Baidu
Based on search engine, empowering AI applications and cloud services through ERNIE large models and PaddlePaddle platform
Leveraging massive Chinese-language data and knowledge graphs accumulated from search scenarios, building a full-stack closed loop of ERNIE large model + PaddlePaddle deep learning
Company Overview
Baidu was founded in 2000, listed on NASDAQ in the United States in 2005 (BIDU), and completed a secondary listing in Hong Kong in 2021 (9888.HK), with its headquarters in Beijing. The company started from a Chinese search engine and has gradually evolved into a comprehensive AI enterprise spanning AI Cloud, autonomous driving, large models, and smart hardware. In the "Models and Middleware Layer" of the AI industry chain, Baidu has become one of the core players in the domestic large-model competition, powered by its self-developed PaddlePaddle deep learning platform and the ERNIE large model series. In 2023, the company released ERNIE Bot, becoming the first to open it to the public, and in 2024, launched multiple version iterations, actively driving the deployment of large models across search, cloud services, autonomous driving, and other business lines.
Baidu's core revenue source remains search engine and information feed advertising within the Baidu App. Its AI large models are deeply transforming the search experience through "AI Search," improving user retention and ad matching efficiency. In 2024, Baidu Search began integrating Ernie Bot's summary generation and multi-turn dialogue capabilities, directly competing with the search products of ByteDance and Alibaba.
Baidu AI Cloud, powered by the PaddlePaddle platform and Ernie large models, provides AI cloud services such as model training, inference deployment, and MaaS (Model as a Service). In 2023, AI Cloud revenue reached approximately RMB 18.7 billion, up 15% year-over-year. In 2024, Baidu aggressively cut prices (Ernie Bot API pricing as low as RMB 0.003 per thousand tokens) to win enterprise customers, although gross margins came under pressure.
The Apollo autonomous driving platform follows a "Vehicle-Road-Cloud Integration" approach. In 2024, Apollo Go (autonomous ride-hailing) achieved large-scale operations in cities such as Wuhan, with cumulative orders surpassing 1 million. Large models are used for perception, decision-making, and simulation training, representing a vertical application scenario of AI technology.
| Product Line | Revenue Share (2024E) | Core Customers | Gross Margin |
|---|---|---|---|
| Search & Mobile Ecosystem | ~60% | Advertisers, individual users | ~55% |
| Baidu AI Cloud | ~18% | Enterprises, developers, government & enterprise clients | ~20% |
| Autonomous Driving (Apollo) | ~3% | Local governments, mobility platforms | Negative (R&D investment phase) |
| Others (iQiyi, Xiaodu, etc.) | ~19% | Consumers, telecom operators | ~10% |
Technical Moat
Moat 1: PaddlePaddle Ecosystem
PaddlePaddle is the first open-source deep learning platform in China, with over 230 million downloads and more than 5 million developers as of 2024. It provides a full-process toolchain from training to inference, deeply optimized for NVIDIA GPUs as well as domestic Ascend and Cambricon chips, forming a three-tier binding of "chip-framework-model." Once developers complete model development on PaddlePaddle, the migration cost is relatively high.
Moat 2: Massive Chinese Data and Knowledge Graphs
Baidu Search has accumulated more than 20 years of Chinese web pages, text, images, and user behavior data, which are used to build domain-specific knowledge graphs. ERNIE (Wenxin) large models have a natural advantage in tasks such as Chinese understanding, common-sense reasoning, and long-text processing, ranking among the top in Chinese benchmark tests such as C-Eval and MMLU.
Moat 3: Full-Stack AI Closed Loop (Cloud + Edge + Vehicle)
Baidu has established an end-to-end chain from underlying chips (Kunlun), mid-level frameworks (PaddlePaddle), upper-level large models (ERNIE), to application layers (Search, Cloud, Autonomous Driving, Xiaodu). Compared to startups that only offer large model APIs, Baidu can leverage its own scenarios to rapidly iterate models and achieve a data flywheel across its three core battlegrounds: Search, Cloud, and Vehicle.
| Dimension | Data |
|---|---|
| Domestic large model market share | Approximately 15-20% (Q3 2024, second only to Alibaba Qwen) |
| Industry ranking | Ranked 2nd-3rd (in the same tier as ByteDance Doubao, Tencent Hunyuan, and Zhipu AI) |
| Main competitors | Alibaba (Qwen), ByteDance (Doubao), Tencent (Hunyuan), Zhipu AI (GLM) |
| Downstream customers | Internet enterprises, finance, healthcare, education, local governments, autonomous driving partners |
| API pricing | ERNIE Bot API RMB 0.003 per 1,000 tokens (after the 2024 price cut), lower than Zhipu AI (RMB 0.005) but higher than ByteDance Doubao (partially free) |
Financials and Growth
| Metric | Data |
|---|---|
| Revenue (2023) | RMB 134.56 billion (YoY +8.8%) |
| Gross Margin (2023) | 51.6% (dragged down by low margins in Intelligent Cloud) |
| Net Margin (2023) | 16.3% (adjusted) |
| R&D Investment (2023) | RMB 27.4 billion (20.4% of revenue) |
| Core Growth Logic | ① Large model-empowered search boosts ad revenue; ② Intelligent Cloud MaaS captures enterprise AI demand; ③ Apollo autonomous driving commercializes faster; ④ iQIYI and Xiaodu AI terminal monetization |
Key Risks:
- Price War Pressure: Tech giants such as Alibaba, ByteDance, and Tencent have launched ultra-low-priced APIs or free plans. Ernie Bot's API gross margin may decline from over 50% to below 10%, impacting the profitability of Intelligent Cloud.
- Underwhelming Commercialization: The paying user base for Ernie Bot in the C-end market remains limited (approximately 5 million paying users in 2024, far below ChatGPT's tens of millions). Despite price cuts in the enterprise segment, growth in incremental customers has slowed.
- Autonomous Driving Profitability Remains Distant: Apollo continues to burn cash, overseas operations face setbacks (testing scaled back in California), and although domestic policies are favorable, high operational costs make near-term profit contribution unlikely.
- Regulatory and Compliance Risks: Concerns over the authenticity and privacy of large model-generated content, coupled with tightening regulations, may restrict deployment in certain industries.
Core Investment Thesis / Industry Value Summary:
Baidu is one of the few "full-stack" players in the AI industry chain, possessing a complete closed loop from chips (Kunlun) to frameworks (PaddlePaddle) to applications (search, cloud, and autonomous driving). Despite facing intense price wars from major competitors and commercialization uncertainties, its monopoly position in the search scenario and data flywheel effect are difficult to replicate. In the long term, AI-powered search will reshape traffic entry points, Intelligent Cloud MaaS is expected to achieve economies of scale after price reductions, and autonomous driving offers high-elasticity optionality. The current valuation (approximately 15x PE in 2024) already reflects pessimistic expectations. If large model application penetration exceeds forecasts, Baidu will be among the most beneficiaries among Chinese companies.