The North Atlantic Ocean is a region of intense uptake of atmospheric CO2. To assess how this CO2 sink has evolved over recent decades, various approaches have been used to estimate basin-wide uptake from the irregularly sampled in situ CO2 observations. Until now, the lack of robust uncertainties associated with observation-based gap-filling methods required to produce these estimates has limited the capacity to validate climate model simulated surface ocean CO2 concentrations. After robustly quantifying basin-wide and annually varying interpolation uncertainties using both observational and model data, we show that the North Atlantic surface ocean fugacity of CO2 (fCO(2-ocean)) increased at a significantly slower rate than that simulated by the latest generation of Earth System Models during the period 1992-2014. We further show, with initialized model simulations, that the inability of these models to capture the observed trend in surface fCO(2-ocean) is primarily due to biases in the models' ocean biogeochemistry. Our results imply that current projections may underestimate the contribution of the North Atlantic to mitigating increasing future atmospheric CO2 concentrations.