The booming literature on civil wars has paid considerably less attention to the processes through which civil wars come to an end. This paper tries to fill the gap by theorizing the conditions under which incumbent regimes lose wars and collapse. We discuss three arguments. First, the odds of collapse should increase as the incumbent´s military capabilities decline. A clear manifestation of this process occurs when foreign patrons withdraw their material support for the regime. Second, we explore whether political linkages between central officials and local rulers (or warlords) have an impact on regime collapse. Counterintuitively, we hypothesize that autonomous regional actors face a higher cost of rebel takeover, so they will fight longer for their survival. Finally, strategic misfires such as wrong-headed prisoner swaps change the frontline momentum and inadvertently help the rebels to consolidate their gains. We empirically explore these arguments by looking at the case of the collapse of the NATO-backed regime in Afghanistan. We match data on rebel control and violence with indicators on socio-economic and political factors at the district level from 2014 to August 2021. Survival analyses show that the fall of the Afghan regime was driven by a combination of strategic misfires, loss of foreign support, and lack of robust alliances with local warlords. These findings strikingly portray the risks of enforcing power centralization schemes as the main instrument for regime consolidation when local armed actors are the first line of defence against rebel actors with popular support.