As large scale penetration of renewables into electric systems requires increasing flexibility from dispatchable production units, the electricity mix must be adapted to brutal variations of residual demand. Using tools from distributionally robust optimization (DRO), we propose a trajectory ambiguity set including residual demand trajectories answering both support and variability criterion using quantile information, and approximate the level-maximizing and variability-maximizing residual demand trajectories using two simple algorithms. These two limiting trajectories allow us to make investment decisions robust to extremely high levels and brutal variations of residual demand. We provide a numerical experiment using a MILP investment and unit commitment model in the case of France and discuss the results.