This paper investigates the evolving structure of the inter-city knowledge flow network underlying the full life cycle of mRNA vaccine innovation, from early basic research in the mid-1980s to the commercialisation of COVID-19 vaccines in 2020. Drawing on 24,442 patents and 4,412 inventors across 46 countries and 1,281 cities, we construct a directed temporal network and analyse its structural evolution relying on the innovation life cycle model. Our results show that concentration is not an enduring structural property but an emergent outcome: the initial phase of the innovation process is characterised by dispersed, non-hierarchical flows and weakening hub dominance, while the second phase consolidates around a stable hierarchy of dominant cities. These findings suggest that innovation policy should account for the phase-dependence of network structure, avoiding premature concentration of resources in the early stages of technological development.