The impact of AI technology collaboration network embeddedness on urban digital innovation resilience: A resource orchestration theory perspective

In the digital economy era, constructing a robust safety barrier for digital innovation systems is imperative. Artificial Intelligence (AI) serves as a foundational pillar in this context; however, does embedding in the AI technology collaboration network necessarily act as a “power source” for enhancing urban digital innovation resilience? To address this, drawing on Resource Orchestration Theory, we construct an analytical framework of “network resource acquisition – digital capability reconfiguration – innovation resilience escalation.” Through this framework, it conducts an in-depth analysis of the empowerment mechanism that the embedding depth of the AI technology collaboration network exerts on urban digital innovation resilience. The findings reveal an inverted U-shaped relationship between the embedding depth of the AI technology collaboration network and urban digital innovation resilience. The R&D capability and commercialization capability of key digital technologies play a mediating role during the moderate embedding stage. However, in scenarios of over-embeddedness, the transmission mechanisms of both capabilities are rendered ineffective. Notably, this inverted U-shaped relationship exhibits multi-dimensional context-dependency. Heterogeneity analysis confirms that network embedding depth exerts a more pronounced promotional effect on peripheral-tier cities, whereas poor-quality public data openness significantly exacerbates the suppressive effect of over-embeddedness. Furthermore, the moderating effect indicates that urban absorptive capacity acts as a crucial dynamic regulatory valve. Once it crosses a specific critical threshold, it drives the original inverted U-shaped relationship to undergo a structural shape-flip into a U-shaped pattern. Accordingly, this study proposes relevant policy implications focusing on four aspects: evading network traps, dismantling data constraints, consolidating the dual-capability pillars, and constructing risk-buffering mechanisms.