The paradox of AI investment in innovation-to-performance pathways

Purpose This study examines the paradoxical and stage-dependent role of artificial intelligence (AI) investment along the innovation-to-performance pathway. While AI is widely assumed to enhance firms’ innovation outcomes, emerging evidence suggests that its effects are uneven and potentially contradictory. This research investigates how AI strengthens the transformation of external knowledge into internal innovation capabilities, yet may simultaneously constrain the downstream realization of performance. By distinguishing total AI investment from dedicated AI hardware investment, the study clarifies how heterogeneous AI resources generate divergent impacts across stages of the innovation process. Design/methodology/approach The empirical analysis utilizes data from 314 Chinese A-share listed firms engaged in open innovation and AI-related activities. A moderated mediation model is applied to evaluate (1) the mediating role of internal innovation in converting external knowledge into performance and (2) the stage-specific moderating effects of total AI investment and AI hardware investment on the innovation pathway. All analyses are conducted using PROCESS Macro and adhere to established methodological standards for multi-stage innovation research. Findings The results reveal a clear stage-dependent paradox in the effects of AI investment. Total AI investment enhances firms’ ability to assimilate external knowledge and generate internal innovation, but it simultaneously hinders the translation of such innovation into realized performance due to coordination overload, procedural rigidity, and misalignment between digital technologies and organizational routines. In contrast, dedicated AI hardware investment strengthens the innovation-to-performance link by providing scalable computational infrastructure. The mediated effect of internal innovation is therefore contingent on how firms configure their AI resource portfolios. Originality/value This study advances paradox theory and the resource-based view by demonstrating that AI investments are neither uniformly beneficial nor functionally homogeneous. It identifies a hidden form of innovation failure—the inability to convert innovation output into performance—arising from misaligned or excessive digital investment. By differentiating between total AI investment and AI hardware investment, the study sharpens conceptualizations of digital resource heterogeneity and offers actionable insights for firms seeking to strategically balance innovation gains with performance risks. The findings thus provide a more realistic and theoretically robust understanding of AI’s role within multi-stage innovation processes.