经济观察报·2026-03-08 03:49Core Insights - The article highlights a significant shift in the usage of artificial intelligence (AI) models, particularly in the context of token consumption, indicating a transition from traditional question-answering models to more complex agent-based models that perform multi-step tasks [2][3][6]. Token Consumption Trends - In February 2026, China's AI models achieved a weekly token consumption of 5.16 trillion, surpassing the U.S. models for the first time, reflecting a 127% increase over three weeks [2]. - The OpenRouter platform, which aggregates global AI model interfaces, reported that programming task tokens rose from 11% to over 50% of total token usage, indicating a shift towards more complex applications [6]. Application Shifts - The transition to agent-based models allows for continuous task execution in the background, leading to exponential growth in token consumption per session [6][8]. - Multi-modal applications, such as video generation, significantly increase token consumption, with a 10-second video requiring approximately 350,000 tokens [7][8]. Industry Impact - The surge in token consumption is primarily supported by high-frequency, scalable commercial applications across sectors like internet, finance, cross-border e-commerce, and entertainment [9]. - Key application areas include enterprise-level solutions (e.g., intelligent customer service, marketing automation), content generation services, and AI-generated content tools [9]. Market Dynamics - The competitive landscape is shifting as domestic AI chip manufacturers gain opportunities due to changing procurement standards, moving from a focus on training costs to operational efficiency and cost-effectiveness [4][14]. - The pricing of Chinese AI models is significantly lower than their overseas counterparts, with input prices around $0.3 per million tokens compared to $5 for similar foreign products [10]. Infrastructure Developments - The launch of the Zhengzhou core node, capable of supporting over 30,000 domestic acceleration cards, marks a significant advancement in China's AI computing infrastructure [18]. - The deployment of large-scale AI computing clusters is enabling various applications, including simulations for the automotive industry and support for luxury brands [19]. Financial Performance - Domestic AI chip manufacturers are experiencing explosive growth, with companies like Cambrian achieving a revenue increase of 453.21% in 2025, driven by rising demand for AI computing power [20]. - The market penetration of domestic AI chip brands has increased from approximately 29% in 2024 to 42% in 2025, indicating a shift towards local solutions in the AI sector [21].