Detailed Narrative
Foundation Model Development and ERNIE
Baidu is focused on application-driven foundation model development, with ERNIE 4.5 Turbo powering digital human technology for live streaming e-commerce, generating tens of millions in GMV. They are exploring frontiers of foundation model research and pushing AI capabilities, with value creation expected at the application level. The company is working on the next flagship ERNIE version with significant improvements, to be launched when ready, alongside ongoing iterations.
AI Cloud and 4-Layer Architecture
Baidu's unique 4-layer end-to-end AI architecture is a core competitive advantage, driving healthy growth in AI Cloud. A breakthrough in prefiled detailed separation architecture significantly improves inference efficiency and reduces costs, reinforcing its position as a top-tier AI cloud provider. This allows for better performance at lower cost and more competitive pricing for enterprise customers, creating a virtuous cycle.
Qianfan MaaS Platform Evolution
Qianfan continues to evolve, offering a comprehensive model library including ERNIE 4.5 series and third-party models. It provides enhanced stability, higher concurrency, and lower inference costs, solidifying its leadership as a MaaS platform. Qianfan's tool chains are among the most comprehensive, supporting a wider range of AI tools and functions via MCP or API, simplifying AI application development.
Autonomous Driving (Apollo Go) Global Expansion
Apollo Go provided over 2.2 million fully driverless rides, up 148% YoY, marking its strongest quarterly growth in 2 years. Strategic partnerships with Uber and Lyft will deploy thousands of autonomous vehicles in international markets (Asia, Middle East, Europe), leveraging an asset-light model and strong unit economics achieved in China. Apollo Go leads in both left-hand and right-hand drive robotaxi markets, with its RT6 vehicle offering the lowest unit cost globally for Level 4 autonomous driving.
AI Transformation of Baidu Search
Baidu is revolutionizing search by replacing static hyperlinks with intelligent, structured, multi-model first AI-generated responses. AI-generated content reached 64% of mobile search result pages by July, reaching over 90% of MAUs, with over 60% starting with rich media. This transformation aims to expand search boundaries from providing answers to completing tasks and connecting real-world services through agents.
Internal AI Productivity and External Democratization
Internally, Comate, an AI coding assistant, contributed to generating over 45% of new code in July, significantly boosting R&D efficiency. Externally, Miaoda extends AI development capabilities to the broader community, offering no-code functionalities. As of July, users have created around 200,000 applications on Miaoda completely without writing a single line of code, democratizing AI innovation.
AI Search Monetization and Digital Humans
Early small-scale monetization testing for AI Search shows potential, especially for previously difficult-to-monetize queries. Agents grew revenue for advertisers by 50% QoQ, contributing 13% of online marketing revenue. Digital humans increased revenue by 55% QoQ, contributing 3% of online marketing revenue, adopted in sectors like healthcare, legal, education, and automotive for boosting conversion performance.