Decoding AI innovation: How R&D alliances drive technological breakthrough

Artificial intelligence (AI) innovation is central to technological progress in the digital economy, yet little is known about how different R&D alliance types shape this process. Drawing on resource dependence theory (RDT), we examine how market- and research-oriented alliances affect AI innovation using patent-based measures constructed via a bag-of-words (BoW) model and panel data on Chinese listed firms. Results show that market alliances significantly enhance AI innovation, including generative AI (Gen_AI), whereas research alliances hinder general AI innovation and exhibit no significant relationship with Gen_AI. These results remain consistent across different robustness (replacement variables, exclusion of financial volatility, stricter fixed effects, dual machine learning) and endogeneity tests (GMM and instrumental variables). Moderation analyses reveal that knowledge path dependence weakens the benefits of market alliances and amplifies the drawbacks of research alliances, while dynamic capabilities reverse these effects by enabling knowledge integration and redeployment. Therefore, we extend RDT by revealing how internal inertia and adaptive capacity influence the effectiveness of R&D alliances in driving AI innovation through moderation effects. These findings offer theoretical insights and practical guidance for firms leveraging R&D alliances to sustain AI innovation in fast-changing environments.