Abstract Future changes in the frequency of extreme drought events are of vital importance for risk assessment and relevant policy making. But a reliable estimation of their probability is intrinsically challenging due to limited available observations or simulations. Here, we use two large ensemble simulations, 50 members from CanESM2 and 40 members from CESM1 under the future RCP8.5 scenario, to elaborate a reliable projection of the 100-year drought events (once in a century) under different warming levels. It is however necessary to firstly remove systematic biases for the simulated temperature and precipitation through a bias-correction method based on quantile mapping. Droughts are diagnosed with the Standardized Precipitation Evapotranspiration Index (SPEI), which considers both precipitation and potential evapotranspiration (PET, involving temperature). The results show that the frequency of extreme droughts increases with the continued global warming. Some differences between the two ensembles are also observed, especially for high warming levels. The China-averaged probability of 100-year droughts that occur once in a century in the current climate increases by factors of 1.52 (1.90) and 1.44 (2.02) in the two ensembles under 1.5°C and 2°C warming levels, respectively. A simple statistical scheme shows that the increasing future risk of extreme droughts is mainly due to the increasing effect of PET on the occurrence of extreme drought events, while the effect of precipitation almost keeps constant with global warming.