CGI深度 | 人工智能产业创新：强者的游戏？

Core Viewpoint - The article analyzes the innovation and competition landscape of the AI industry, highlighting the varying degrees of "Matthew Effect" across different segments, specifically in AI chips, foundational models, and vertical applications. It emphasizes that the chip and foundational model sectors exhibit strong Matthew Effect characteristics, while vertical applications show weaker effects [3][32]. Group 1: AI Industry Segmentation - The AI industry consists of three main segments: chip layer, foundational model layer, and vertical application layer, each exhibiting different innovation and competition dynamics [3]. - The chip and foundational model sectors are characterized by a high concentration of leading firms, indicating a strong Matthew Effect, while the vertical application layer is more fragmented with numerous players [4][6]. Group 2: Innovation Models - The article introduces Schumpeter's innovation models, categorizing industries into "Schumpeter Mark I" (low concentration, unstable competition) and "Schumpeter Mark II" (high concentration, stable competition). The AI chip and foundational model sectors fall under the Schumpeter Mark II category, indicating a strong Matthew Effect [5][8]. - The analysis of patent data from 2000-2023 shows that both AI chips and foundational models exhibit a high concentration of innovation activities, reinforcing their classification as Schumpeter Mark II industries [11][13]. Group 3: Economic Logic of the Matthew Effect - Four characteristics influence the strength of the Matthew Effect in an industry: convergence of dominant designs, sources of innovation knowledge, product generality, and customer switching costs. A higher convergence of designs, reliance on practical knowledge, high product generality, and high switching costs lead to a stronger Matthew Effect [4][14][24]. - The AI chip sector shows high design convergence and product generality, while foundational models also exhibit similar characteristics, contributing to a strong Matthew Effect in both sectors [26][31]. Group 4: Policy Recommendations - The article suggests that the government should focus on supporting leading domestic firms in the AI chip and foundational model sectors, as these industries demonstrate a strong Matthew Effect. Concentrated investment in top-tier companies is deemed more effective than dispersed investments [38][40]. - Demand-side policies, such as public procurement and subsidies, are recommended to create a favorable environment for domestic AI chip and model manufacturers, encouraging the use of local products [41][42].