A large-scale clustered massive MIMO network is proposed for improving the spectral efficiency of the nextgeneration wireless infrastructure by maximizing its sum-rate. Our solution combines the advantages of the centralized processing architecture and massive MIMO. Explicitly, the network is divided into multiple clusters; each cluster is handled by a centralized processing unit, which connects to a certain number of massive MIMO-aided BSs, where only limited information is exchanged among the clusters; each user of a cluster can be served by several nearby BSs in a user-centric way.We analyze the maximum sum-rate of the network with multiple antennas at BSs and UEs, relying on the optimal transmit precoder matrix of each BS configured for each user, and on the optimal frequency-domain power sharing scheme of each cluster. The optimizations are conceived for multiple coordination schemes that were widely studied in literature, namely the coherent-joint-transmission (CJT) scheme, the noncoherent-joint-transmission (NCJT) scheme and the coordinatedbeamfoming/scheduling (CBF/CS) scheme. Our simulations show that the optimal CJT achieves 2.2 -4.5 times higher average sumrate than its non-cooperative massive MIMO network counterpart, while the optimal NCJT and the optimal CBF/CS achieve at most a factor 1.3 average sum-rate gain. The popular signalto-leakage-and-noise-ratio (SLNR) scheme is also extended to the multi-antenna UE scenario and achieves a factor 1.1 -1.2 gain. Index Terms-Large-scale clustered MIMO network, cell free MIMO, weighted sum rate maximization, optimal precoding matrix, optimal power sharing.