Massive MIMO is a promising technique to increase the spectral efficiency of cellular networks, by deploying antenna arrays with hundreds or thousands of active elements at the base stations and performing coherent beamforming. A common rule-of-thumb is that these systems should have an order of magnitude more antennas, $N$, than scheduled users, $K$, because the users' channels are then likely to be quasi-orthogonal. However, it has not been proved that this rule-of-thumb actually maximizes the spectral efficiency. In this paper, we analyze how the optimal number of scheduled users, $K^\star$, depends on $N$ and other system parameters. The value of $K^\star$ in the large-$N$ regime is derived in closed form, while simulations are used to show what happens at finite $N$, in different interference scenarios, and for different beamforming.