The size of the terrestrial sink remains uncertain. This uncertainty presents a challenge for projecting future climate–carbon cycle feedbacks1, 2, 3, 4. Terrestrial carbon storage is dependent on the availability of nitrogen for plant growth5, 6, 7, 8, and nitrogen limitation is increasingly included in global models9, 10, 11. Widespread phosphorus limitation in terrestrial ecosystems12 may also strongly regulate the global carbon cycle13, 14, 15, but explicit considerations of phosphorus limitation in global models are uncommon16. Here we use global state-of-the-art coupled carbon–climate model projections of terrestrial net primary productivity and carbon storage from 1860–2100; estimates of annual new nutrient inputs from deposition, nitrogen fixation, and weathering; and estimates of carbon allocation and stoichiometry to evaluate how simulated CO2 fertilization effects could be constrained by nutrient availability. We find that the nutrients required for the projected increases in net primary productivity greatly exceed estimated nutrient supply rates, suggesting that projected productivity increases may be unrealistically high. Accounting for nitrogen and nitrogen–phosphorus limitation lowers projected end-of-century estimates of net primary productivity by 19% and 25%, respectively, and turns the land surface into a net source of CO2 by 2100. We conclude that potential effects of nutrient limitation must be considered in estimates of the terrestrial carbon sink strength through the twenty-first century.