosystem and search).

Roadmap: Plans to open-source ERNIE 4.5 on June 30 to expand community engagement and explore new use cases. Ongoing iteration of ERNIE models, with next-gen R&D already underway, aiming to accelerate release cycles.

Inference costs: Continued cost reduction via full-stack AI capabilities (e.g., chips, frameworks, model optimization). Recent models already offer significant price cuts (e.g., ERNIE X1 priced at half of competitors'), with plans to maintain cost advantages.

Q: What drove Q1's strong cloud growth? Sustainability? Can you break down cloud revenue by infrastructure/industry solutions? 2025 cloud outlook? Impact of US AI chip restrictions?

A: Drivers: Surging demand for GenAI and foundation model training/inference across industries, especially accelerated model iterations. Baidu's leading AI infrastructure and Qianfan's capabilities (lowering inference costs, improving toolchain efficiency) attracted clients.

Sustainability: Enterprise cloud's subscription-based revenue dominates, providing recurring income, with GenAI-related revenue growing triple-digit YoY for multiple quarters. Long-term, subscription share will rise, supporting sustainable growth.

Breakdown: AI cloud splits into consumer and enterprise, with the latter driving most revenue and growth. Enterprise further divides: subscriptions dominate, while project-based revenue fluctuates and will decline long-term.

Outlook and margins: AI cloud Non-GAAP op margin expanded YoY, now in double digits, driven by high-value product mix.

Chip restrictions: Limited impact. Baidu's "application-driven" strategy focuses on AI application value, with full-stack capabilities enabling cost-effective solutions. AI infrastructure supports high GPU utilization and multi-chip flexibility (especially inference). Long-term, domestic chips and software will reduce external reliance.

Q: Logic behind accelerating AI search? AI answer penetration expectations? Q2 monetization tests? H2 monetization and user behavior trends?

A: Logic: Core goal is enhancing user experience via innovation, where superior UE and metrics underpin sustainable growth. Rapid AI advancements and diverse search behaviors necessitate agile innovation.

Penetration: ~35% of mobile search results are AI-generated (vs. 22% in Jan), expected to rise further in Q2. Model improvements (multimodal generation, lower pricing) and multimedia integration drive penetration.

Monetization tests: Early-stage preparations ongoing, as AI search differs from traditional search, requiring new monetization models. No large-scale tests yet. Exploring AI-native ad formats (e.g., non-intrusive, multimodal-compatible) for seamless integration.

H2 trends:

(1) Monetization: AI search monetizes long-tail queries and non-commercial scenarios untapped by traditional search. New ad formats may test in H2, potentially boosting long-term efficiency but pressuring near-term margins.

(2) User behavior: Rising acceptance of AI results, with higher info efficiency, richer queries, and improved retention.

Q: Robotaxi competition? Apollo RT6 differentiation? Partnership models? 2024 expansion plans? Unit economics and long-term profitability?

A: Competition: Avoided direct comparison, highlighting Apollo Go as a global leader in scaled operations.

Differentiation: World's first L4 mass-produced vehicle, designed for full autonomy, with in-house hardware/algorithms/software and top safety redundancy. Sub-$30k cost per vehicle beats global peers. Full-stack capabilities enable efficient scaling.

Partnerships: Exploring models with ride-hailing platforms and fleet operators (e.g., CAR Inc), focusing on rapid scaling.

Expansion: Expect accelerated growth, with 1,000+ vehicles in 15 cities (including new Dubai/Abu Dhabi entries). Flexible expansion into regulation-friendly markets.

Long-term: No specific targets, but fleet size, city coverage, and orders to grow. Clear path to profitability via cost declines and scale efficiencies.

Q: How does Baidu compete with AI apps/platforms (e.g., enhanced models, deep search agents, super-app traffic)?

A: Early AI search innovation addresses industry shifts. Chatbots are exploratory, not the endgame. Competition centers on redefining search value in the AI era, focusing on: (1) rich multimodal content for better UE, (2) agents solving complex tasks beyond info retrieval.

Strategically, open ecosystems enhance capabilities, integrating MCP/third-party agents to connect tools/services while empowering partners (e.g., e-commerce MCP). Innovations expand search boundaries, creating long-term value.

Q: China cloud TAM with AI adoption? Baidu Cloud differentiation? Qianfan updates? Fastest-growing AI verticals? Long-term AI potential?

A: TAM: Driven by foundation models' massive compute needs, where GPU cluster efficiency is key. Model diversification favors full-stack cloud platforms.

Differentiation: (1) Strategy: "Application-driven," tailoring full-stack AI solutions to industries. (2) Tech: Rare global full-stack AI cloud, with China's most efficient infrastructure. (3) Clients: Deep ties with leaders + mid-market appeal.

Qianfan: Among the most advanced AI dev platforms, now covering Baidu + third-party/open models. Q2 toolchain upgrades expanded training/fine-tuning/distillation, especially for multimodal/inference models.

Fastest-growing: Internet, tech, online ed.

Highest potential: Auto, finance, utilities, public sector.

Q: 2025 capital allocation priorities?

A: Increasing AI investments, with 2024 total (capex + OCF) up significantly vs. 2023, and 2025 plans to further strengthen foundations.

Focus areas:

AI cloud: Expand infrastructure amid surging demand (Q1 revenue +42%, margins in double digits).

ERNIE: Optimize tech for low-cost, high-quality models.

Autonomous driving: Scale Apollo Go in China/Middle East post-model validation.

AI search: Prioritize UE despite near-term pressure.

Shareholder returns: Accelerated buybacks in 2025, reflecting commitment.

<End>

Disclosures & Disclaimers: Dolphin Research Disclosures