This paper seeks to answer a simple but fundamental question: what role can non-orthogonal multiple access (NOMA) play in massive multi-in multi-out (MIMO)? It is well established that power-domain NOMA schemes can outperform conventional orthogonal multiple access schemes in cellular networks. However, this fact does not imply that NOMA is the most efficient way to communicate in massive MIMO setups, where the base stations have many more antennas than there are users in the cell. These setups are becoming the norm in future networks and are usually studied by assuming spatial multiplexing of the users using linear multi-user beamforming. To answer the above-mentioned question, we analyze and compare the performance achieved by NOMA and multi-user beamforming in both non-line-of-sight and line-of-sight scenarios. We reveal that the latter scheme gives the highestaverage sum rate in massive MIMO setups. We also identify specific cases where NOMA is the better choice in massive MIMO and explain how NOMA plays an essential role in creating a hybrid of NOMA and multi-user beamforming that is shown to perform better than two standalone schemes do.