The elevation gradient has long been known to be vital in shaping the structure and function of terrestrial ecosystems, but little is known about the elevation-dependent pattern of net CO2 uptake, denoted by net ecosystem productivity (NEP). Here, by analyzing data from 203 eddy covariance sites across China, we report a negative linear elevation-dependent pattern of NEP, collectively shaped by varying hydrothermal factors, nutrient supply, and ecosystem types. Furthermore, the NEP shows a higher temperature sensitivity in high-elevation environments (3000-5000 m) compared with the lower-elevation environments (<3000 m). Model ensemble and satellite-based observations consistently reveal more rapid relative changes in NEP in high-elevation environments during the last four decades. Machine learning also predicts a stronger relative increase in high-elevation environments, whereas less change is expected at lower elevations. We therefore conclude a varying elevation-dependent pattern of the NEP of terrestrial ecosystems in China, although there is significant uncertainty involved.