Carbon stocks in karst regions are critical components of the global carbon cycle. Understanding their spatiotemporal dynamics and underlying driving mechanisms is essential for assessing the effectiveness of ecological restoration efforts and improving regional carbon sink capacity. This study investigates the spatiotemporal evolution and drivers of ecosystem carbon stocks under different grades of karst rocky desertification (KRD) in Chongqing's karst region, China, from 2001 to 2020. To achieve this, the Integrated Valuation of Ecosystem Services and Tradeoffs (InVEST) model was used to estimate the carbon stocks, and the Sen's slope estimation and the Mann-Kendall test were used to investigate the trend and transition point. The Optimal Parameter-based GeoDetector (OPGD) model was then employed to identify dominant drivers and their interactive effects, while random forest-based Partial Dependence Plot (PDP) was applied to explore nonlinear responses. The results revealed that ecosystem carbon stocks initially declined slightly and then gradually recovered, increasing from 413.7 to 417.4 Tg. Concurrently, overall KRD conditions improved, with 2014 identified as a critical turning point. Spatially, carbon storage exhibited a significant negative correlation with KRD (Moran's I = -0.37 to -0.44, p < 0.01), indicating marked clustering of areas characterized by high desertification grades and low carbon storage. Regarding the driving factors, NDVI was the most influential in explaining the spatial heterogeneity in carbon storage (q = 0.64), and all significant interactive effects involved NDVI. Furthermore, PDP revealed that carbon storage levels tend to plateau when NDVI exceeds 0.835. These findings suggest that ecological restoration initiatives in Chongqing's karst region have effectively enhanced carbon sink capacity, primarily through vegetation recovery. Overall, this study offers critical insights for improving rocky desertification control and optimizing carbon sequestration strategies in fragile karst ecosystems.