This paper investigates the non-uniform constellation (NUC) optimization adapted for Hierarchical modulation (HM) without using Successive Interference Cancellation (SIC). This approach reduces system demod/decode delay in comparison to Layered Division Multiplexing (LDM). The objective is enabling the capacity achieved by HM comparable to LDM. To achieve this goal, the constellation constrained capacity of Enhanced Layer (EL) service in HM is maximized, while the capacity of Core Layer (CL) service are approximately the same in HM and LDM. Particle Swarm optimization (PSO) algorithm is used to solve this problem. To accelerate the optimization, initial constellation is selected from regular NUCs or the combination of CL and EL constellations of LDM in ATSC 3.0. The results imply that under certain capacity demands, especially when there is a large difference between the SNR thresholds for correct decoding of CL and EL or the power ratio of CL to EL is high (for example, 10 dB or higher), HM, with lower delay compared to LDM, can achieve capacity close to LDM with the help of NUC. Even if the power ratio of CL to EL is relatively low (for example, 3 dB), the capacity loss can be reduced with properly designed NUC and the SNR threshold loss of EL can be lower than 1 dB with respect to LDM. However, LDM is still superior to HM when the difference between the SNR thresholds of CL and EL is relatively low.