Using the standard two‐stage game of process R&D and Cournot competition with R&D output spillovers, this paper provides a thorough second‐best welfare analysis. The planner's solution is compared with the standard noncooperative scenario, the R&D cartel, the cartelized research joint venture, and the social research joint venture solution in terms of propensities for R&D as well as welfare levels. The main result is that, when spillovers are not too high, a cartelized joint venture unexpectedly outperforms the planner's solution in terms of propensities for R&D and resulting welfare level, though it is the only market scenario to do so. We also assess the performance of a social joint venture, relative to all the well‐known scenarios for the organization of R&D. Finally, we observe that the gap between market outcomes and planner's solutions, in terms of welfare levels, increases as R&D becomes less appropriable.