How would a cellular network designed for high energy efficiency look like? To answer this fundamental question, we model cellular networks using stochastic geometry and optimize the energy efficiency with respect to the density of base stations, the number of antennas and users per cell, the transmit power levels, and the pilot reuse. The highest efficiency is neither achieved by a pure small-cell approach, nor by a pure massive MIMO solution. Interestingly, it is the combination of these approaches that provides the highest energy efficiency; small cells contributes by reducing the propagation losses while massive MIMO enables multiplexing of users with controlled interference.