We show that the negative relation between real investments and future stock returns is primarily driven by the subsample of firms building additional capacity. We develop a real options model to rationalize that evidence based on the premise that firms need to learn how to best operate modern capacity vintages, inducing idiosyncratic uncertainty in that capacity’s production costs over the learning period. Conversely, the uncertainty lowers the expected return of firms with newly built capacity until it is resolved. Further evidence based on profit sensitivities to aggregate conditions; analyst forecast-error volatilities; and high- versus low-tech industry subsamples supports our uncertainty explanation.