The globally optimal solution regarding the energy management of a fuel cell hybrid electric vehicle forms the basis for benchmarking and deriving energy management strategies. The corresponding dynamic optimization problem is often solved with dynamic programming (DP), the main drawback of which is the high computational complexity. This work proposes an algorithm that adapts the DP step size depending on the power demand signal. In this way, the computational time and the memory requirements can be reduced considerably compared to DP with equidistant step size while preserving the accuracy. The benefits of the proposed adaptive step size DP are demonstrated using measurements of a real-world driving cycle.