The Asymmetric Crowding-Out Effect of Ambidextrous Artificial Intelligence Attention Allocation on Breakthrough Innovation

How firms allocate resources between artificial intelligence (AI) technology exploration and application exploitation is a critical issue in the pursuit of breakthrough innovation. Using data from Chinese A-share listed firms from 2007 to 2024, this study examines the impact of ambidextrous AI attention allocation on firms' breakthrough innovation. The results show that attention to either AI technology exploration or AI application exploitation, when considered independently, promotes breakthrough innovation output. However, under resource constraints, simultaneous high levels of attention to both dimensions may generate resource competition and cross-departmental coordination frictions, thereby producing a crowding-out effect that suppresses innovation.This crowding-out effect exhibits an asymmetric evolutionary pattern. The effect of application exploitation on technology R&D follows a relatively smooth and linear trajectory, whereas the constraining effect of technology exploration on application exploitation displays a threshold pattern. Once attention to technology exploration exceeds a certain threshold, it may induce resource lock-in and increase organizational coordination costs, thereby reducing the marginal efficiency with which application exploitation is transformed into innovation output. Moderation analyses further indicate that greater AI investment intensity creates a dedicated resource buffer, while deeper digital transformation mitigates internal organizational frictions by enhancing information sharing and reducing cross-departmental coordination costs. Both mechanisms alleviate the crowding losses associated with ambidextrous AI attention allocation.Heterogeneity analyses show that the crowding-out effect of ambidextrous AI attention allocation is concentrated primarily among high-tech firms and firms led by CEOs without information technology backgrounds. This study provides a dialectical assessment of the boundary conditions under which firms can simultaneously advance AI technology exploration and scenario-based application exploitation. It also offers decision-making references for firms with different resource endowments, technological foundations, and governance capabilities to reasonably arrange the pace and coordination mechanisms of AI technology exploration and scenario-based application exploitation.