Objective To use a physics-informed neural network (PINN)-based model to predict hemodynamics in intracranial aneurysms and address the long simulation time and high computational cost of traditional computational fluid dynamics (CFD) simulations. Methods The PINN model was trained using only the computational domain coordinates and sparse velocity measurement points from CFD data of clinical patients. The predicted blood flow velocity, pressure, and wall shear stress (WSS) from the PINN model were compared with CFD simulation results. Results The proposed method was tested and validated on data from four different patients. For velocity prediction, the average mean absolute error (average MAE) was 4.60%, the average mean relative error (average MRE) was 6.61%, and the average mean squared error (average MSE) was 0.229%. For WSS prediction, the average MAE was 5.54%, the average MRE was 8.58%, and the average MSE was 0.510%. The PINN model demonstrates good generalization capability across different aneurysm models and can reduce the computation time of hemodynamics from several hours to just a few seconds. Conclusions The PINN model can effectively compensate for incomplete measurement data through physical constraints, even when boundary conditions are unknown and measurement data are sparse. It can rapidly and accurately simulate the hemodynamics of intracranial aneurysms. This method has the potential to provide effective support for clinical risk prediction in intracranial aneurysms.