The literature on war economies argues that prolonged civil wars have an economic logic: certain groups may obtain material gains from committing acts of violence and hence resist peacebuilding efforts. Objective tests of these predictions have so far been limited, as corruption and conflict prevent the collection of reliable economic data on the ground. Remote sensing and Geographic Information Science techniques enable us to overcome these problems of terrestrial data collection. Electricity consumption manifested as night-time light emissions recorded in satellite images is proposed as a proxy for changes in disposable income in Somalia's cities. The nightlight images provide striking illustrations of economic decline and recovery and clearly show the contrast between the stable regions of Northern Somalia and the chaos and anarchy of Southern Somalia. Based on geospatial analyses of settlement patterns in Somali cities, we argue that specific metrics of light output can be used to proxy for the incomes of different social groups. We use geo-coded conflict event data to analyze the economic impact of conflict on local light output and therefore incomes. We find a significant peace dividend for poorer households located at the margins of cities, which benefit both from local stability and more peaceful conditions in the country as a whole. By contrast, the central business districts are relatively well insulated from the effects of local conflict, and violence in Mogadishu has positive effects on light output from cities where humanitarian aid agencies are located. Future peace initiatives need to confront these economic incentives for continued conflict and state failure in Somalia.