In this paper we explore the use of prediction and delay-tolerance for offloading of large, socially recommended contents from 3G networks. We show that the two algorithms we design enable users to trade delay for energy and easily reduce battery consumption coming from 3G transfers of delay-tolerant content for 50%. We show that the real-time offloading requires Wi-Fi coverage of 3 to 4 times more cells, than our delay-tolerant algorithm. We find that both algorithms have lowest delay during the peak hours, when offloading is most needed. We also demonstrate how operators can benefit the collected data to offer cloud solutions, appealing to users (extending battery lifetime) and to the operators (load balancing between orthogonal technologies).