Abstract: In practical engineering applications, the hollow core sandwich concrete GFRP-steel tube composite column (GCS) can be employed in the pile foundation platforms of offshore wind turbines operating in corrosive environments. In this study, a finite element model of GCS subjected to axial compression was established and experimentally validated. Using the foundational data provided by the finite element model, an artificial neural network (ANN) was trained. The trained ANN was then integrated with genetic algorithm (GA), using GFRP tube thickness (t_G), GFRP tube inner diameter (D_G,i), steel tube thickness (t_s), steel tube outer diameter (D_S,o), GFRP tube strength (f_G), concrete strength (f_C), and steel tube strength (f_S) as the optimization variables for the GA. The optimization objective was to minimize the absolute difference between the target loading capacity (F^*) and the actual loading capacity (F) of GCS columns, relative to the manufacturing cost (P). The optimization focused on the structural design of GCS columns under five specific loading capacities. A comparison of the cross-sectional information for GCS columns under these specific loading capacities was performed. The optimization results from the GA-ANN algorithm were verified using finite element simulation software. The research findings indicate that, at the same loading capacity, there is a significant difference in the price (P) of GCS columns, emphasizing the necessity of optimization. The loading capacity error was largely controlled within 5%, confirming the accuracy of the GA-ANN algorithm. These research achievements not only demonstrate the effectiveness of the GA-ANN algorithm in the optimization of composite material structures, but also provide new technical ideas for the intelligent design of offshore engineering structures.