<p>In the digital paradigm shift, university–industry (U–I) collaboration plays a crucial role in the innovation ecosystem for knowledge generation and commercialization. Firms, universities, and other organizations are increasingly engaging in cross-geographical collaborations, forming innovation communities that serve as meso-level innovation units. Although cross-community collaboration offers a promising solution to exchange knowledge and overcome path dependence and technology lock-in during this transition, it is notably underexplored in existing research. Therefore, this study aims to unveil the impact of cross-community collaboration on innovation performance. We propose a framework of innovation communities in a dual-layer innovation ecosystem, including technology and business layers, based on which we put forward the conceptual model of the effects of intensity of cross-community collaboration and breadth of cross-community collaboration on innovation performance. Employing large-scale network analysis and negative binomial regression, we conduct an empirical investigation in the Chinese Computer Numerical Control machine tool industry. Our findings suggest that intensive and broad cross-community collaborations in either the technology or business layer are conducive to innovation. However, simultaneously engaging in intensive cross-community collaborations in both layers may be detrimental. This research highlights strategies for enhancing innovation through cross-community collaboration in the digital manufacturing era.</p>