We consider the content delivery problem in a fading multi-input single-output channel with cache-aided users. We are interested in the scalability of the equivalent content delivery rate when the number of users, K, is large. Analytical results show that, using coded caching and wireless multicasting, without channel state information at the transmitter, linear scaling of the content delivery rate with respect to K can be achieved in some different ways. First, if the multicast transmission spans over L independent sub-channels, e.g., in quasi-static fading if L = 1, and in block fading or multi-carrier systems if L 1, linear scaling can be obtained, when the product of the number of transmit antennas and the number of sub-channels scales logarithmically with K. Second, even with a fixed number of antennas, we can achieve the linear scaling with a threshold-based user selection requiring only one-bit feedbacks from the users. When CSIT is available, we propose a mixed strategy that combines spatial multiplexing and multicasting. Numerical results show that, by optimizing the power split between spatial multiplexing and multicasting, we can achieve a significant gain of the content delivery rate with moderate cache size.