The production of knowledge relies on collaborations between researchers. However, we do not know to what extent policies may stimulate these interactions. In this paper we show how a large-scale public “research cluster” policy in France, which funds local communities of researchers working on a common theme, affects the organization of research. Relying on an identification strategy based on grades awarded by reviewers, we show that members of financed clusters increase their collaborations with other cluster members by up to 30%. Paradoxically, researchers not at the core of the cluster topic benefit the most from the policy.