Data-driven generalizable aerodynamic analysis models demonstrate strong capability in performing fast aerodynamic analyses under arbitrary aerodynamic conditions, which provides an emerging technology for intelligent aircraft design optimization. However, training generalizable analysis models for complex aerodynamic shapes requires a large amount of aerodynamic data due to the curse of dimensionality issue, which impedes practical applications of this approach in the industry. Two tasks related to data-driven rapid optimization of airfoil and wing shapes were focused. By providing a proper representation of the aerodynamic shape design space, it effectively avoided the adverse effects of the “curse of dimensionality”. Demonstrations with approximately 100 000-scale computational fluid dynamics training datasets were provided, which enabled fast aerodynamic shape optimization of airfoils and wings.