This paper presents a unified, coordinated, thermal-computational approach to the data center energy management problem. A data center is modeled as two coupled net-works: a computational and a thermal network. The first network describes the computation performance of the data center, while the latter describes temperature evolution of the devices housed in the data center, based on their electrical power consumption and cooling. Server power states influence both networks. We formulate the energy control problem as a Markov decision process (MDP) and compare through an example, the performance of a controller derived using the proposed unified thermal-computational approach against a controller that disregard the coupling between the two networks. Simulation results suggest further research directions.