Underground supermarkets are characterized by high occupant density, long operating hours, and substantial internal loads, creating competing demands for energy efficiency, thermal comfort, and carbon reduction. This study investigates the coordinated optimization of these performance objectives for a single-level underground supermarket in Xuzhou, China. Energy use intensity (EUI), occupied-hour discomfort time ratio (DTR), and carbon emissions from retrofit-material production (A1–A3) and operational energy use (B6) were adopted as objectives. Eleven envelope, overburden, morphology, and operational variables were evaluated using 2000 Latin hypercube samples through a Rhino-Grasshopper/Ladybug Tools-Honeybee workflow. Independent back-propagation neural-network surrogate models were combined with Morris sensitivity analysis and NSGA-III, and entropy-weight TOPSIS was used to rank Pareto solutions. The selected compromise solution reduced EUI by 50.9% and carbon emissions by 44.6%, while decreasing DTR from 49.29% to 23.96% relative to the baseline. External-wall insulation thickness had the strongest influence on EUI and DTR, whereas roof insulation thickness most strongly affected carbon performance because of the trade-off between operational energy savings and material-related emissions. These results demonstrate the need for coordinated envelope and operational strategies when balancing energy, comfort, and carbon performance in underground commercial buildings.