In contrast to the conventional feedback approach, the energy balancing task of a grid-side modular multilevel converter (MMC) with half-bridge cells and an isolated ac star point is considered as an optimization problem. As a result, nominal trajectories for circulating currents and common-mode voltage are obtained that inherently steer the system back to a balanced operation within finite time. The method relies on an MMC arm energy model, allowing for algebraic parameterization of almost all MMC variables during optimization, which considerably reduces the computational cost. Thus, the optimized solution is the one selected from a family of trajectories that meets the balancing goal. Owing to the trajectory planning of the MMC energies, the search for a solution is inherently limited to the domain of realistic energy variations and no balancing error remains even during transfers between operating regimes, i.e., the task of the balancing controller is reduced to the compensation of parameter uncertainties and disturbances. Here, a grid-side MMC is considered in the optimization in contrast to the previous work, which has been restricted to passive RL loads. Measurement results reveal drawbacks of the previous solution after adoption of a grid-side application. A dedicated candidate transfer for grid-side applications eliminate the drawbacks and regains the improved balancing performance as shown by measurements.