Tropical forests play a crucial role in the global carbon cycle, accounting for one-third of the global NPP and containing about 25% of global vegetation biomass and soil carbon. This is particularly true for tropical forests in the Amazon region, as these comprise approximately 50% of the world’s tropical forests. It is therefore important for us to understand and represent the processes that determine the fluxes and storage of carbon in these forests. In this study, we show that the implementation of phosphorus (P) cycle and P limitation in the version 1 of the E3SM land model (ELM v1) improves the simulated spatial pattern of wood productivity. The P-enabled ELM v1 is able to capture the declining west-to-east gradient of productivity, consistent with field observations. We also show that by improving the representation of mortality processes using soils data, ELMv1 is able to reproduce the observed spatial pattern of above-ground biomass. Our model simulations show that the consideration of P availability lead to a smaller carbon sink associated with CO2 fertilization effect, and lower carbon emissions due to land use and land cover change (LULCC). Our simulations suggest P limitation would significantly reduce the carbon sink associated with CO2 fertilization effects through the twenty-first century. We conclude that P cycle dynamics affect both sources and sinks of carbon in the Amazon region, and the effects of P limitation would become increasingly important as CO2 increases. Therefore, P limitation must be considered for projecting future carbon dynamics in tropical ecosystems.