The oceans have absorbed a large fraction of anthropogenic carbon dioxide emissions, having consequences for ocean biogeochemistry and ecosystems via ocean acidification. Simulations with Earth System Models can be used to predict the future evolution of ocean carbon uptake and acidification in the coming decades and beyond, but there is substantial uncertainty in these model predictions, particularly on regional scales. Such uncertainty challenges decision makers faced with protecting the future health of ocean ecosystems. Uncertainty can be separated into three component parts: (1) uncertainty due to internal variability, (2) uncertainty due to model structure, and (3) uncertainty due to emission scenario. Here, we isolate and quantify the evolution of these three sources of prediction uncertainty in ocean carbon uptake over the next century using output from two sets of ensembles from the Community Earth System Model (CESM) along with output from models participating in the Fifth Coupled Model Intercomparison Project (CMIP5). We find that the three sources of prediction uncertainty in ocean carbon uptake are not constant, but instead vary with prediction lead time and the scale of spatial averaging. In order to provide valuable predictions to decision makers, we should invest in reducing the main sources of uncertainty.