Smart meters enable a fine-granular monitoring of power consumption and distributed power production in costumers' premises, which are used to predict the power requirements for the near future. The goals are to offer more security of supply as well as to minimize the power requirement estimation errors. However, to benefit from this information, the communication infrastructure that transmits the energy-related data needs to fulfill stringent requirements with respect to dependability, while remaining monetarily feasible. This paper discusses the usage of network function virtualization (NFV) technologies and constructs a virtual advanced metering infrastructure (AMI) network to transmit energy-related information in a dependable and cost-effective way. After the discussion of dependability requirements of AMI and the shortcomings of current approaches, the reliability and availability of a new architecture based on NFV is analyzed using analysis. Finally, a cost model is developed to compare the Virtual Network Function approach to current AMIs.