Understanding the effects of drought on the coupling between ecosystem carbon uptake and water loss is critical for predicting water sustainability in regions with high elevation and limited water. In this study, the spatiotemporal dynamics, drought responses, and dominant drivers of summer water use efficiency (WUE) across the Qinghai-Tibet Plateau (QTP) from 2001 to 2020 were analyzed using three complementary metrics: ecosystem WUE (EWUE), transpiration-based WUE (TWUE), and intrinsic WUE (IWUE). Remote sensing products were combined with machine learning attribution and causal network analysis to disentangle hydrological and physiological controls. The results showed that IWUE exhibited the most spatially extensive and consistent increase, with significant positive trends across 62.4% of the QTP, whereas EWUE increases were weaker and TWUE trends were spatially heterogeneous. During soil moisture droughts, IWUE increased by 21%, while TWUE declined by 6% and EWUE showed limited change. Causal analyses indicated that EWUE and TWUE were strongly associated with evapotranspiration and canopy transpiration, respectively, whereas IWUE was mainly related to canopy conductance and atmospheric demand across more than 60% of the region. The strength of stomatal control on IWUE substantially exceeded that of carbon assimilation, indicating a partial decoupling induced by drought in which transpiration was suppressed more strongly than photosynthesis. These findings demonstrate that rising WUE on the QTP primarily reflects intensified physiological regulation rather than enhanced ecosystem productivity. Explicit representation of stomatal drought feedbacks and atmospheric demand is therefore essential for improving hydrological and Earth system model projections in high-elevation drylands.