To bridge the persistent “valley of death” in medical innovation, this study highlights the role of research hospitals in knowledge production and calls for a collaborative, knowledge-sharing innovation ecosystem. Drawing on theories of network embeddedness and innovation ecosystems, it develops a tripartite evolutionary game model involving research hospitals, government, and medical technology enterprises, inspired by the institutional logic of translational medicine centers. Parameter settings are informed by expert interviews and simulation is used to explore the impact of initial intentions, governmental incentives, and self-organizing constraints. Results reveal four stable evolutionary paths: resource-convergent co-construction, policy-driven government leadership, actor-dispersed disengagement, and locally stable self-organization. Research hospitals prefer long-term gains through deeper network embedding; enterprises gradually move from relational to structural embeddedness, thus becoming stronger endogenous drivers than government support. Meanwhile, self-organized ecosystems impose internal constraints through sanctions against opportunism, allowing government to shift from direct intervention to supervisory governance. The study proposes strategic actions to support the coordinated evolution of translational medical innovation ecosystems.