Bottom bending is a widely used sheet metal forming process in several industries. Although finite element analysis is commonly employed for process optimization, its high cost and the requirement for specialized expertise limit its widespread adoption, leading many manufacturers to rely on trial-and-error methods [1]. Recently, machine learning techniques have been explored to predict forming results [2], but their applicability is often constrained by limited generalization to new materials and designs. This work proposes an AI-driven strategy for the metamodeling and optimization of the bottom bending process, enabling automatic design and real-time optimization. A fully parameterized finite element model was used to generate an extensive database covering a wide range of materials, tool geometries, and process parameters. Artificial neural networks were trained on this dataset to predict the final bending angle, maximum punch force, and maximum thickness reduction. The metamodel achieved R² values exceeding 0.9 for all outputs. Subsequently, the metamodel was coupled with a grid search–based optimization algorithm to identify configurations that maintain the target bending angle while minimizing punch force and thickness reduction. The strategy was applied to several case studies, and the optimized solutions achieved average reductions of 28% in maximum punch force and 24% in maximum thickness reduction, with an identical final bending angle (average difference of 1.4%). These results demonstrate that ANN-based metamodels provide an efficient and accurate alternative to conventional simulation-based optimization, accelerating process design and improving bending performance.