Almost two billion people depend on freshwater provided by the Asian water towers, yet long-term runoff estimation is challenging in this high-mountain region with a harsh environment and scarce observations. Most hydrologic models rely on observed runoff for calibration, and have limited applicability in the poorly gauged Asian water towers. To overcome such limitations, here we propose a novel data-driven model, SM2R (Soil Moisture to Runoff), to simulate monthly runoff based on soil moisture dynamics using reanalysis forcing data. The SM2R model was applied and examined in 20 drainage basins across seven Asian water towers during the past four decades of 1981‒2020. Without invoking any observations for calibration, the overall good performance of SM2R-derived runoff (correlation coefficient ≥ 0.74 and normalized root mean square error ≤ 0.22 compared to observed runoff at 20 gauges) suggests considerable potential for runoff simulation in poorly gauged basins. Even though the SM2R model is forced by ERA5 reanalysis data, it largely outperforms the ERA5-estimated runoff across the seven Asian water towers, particularly in basins with widely distributed glaciers and frozen soil. The SM2R approach is highly promising for constraining hydrologic variables from soil moisture information. Our results provide valuable insights for not only long-term runoff estimation over key Asian basins, but also understanding hydrologic processes across poorly gauged regions globally.