The very-high-cycle fatigue (VHCF) behavior of material BG2532, used in oil and gas completion strings, was investigated under both non-corrosive and hydrogen sulfide (H2S) gas corrosion conditions. During the experiment, the material's fatigue property and fatigue fracture characteristics were studied. Additionally, the microstructure on the axial cross-section, perpendicular to the fatigue fracture surface, was analyzed to explore the mechanism of corrosion-induced VHCF crack initiation. To enable unified VHCF life prediction for the material under both corrosive and non-corrosive conditions, different VHCF life prediction models were developed. Fatigue fracture characteristics, including the number of grains per unit area on fatigue source and the facet ratio on propagation area, were proposed as key parameters for VHCF modeling. Two artificial intelligence (AI)-fatigue models incorporating corrosion effects were developed and compared. The results show that integrating fatigue source and propagation characteristics using deep learning and convolutional neural networks significantly enhances the accuracy of VHCF life predictions, with errors remaining within a factor of two. This model effectively predicts the VHCF life of BG2532 alloy under both corrosive and non-corrosive conditions.