Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for reducing latency in presenting digital videos. In one aspect, a system includes a front-end server that provides videos and local servers that each include a cache for storing videos that have been partially presented at client devices located in a same geographic location as the local server. The system includes a back-end server that selects a video in response to a content request received from a client device located in the same geographic location as a local server. The back-end server provides, to the client device, at least a portion of the video and determines that less than all of the video was presented at the client device. The back-end server reduces latency in presenting additional portions of the video at the client device by storing the video in the cache of the local server.