AbstractNet primary productivity (NPP) in global grasslands is a critical component of terrestrial carbon cycling and the primary source of food for herbivores. However, the size and spatial distribution of NPP across global grasslands remain unclear, especially for belowground NPP (BNPP), which limits our understanding of above‐ and belowground carbon cycling and the assessments of herbivore food security. Here, we compiled a comprehensive grassland NPP database with 1,467 field measurements to estimate the spatial distributions of aboveground NPP (ANPP), BNPP, total NPP (TNPP), and the fraction of BNPP (f_BNPP) using the random forest (RF) model. The global mean grassland ANPP, BNPP, TNPP, and f_BNPP were 433 ± 31 g m⁻² yr⁻¹, 593 ± 47 g m⁻² yr⁻¹, 979 ± 78 g m⁻² yr⁻¹, and 0.54 ± 0.02, respectively. The total ANPP, BNPP, and TNPP over global grasslands were 6.84 ± 0.49, 9.36 ± 0.74, and 15.46 ± 1.23 Pg C yr⁻¹, respectively. ANPP, BNPP, and TNPP exhibited decreasing trends from low latitudes toward the poles. The spatial pattern of f_BNPP was almost opposite to that of ANPP. Climate was a major determinant in shaping the spatial distribution of ANPP and TNPP, while soil and vegetation had significant impacts on that of BNPP and f_BNPP. Our findings suggest that field data‐driven estimation of NPP using the RF model could be a useful approach for obtaining spatially explicit NPP of grasslands, particularly BNPP products, and that more attention should be paid on belowground and non‐climatic factors to better assess the carbon cycle in global grasslands.