With more and more activities taking place online, concern over the environmental impact of digital services has drawn attention to the energy intensity of the network. Estimating the network energy intensity has been the subject of research for some time but results have differed widely, thus weakening the robustness of any conclusions drawn from assessments. A review of past studies shows two separate communities at work, applying different methods and assumptions. In this article we consider the approaches of top-down and bottom-up modeling. Top-down models have in the past usually given higher estimates of energy intensity than bottom-up models. We find that among the main reasons for the difference are varying system boundaries, and assumptions on the number and energy efficiency of routers and optical transmission equipment. Through application of consistent system boundaries around the metro and core networks and excluding access networks and customer equipment, we reduce the difference between the energy intensity estimates of the alternative approaches. Additionally, we review the varying assumptions in existing bottom-up models and combine them in a meta-model. Through Monte Carlo simulation over the distributions behind the varying assumptions we provide a more robust estimate of approximate energy efficiency for networks of 0.02 kWh/Gbyte that can be used in the environmental impact assessment of digital services.