We introduce a model of strategic experimentation on social networks where forward-looking agents learn from their own and neighbors' successes. In equilibrium, a private discovery phase is followed by a social diusion phase; the anticipation of future social information crowds out agents' own experimentation. We rst study tree-like networks, characterize learning dynamics via ODEs, and draw tight comparisons between directed, undirected and clustered networks. We then turn to general large random networks, with density ranging from sparse trees to dense cliques. We show that information aggregates if network density is low, which motivates private discovery. In contrast, welfare attains a second-best benchmark if network density is intermediate, which allows for a quicker diusion of discoveries.