A network node jointly precodes multi-user (MU) multiple-input multiple-output (MIMO) transmissions simultaneously sent from geographically distributed base stations to a plurality of mobile terminals over associated downlink MU-MIMO channels. The node receives feedback that describes statistics of the downlink MU-MIMO channels, including channel mean and covariance. The node then computes, based on the channel means and covariances, uplink input covariances for the mobile terminals that would collectively maximize a first or second-order approximation of the ergodic capacity of dual uplink MU-MIMO channels, subject to a global transmit power constraint that comprises the sum of individual transmit power constraints for the base stations. Notably, the node adjusts the uplink input covariances as needed to satisfy the individual transmit power constraints for the base stations, maps the uplink input covariances to corresponding downlink input covariances, and jointly precodes MU-MIMO transmissions sent over the downlink MU-MIMO channels based on those downlink input covariances.