While power outages caused by tropical cyclones (TCs) already pose a great threat to coastal communities, how—and why—these risks will change in a warming climate is not well understood. To address this need, we develop a robust machine learning model to capture TC-induced power outage risk, which we call the Electric Power Outages from Cyclone Hazards (EPOCH) model. We then apply the EPOCH model to 900,000 synthetic TCs, downscaled from simulated historical and future climate conditions under a strong warming scenario. The results indicate that outage risk in the United States and Puerto Rico is expected to increase broadly by the end of the century, with some states seeing increases of 50% or more compared to historical levels. Feature importance analysis reveals that rising TC rainfall rates will play an increasingly important role in power outage risk as the climate changes, explaining more than half of the projected change in risk in some regions. Our findings can help guide decision-makers in future outage risk investment and mitigation strategies.