Corrugated-web steel I-girders (CWGs) have better shear and torsional strengths, offering additional stability against out-of-plane web buckling and local crippling under patch loads than flat web girders. CWGs are lightweight and more cost-effective than flat web beams for applications in buildings and bridges. The existing design models for estimating patch-loading resistance in CWGs assume equal widths in web corrugation folds for a simply supported span. This study examined the parametric influence of unequal trapezoidal-web folds on the nonlinear behavior of CWGs under compressive patch loading. A total of 533 steel I-girders with simply supported (SS) and 478 cantilever spans (CS) were analyzed using a calibrated nonlinear finite element (FE) methodology. The generated datasets of CWGs consisting of governing parameters (i.e., girder geometry and material properties) were mapped to refined patch-loading resistance using artificial neural networks (ANNs) for both CS and SS-type spans. The proposed ANN formulation showed better prediction accuracy (mean absolute error of about 2–4% in this study) than the existing design models for estimating the patch-loading resistance of CWGs. For practical applications and conservative patch-loading resistance, prediction modification factors were also proposed for ANN formulation and existing design models within a targeted error margin.