Atmospheric CO2 has more than doubled since the last glacial maximum (LGM) and could double again within this century, largely due to anthropogenic activity. It has been suggested that low [CO2] contributed to reduced tree cover in savanna and grassland biomes at LGM, and that increasing [CO2] over the last century promoted increases in woody plants in these ecosystems over the past few decades. Despite the implications of this idea for understanding global carbon cycle dynamics and key global role of the savanna biome, there are still very few experimental studies quantifying the effects of CO2 on tree growth and demography in savannas and grasslands. In this paper we present photosynthetic, growth and carbon allocation responses of African savanna trees (Acacia karroo and Acacia nilotica) and a C4 grass, Themeda triandra, exposed to a gradient of CO2 concentrations from 180 (typical of LGM) to 1000 µmol mol−1 in open-top chambers in a glasshouse as a first empirical test of this idea. Photosynthesis, total stem length, total stem diameter, shoot dry weight and root dry weight of the acacias increased significantly across the CO2 gradient, saturating at higher CO2 concentrations. After clipping to simulate fire, plants showed an even greater response in total stem length, total stem diameter and shoot dry weight, signalling the importance of re-sprouting following disturbances such as fire or herbivory in savanna systems. Root starch (per unit root mass and total root starch per plant) increased steeply along the CO2 gradient, explaining the re-sprouting response. In contrast to the strong response of tree seedlings to the CO2 gradient, grass productivity showed little variation, even at low CO2 concentrations. These results suggest that CO2 has significant direct effects on tree recruitment in grassy ecosystems, influencing the ability of trees to recover from fire damage and herbivory. Fire and herbivore regimes that were effective in controlling tree increases in grassy ecosystems could thus be much less effective in a CO2-rich world, but field-based tests are needed to confirm this suggestion.