Proximity-based content caching and distribution in wireless networks has been identified as a promising traffic offloading solution for improving the capacity and the quality of experience (QoE) by exploiting content popularity and spatiotemporal request correlation. In this paper, we address the following question: where should popular content be cached in a wireless network? For that, we model a wireless cellular network using stochastic geometry and analyze the performance of two network architectures, namely caching at the mobile device allowing device-to-device (D2D) connectivity and local caching at the radio access network edge (small cells). We provide analytical and numerical results to compare their performance in terms of the cache hit probability, the density of cache-served requests and average power consumption. Our results reveal that the performance of cache-enabled networks with either D2D caching or small cell caching heavily depends on the user density and the content popularity distribution.