With the increasing of the train speed, the aerodynamic drag rapidly increases and the drag reduction design has become one of the main objectives for the train aerodynamic shape design. Simulation is the main method of evaluating the train aerodynamic performance, hundreds or even thousands of aerodynamic drag simulations are required in the process of train design. The traditional train aerodynamic simulation faces a prominent challenge in balancing accuracy-efficiency-cost. High-precision simulation results rely on refined modeling strategies and high-fidelity physical model algorithms, leading to large computation. Using supercomputing resources significantly increases simulation costs, while simulation efficiency still cannot meet design requirements. The development of AI technology provides technological directions for solving these issues. In the present paper, the data-driven intelligent simulation technology was proposed to rapidly predict the aerodynamic drag of the high-speed train. The standard dataset was established using the surface mesh data, pressure data and shear stress data of trains. An AI model geometry-informed neural operator (GINO) with 250-billion parameters was constructed to predict the surface pressure and shear stress of the train, and the aerodynamic drag can be obtained by integrating the surface pressure and shear stress. The proposed AI model was used to predict the aerodynamic drag of high-speed trains with different shapes, and the results show that the data-driven intelligent simulation technology proposed in the present paper has good geometric generalization and can rapidly evaluate the aerodynamic drag of different train shapes. The computation error is less than 5% compared with wind tunnel tests, and the computation time has been reduced from days for the traditional CFD simulation on supercomputing platforms to seconds on a single GPU.