Shortwave (SW) cloud feedback (SWFB) is the primary driver of uncertainty in the effective climate sensitivity (ECS) predicted by global climate models (GCMs). ECS for several GCMs in the Sixth Coupled Model Intercomparison Project (CMIP6) exceed 5K, above the fifth assessment report (AR5) ‘likely’ maximum (4.5K) due to extratropical SWFB’s that are more positive than those simulated in previous generation CMIP5 GCMs. Here we show that across 57 GCMs Southern Ocean SW_FB; can be predicted from the sensitivity of column-integrated liquid water mass (LWP) to moisture convergence and to surface temperature. The response of LWP to moisture convergence and the response of albedo to LWP anti-correlate across GCMs. This is because GCMs that simulate a larger response of LWP to moisture convergence tend to have higher mean-state LWPs, which reduces the impact of additional LWP on albedo. Observational constraints suggest a modestly negative Southern Ocean SWFB— inconsistent with extreme ECS.