Increasing evidence shows that climate change promotes the development of harmful cyanobacterial blooms, and affects the competition of toxic and non-toxic strains. Previous studies have found low CO2 favors toxic strains, but how changing dissolved CO2 (CO2 [aq]) in water body influences the succession of toxic and non-toxic strains remains uncertain. Here, we combined laboratory competition experiments, field observations, and a machine learning model to reveal the links between CO2 changes and the succession. Laboratory experiments showed under low CO2 conditions (100–150 ppm), toxic strains better utilized CO2 (aq) and dominated. Non-toxic strains demonstrated a growth advantage as CO2 concentration increased (400–1000 ppm). Field observations from May to November in Taihu Lake showed the percentage of toxic strains increased as CO2 (aq) decreased. Machine learning highlighted links between the inorganic carbon concentration and the proportion of advantageous strains. Our findings provide new insights for cyanoHABs prediction and prevention.