Abstract: An AI-based computation method for beam bridges is presented based on physics-informed neural networks (PINN), which is integrated with BIM to develop a one-stop parametric modeling and analysis platform. Firstly, the primary challenges in the intelligent computation of bridge structures are identified, which are the complex boundary conditions corresponding to multiple load cases and the structural heterogeneity caused by variable cross-sections. Subsequently, a PINN architecture tailored for variable cross-section beam bridges is proposed. This includes the innovative polynomial kernel function strategy, the design of theoretical loss functions, and the development of calculation methods for various load cases. The BIM platform’s user interaction interface, developed based on the Windows Presentation Foundation framework, enables data input required for bridge analysis without migrating the model. Numerical experiments, using the Changtai bridge project as a case study, demonstrate that the computational results of the intelligent model have an error of less than 5% compared with MIDAS CIVIL, meeting the accuracy requirements for engineering applications and proving the reliability. Additionally, an ablation experiment indicates that conventional PINNs are unsuitable for bridge structure analysis, further validating the effectiveness and necessity of the polynomial kernel function strategy. This platform leverages the efficient parametric modeling advantages of BIM, overcoming its information silo issue, and achieves intelligent computation of responses for beam bridges, significantly enhancing the digitalization of the bridge design process.