Due to its wide‐ranging impacts, predicting decadal variations of sea surface temperature (SST) in the subpolar North Atlantic remains a key goal of climate science. Here, we compare the representation of observed subpolar SST variations since 1960 in initialized and uninitialized historical simulations from the 5th and 6th phases of the Coupled Model Intercomparison Project (CMIP5/6). Initialized decadal hindcasts from CMIP6 explain 88% of observed SST variance post‐1980 in the subpolar gyre at lead years 5–7 (77% in uninitialized simulations) compared to 42% (8%) in CMIP5, indicating a more prominent role for forcing in driving observed subpolar SST changes than previously thought. Analysis of single‐forcing experiments suggests much of this correlation is due to natural forcing, explaining ∼55% of the observed variance. The amplitude of observed subpolar SST variations is underestimated in historical simulations and improved by initialization in CMIP6, indicating continued value of initialization for predicting North Atlantic SST.