Understanding how planting density regulates water-use strategies is crucial for sustainable afforestation in arid regions. In the Kubuqi Desert, stable isotope (delta 2H, delta 18O) analysis, Bayesian mixing models (MixSIAR), and machine learning were integrated with structural equation modelling (SEM) to unravel the mechanisms driving water uptake in Caragana korshinskii plantations across a density gradient (low, 822-944; medium, 1600-1733; high, 2667-2889 plants & sdot;hm-2). A density-dependent shift in water-use strategies was identified, from predominantly shallow-soil water uptake (0-40 cm) in low-density stands, through flexible source switching in mediumdensity stands, to a heavy, consistent reliance on deep soil water (80-100 cm) in high-density stands. This strategic shift was mechanistically explained by SEM, which revealed that increasing density intensified root competition, thereby altering the primary drivers of root development. In low-density stands, root growth was nutrient-driven, supporting efficient water uptake from moist upper layers. In stark contrast, under high-density stress, root biomass became primarily governed by soil moisture scarcity, forcing a shift towards deeper, yet less efficient, water exploitation. Consequently, isotopic signatures and plant water lines indicated significantly stronger evaporation and physiological fractionation in denser stands. These findings demonstrate that afforestation density controls water-use sustainability by modulating the fundamental 'soil-root-water' nexus. Optimising stand density to maintain a nutrient-driven, efficient water-use strategy is therefore critical for the long-term viability of desert plantations.