This paper constructs a novel metric on enterprise AI application leveraging recent breakthroughs in large language models. Based on this metric, we examine the impact and mechanism of AI application on enterprise global value chain (GVC) embeddedness. The findings indicate that AI application significantly promotes enterprise GVC embeddedness, a conclusion robust to alternative specifications including the replacement of large language models, adjustments to the dependent variable, and changes in estimation methods. The mechanism through which AI application drives GVC embeddedness lies in its enhancement of productivity, which further accelerates technological innovation and mitigates resource misallocation. The positive effect of AI application on GVC embeddedness is more pronounced among smaller enterprises, non-state-owned enterprises, firms located in regions with lower marketization levels, and capital-intensive enterprises. The evolving uncertainty of the GVC system in a time of great change highlights the pressure on industrial chain security resulting from AI's promotion of GVC embeddedness.