As mobile data and video traffic is fast growing, the next-generation mobile communication (5G) networks are expected to further support 1000-fold more traffic than today's mobile networks. Meanwhile, the energy consumption of mobile networks has emerged as one of the major parts of the CO2 emission worldwide. As a result, the next-generation cellular networks have to be much more energy-efficient (green) than today in order to support the 1000-fold more traffic. For this purpose, the physical layer approaches only are not sufficient anymore and a system-level approach is the must. In this talk, we first claim that the existing cellular networks is spectrum-efficient but not energy-efficient because the static cellular architecture cannot adapt to the traffic dynamics by using real measurement data from commercial mobile networks. As cell size is getting even smaller, the cell traffic will get much more dynamic, both in temporal and spatial domains. In order to make the cellular architecture more adaptive to the traffic dynamics without worry about the coverage, we then propose a so-called hyper-cellular networks (HCN). The key idea behind is to separate the coverage for control signals from the coverage for data traffic, so that the data BSs (DBSs) can be turned off (into sleep mode) when traffic load is getting very low without concerning on the coverage of control signals. The network coverage will be guaranteed by the control BSs (CBSs) which in general cover a relatively wide area without sleeping. This can be considered as the further extension of the existing C-RAN concept and one of the key candidate technologies for 5G mobile communications systems. Some preliminary results have shown that this new paradigm has a great potential in the capacity enhancement and energy savings.