Precipitation variability induced by climate change has a profound impact on soil carbon dynamics in dryland agroecosystems. To elucidate mechanisms of carbon sequestration in rhizosphere and non-rhizosphere soils under precipitation variability, we conducted an eight-year precipitation manipulation experiment (-50 %,-25 %, ambient, +25 %, +50 %) in croplands of China's Loess Plateau. This research provided a comprehensive assessment of carbon fraction dynamics and the driving factors behind these mechanisms. Our results demonstrate that increased precipitation significantly enhanced soil organic carbon (SOC) in both rhizosphere (+20.5 %) and non-rhizosphere (+22.5 %) soils, although carbon accrual patterns exhibited spatial divergence. Rhizosphere carbon accumulation primarily stemmed from mineral-associated organic carbon (MAOC) (MAOC/ SOC ratio increased from 80.3 % to 88.8 %), whereas non-rhizosphere soil relied on particulate organic carbon (POC) contributions (POC/SOC ratio rose from 12.5 % to 17.5 %). Furthermore, microbial communities displayed functional group decoupling. Precipitation increase shifted bacteria towards r-strategies in both compartments (evidenced by declining oligo/copiotroph ratios and increased rrn copy numbers), while fungal life strategies remained unchanged. Bacterial alpha-diversity increased in both zones, but the response of fungal alpha-diversity diverged, increasing in rhizosphere soil but decreasing in non-rhizosphere soil. Partial least squares path modelling (PLS-PM) revealed compartment-specific sequestration mechanisms. In rhizosphere soil, precipitation stimulated root carbon secretion (70.7 %) and increased the exudate C/N ratio (119.8 %), thereby promoting the proliferation of r-strategists that enhanced MAOC formation and SOC accumulation. In contrast, non-rhizosphere soil depended on precipitation-mediated modulation of microbial diversity and protection of aggregates to facilitate POC formation and SOC accrual. This study proposes divergent carbon sequestration models for the rhizosphere and non-rhizosphere under variable precipitation, providing critical insights for predicting carbon trajectories in climate-sensitive agricultural ecosystems.