This paper studies beamforming design problems for multi-cell multi-user downlink networks with simultaneous wireless information and power transfer. In this system, base stations (BSs) concurrently transfer information and energy to multiple single-antenna information decoding (ID) and energy harvesting (EH) users. We aim to maximize energy harvesting efficiency (EHE), which is defined as the ratio of the harvested energy at the EH users to the amount of energy consumption at the BSs, while guaranteeing quality-of-service constraint for each ID user. First, for the centralized case where global channel state information (CSI) is available at all BSs, we propose a centralized beamforming method based on the semi-definite relaxation techniques. Next, in order to reduce the backhaul signaling overhead, a decentralized algorithm is presented where each BS computes its beamforming vector by only using local CSI. Simulation results show that the proposed algorithm offers a significant EHE performance gain over conventional schemes.