Reliable and efficient operation of power networks is of paramount importance. In this paper we explore communication architectures that leverage Phasor Measurement Units (PMUs) and fusion centers to enable robust real-time monitoring of power networks over wide areas. We examine scalability issues in data aggregation for a power network in which PMUs are attached to every bus. We argue that in power networks, the data recorded by the PMUs are spatially correlated and therefore the data admit a sparse representation in a given basis. Furthermore, the number of bits required to reconstruct the PMU data to within a given accuracy at a remote location grows sublinearly with the density of the power network. Our results are: i) the PMUs should not transmit their data to the fusion center independently or asynchronously - an intermediate data aggregation step is beneficial, and ii) if we perform data aggregation then we can add more PMUs to the network to achieve finer network monitoring without causing network congestion.