In the digital era, technological catch-up is inevitable for firms confronted with intensified competition, rapid technological advancements, and customers’ upgrade requirements. By strengthening their cooperation with external parties and integrating internal knowledge, firms can better absorb internal and external resources, accelerating their technological catch-up performance (TCP). This study mainly explores the influence of collaboration and knowledge networks on the TCP of firms based on machine learning algorithms. First, patent data from the Chinese AI industry from 2013 to 2022 were used to construct collaboration and knowledge networks. Then, the hierarchical clustering algorithm was applied to categorize firms based on six network characteristics. Finally, the classification and regression trees (CART) algorithm was employed to analyze the nonlinear relationship between dual networks and firm TCP. The findings show that firms exhibit distinct network configurations and that the drivers of TCP vary across firm groups. For firms lagging behind, prioritizing knowledge network integration proves more effective than expanding collaborations. Leading firms perform best when maintaining balanced collaboration strategies. This study contributes to both theory and practice by identifying the optimal mix of network characteristics and providing empirically grounded strategies for different firm types.