Based on 134 recently released AI-related job postings in the US library field, this paper employs the Latent Dirichlet Allocation (LDA) topic model to conduct a quantitative analysis of job responsibilities and qualification requirements. It systematically reveals the characteristics of AI competency demands for librarians, and accordingly constructs an AI competency framework for librarians, clarifying the organic interconnections among various competency elements. Additionally, the paper summarizes the characteristics and experiences of AI competency requirements for librarians in US libraries, and proposes recommendations for the development of AI competencies among Chinese librarians. These recommendations include focusing on the cultivation of technical competencies in the Chinese context, promoting the in-depth integration of job responsibilities with library service scenarios, and establishing a governance guarantee and capacity-building system tailored to library contexts. This research aims to provide decision-making support for the talent team construction of Chinese libraries in the process of intelligent transformation.