Terrestrial ecosystems mitigate CO2 accumulation in the atmosphere by annually absorbing ~30% of anthropogenic emissions. The degrees to which CO2 and climate drive this absorption are uncertain, which presents a challenge for future planning around carbon mitigation scenarios. To reduce this knowledge gap, we use a Bayesian model–data integration framework (CARDAMOM) to build and analyze a global terrestrial biosphere reanalysis that optimally reconciles multiple lines of Earth Observations with mechanistic model processes. The Earth observations informing the model dynamics include satellite- and inventory-based constraints on distributions and change in terrestrial C storage (e.g., live biomass, soil organic C, and net biosphere exchange of CO2) and mechanisms of that change (e.g., photosynthesis, deforestation, water storage anomalies, or fire). We find that the impact of 2001–2021’s atmospheric CO2 increase on terrestrial C storage (+38 PgC) opposes and far outweighs the impact of climate trends over over the same period ($-$8.2 PgC). Globally, CO2-induced carbon gains occurred primarily in living biomass pools, while climate-induced losses occurred primarily in dead organic C pools, but relative gains and losses vary regionally. The fact that live biomass and dead organic C exhibit distinct and often opposing responses indicates that mechanistically resolving ecosystem function underlying the terrestrial C cycle’s emergent behavior is crucial for estimating the strength and resilience of the land C sink over the coming decades.