This paper studies how decentralization affects institutional experimentation under uncertainty. Heterogeneous regions independently choose whether to incur a cost to adopt a reform of uncertain effectiveness. If the reform is effective, it facilitates the development of an innovation industry: each adopting region generates technological innovations at a fixed rate. Effective reforms may nevertheless be abandoned if innovation fails to occur for a sufficiently long period. Our main result is that the probability of such an institutional trap depends only on the incentives of the lowest-cost region; experimentation in all other regions is irrelevant for long-run outcomes. In other words, decentralization delegates the continuation of institutional experimentation to the lowest-cost region. This result implies a simple policy lesson: interventions aimed at sustaining long-run institutional change should concentrate resources on a single pioneering region. A Schumpeterian microfoundation of the model implies that institutional traps are more likely to occur in poorer economies and in economies with smaller innovation step sizes. We also show that decentralization is less effective when learning about institutional change is less transferable across regions.