The evolution of the Amazon forest is tightly coupled to its terrestrial water balance (evapotranspiration minus precipitation, or ET-P), as an increase in ET-P reduces soil moisture, increasing water stress. However, large differences of ~ 50% between current monthly estimates of ET-P make it challenging to confidently quantify its spatio-temporal distribution and evolution. Here, we show that new satellite observations of the HDO/H2O ratio of water vapor, spanning 2003 to 2020, constrain estimates of the Amazon water balance with monthly precision of ~ 20%. The HDO/H2O ratio of water vapor is sensitive to the difference between ET and P, rather than to either flux alone, because lighter isotopes preferentially evaporate and heavier isotopes preferentially condense. Consequently, variable bias and sensitivity errors that result from combining different ET and precipitation products are minimized with this proxy. Our analysis demonstrates these data can quantify the spatial patterns of Amazon water balance from monthly to interannual time scales.