Artificial Intelligence for Science (AI4S) is rapidly permeating all stages of the innovation chain, driving national innovation systems (NIS) toward a new stage of intelligent transformation. Based on a three-dimensional analytical framework—activities and functions, actors and organizations, and conditions and environment—this study systematically analyzes the mechanisms through which AI4S induces structural transformation in NIS, and identifies key challenges and corresponding policy responses in the Chinese context. The findings show that AI4S restructures the core mechanisms of knowledge production, diffusion, and application, shifting innovation activities from linear processes to iterative cycles. It also transforms innovation actors from labor-dominated modes to human-machine collaboration, while fostering the emergence of platform-based and ecosystem-oriented hub actors. At the same time, it places new demands on factor allocation and institutional operating arrangements. However, China’s NIS still faces several constraints in this transition, including uneven intelligentization across different stages of the innovation chain, lagging capability transformation of innovation actors, insufficient supply of key intelligent factors, and limited institutional adaptability. To address these challenges, this study proposes coordinated efforts to restructure knowledge operation mechanisms, optimize collaborative patterns among innovation actors, and improve institutional design.