Sou Hu Cai Jing·2025-11-21 15:15Core Insights - Computing power is becoming the "new oil" in the era of large models, with global data center investments expected to reach approximately $580 billion by 2025, surpassing the $540 billion investment in global oil supply this year, reflecting market recognition of AI's development potential [2] - Companies are increasingly realizing that computing power is not only the foundational resource for AI model training and inference but also a core element driving industrial innovation and intelligent upgrades [2][4] - Baidu's recent showcase at the Baidu World Conference highlighted its latest explorations in AI infrastructure, demonstrating a comprehensive, scalable AI Infra solution covering chips, clusters, and platforms [2][3] AI Cloud Construction Trend - The construction of a full-stack AI Infra capability, encompassing chips, clusters, and platforms, is becoming a core focus for leading players in the AI industry, particularly in overseas markets [4] - Companies like Google, Amazon, and OpenAI are investing heavily in self-developed chips and customized AI computing systems to enhance efficiency and reduce costs [4][5] - Baidu is recognized as one of the earliest domestic companies to initiate self-developed AI chips and has been a pioneer in integrating cloud computing with AI since 2020 [5][8] Baidu's Competitive Edge - Baidu's full-stack layout from chips to platforms has established a solid advantage in computing power supply and created unique competitive barriers in AI cloud services [7] - The company has maintained a leading position in China's AI public cloud service market, holding a 24.6% market share as of last year [5] Evolving AI Infrastructure - The rapid evolution of AI models and applications is driving exponential growth in computing demand and system complexity, necessitating continuous upgrades to existing technologies and computing systems [8] - Baidu's Kunlun chip roadmap for the next five years includes the launch of new products and super nodes, with significant improvements in interconnect bandwidth and model training capabilities [9][11] Integration of Hardware and Software - The integration of hardware, computing platforms, supply chains, and teams is essential to meet the new demands posed by evolving architectures and applications [11][12] - Baidu's AI computing platform, Baidu百舸, is designed to optimize the use of domestic computing power and enhance model training and inference performance [13][15] Industry Applications and Collaborations - Baidu's full-stack AI Infra capabilities are being widely applied in its internal operations and are also being offered as scalable computing services to external clients [15][20] - The platform has successfully supported various industry leaders in deploying domestic computing power for intelligent transformation [20] Conclusion - The AI industry is moving towards a vertically integrated structure, where leading companies are increasingly focused on deep integration across applications, models, frameworks, cloud infrastructure, and chips [21] - Baidu has positioned itself as a frontrunner in this new wave of AI cloud competition by delivering a comprehensive full-stack AI Infra solution [21]