1) C4 photosynthesis evolved when grasses migrated out of contracting forests under a declining atmospheric CO2 concentration ([CO2]a) and drying climate around 30 million years ago. C4 grasses are hypothesised to benefit from improved plant–water relations in open habitats like savannas, giving advantages over C3 plants under low [CO2]a. But experimental evidence in a low CO2 environment is limited and comparisons with C3 trees are needed to understand savanna vegetation patterns. 2) To test whether stomatal conductance (gS) and CO2 assimilation (A) are maintained in drier soil for C4 grasses than C3 trees, particularly under low [CO2]a, we investigated photosynthesis and plant–water relations of three C3 tree and three C4 grass species grown at 800, 400 or 200 ppm [CO2]a over moderate wetting–drying cycles. 3) C4 grasses had a lower soil–to–leaf water potential gradient than C3 trees, especially at 200 ppm [CO2]a, indicating reduced leaf water demand relative to supply. Yet the dependence of gS and A on predawn leaf water potential (a measure of soil water availability) was greater for the C4 grasses than trees, particularly under low [CO2]a. 4) Our findings establish that gS and A are not maintained in drier soil for C4 grasses compared with C3 trees, suggesting that this mechanism was not prevailing in the expansion of C4–dominated grasslands under low [CO2]a. This inherent susceptibility to sudden decreases in soil water availability justifies why C4 grasses have not evolved a resistant xylem allowing operation under drought, but instead shut down below a water potential threshold and rapidly recover. We point to this capacity to respond to transient water availability as a key overlooked driver of C4 grass success under low [CO2]a.