Data communications infrastructure will play an important role to transfer various information in smart grid. In this paper, we consider the reliability of the smart grid data communications infrastructure and its impact on the power consumption and supply optimizations. For optimizing the power consumption, we consider a deferrable load scheduling method which is modeled by using a constrained Markov decision process (CMDP) model, taking into account the unavailability of the home area network (HAN) and neighborhood area network (NAN) gateways. For optimizing the power supply, we consider an economic dispatch method which is modeled by using stochastic programming (SP), taking into account the unavailability of the exact power demand and supply information. The power consumption and supply costs are analyzed. In addition, we show how these costs can be reduced through the deployment of component redundancy in the smart grid data communications infrastructure.