Vegetation carbon stock (C veg) in global forests, which is important for C cycle-climate feedbacks, commonly increases with forest age. Due to the allometric growth of plants, the nonlinear increase in C veg with woody fraction (f w) is expected across space. However, it remains unclear whether such a nonlinear relationship between C veg and f w can be constrained by observations and further used to benchmark Earth system models (ESMs). Here, based on the in situ measurements at 1,145 forest sites, we found that the nonlinear relationship between C veg and f w followed an exponential equation (i.e., C veg ¼ be a • f w). Then, we showed that such an exponential dependence of C veg on f w also exists in ESMs of CMIP5 and CMIP6 (all P < 0.01), even though age-dependent processes have not been incorporated in most models. However, the exponential C veg-f w relationship varied greatly among the models, and the coefficient b was systematically lower in the ESMs (0.08 ± 0.11; mean ± SD) than the observations (0.28). Based on a compiled forest age data set, we further found that the observed nonlinear increase of C veg with forest age across the Northern Hemisphere (>30°N) was not captured by ESMs. These findings reveal a high disagreement on the spatially nonlinear relationship between vegetation carbon stock and woody fraction in current ESMs. The exponential relationship based on observations provides one useful benchmark for ESMs when they implement the age-dependent processes in the future. Plain Language Summary Forest age plays an important role in vegetation carbon stock (C veg) predictions. This study detects a nonlinear increase of C veg with woody fraction (f w) in aging forests across 1,145 in situ observations. The nonlinear C veg-f w relationship was then used to benchmark the age impacts on C veg predictions in Earth system models (ESMs). Combined with a global forest age data set, we show that current ESMs divergently represent the relationship between C veg and forest age. Our study suggests that the C veg-f w relationship could be one useful benchmark for evaluating ESMs.