Computer-aided aerodynamic design is crucial for aircraft geometry optimization. To further improve the efficiency of aerodynamic characteristic modeling, an aircraft-oriented intelligent aerodynamic coefficient prediction method, AeroPointNet, was proposed. A three-dimensional point cloud representation of geometric models was employed as input, and a neural network architecture was constructed to efficiently extract both local and global geometric features. To capture variations in flow conditions, physical information was fused with geometric features, and two weighted attention mechanisms were introduced to dynamically adjust the weights, by which the problem of weight imbalance was effectively addressed. Experimental results show that AeroPointNet achieves a computational efficiency improvement of over three orders of magnitude in aerodynamic coefficient prediction compared with traditional numerical methods. The mean relative errors of lift and drag coefficients are kept below 5%.