Energy access in conflict-affected states carries a cost premium ― yet there is little evidence of its magnitude and whether it can, at least partially, be mitigated. Analyzing 923 World Bank?financed solar photovoltaic installations across the Republic of Yemen between 2019 and 2025, this analysis finds that project-level learning systematically reduces conflict-linked costs, including in the most volatile regions. The aggregate price decomposition attributes 16.8 percentage points of cost reduction to project-level factors. More generally, a Shapley Machine Learning decomposition of project-level cost variation confirms that project-level learning is the most powerful predictor, explaining 45.2 percent of cost variation. Critically, the conflict-cost relationship evolves over successive procurement cycles: early packages exhibit a significant positive conflict premium, which is gradually mitigated. Cost trajectories converge regardless of whether governorates experienced escalating or de-escalating violence, confirming that learning operates independently of security trends.