Digital television channels are preemptively cached based on a modeling of a user to reduce delays while switching channels. A current television channel is selected using a first tuner. A future television channel selection of the user is then predicted based on a modeling of the user. The recorded content of the predicted future television channel is preemptively cached using a second tuner. A buffer of the recorded content of the predicted future television channel is displayed when the user switches from the current television channel to the predicted future television channel. The modeling of the user is updated and stored in storage.