Contingent pathways and dynamic mechanisms of collaborative innovation: a multiple-case study of China’s AI foundation model firms

How do Chinese AI foundation model firms organize collaborative innovation in the face of intensifying geopolitical contestation and technological constraints? This paper addresses this issue by conducting a comparative multi-case study of three representative firms from China’s AI sector: DeepSeek, Moonshot AI and ByteDance, with the aim of explaining how Chinese AI foundation model firms form and implement collaborative innovation pathways in response to geopolitical and technological constraints. Our findings reveal that pathway choice is shaped by the interaction of three contingent conditions: resource endowments, strategic orientations, and geopolitical and technological constraints. Under these conditions, firms adopt three distinct pathways: community-driven open-source collaboration, user-centric application enablement, and platform-led industrial chain integration. Despite their organisational differences, these pathways operate through a common reinforcing mechanism consisting of search expansion through collaborative exposure, signal conversion and cumulative learning, and legitimation and ecosystem reinforcement. Through this mechanism, collaborative innovation generates three interrelated outcomes: technological iteration, industrial resilience, and ecosystem construction. This study contributes to research on collaborative innovation and innovation ecosystems by demonstrating how latecomer firms can sustain innovation in the face of geopolitical constraints. It also has implications for AI industry policy and firm strategy in emerging economies subject to technological containment.