The pursuit of climate-neutral aviation demands optimization methods that remain reliable under real-world uncertainty and the highly nonlinear physics of modern aircraft design. This keynote presents two complementary data-driven optimization strategies for robust aerodynamic design. First, we introduce a Distributionally Robust Optimization (DRO) framework that constructs ambiguity sets from operational and environmental data. By optimizing for the worst-case expected performance, DRO provides a statistically principled way to formalize the trade-off between performance and resilience to distributional shifts for real-world engineering. Second, we enhance gradient-based, high-fidelity design optimization by embedding learned design knowledge. Using generative diffusion models, we learn a smooth manifold of physically realistic shapes from existing designs and enforce it as an optimization constraint. Central to the method is the backpropagation of adjoint-computed shape gradients through this generative model to the latent space. This ensures the optimization remains within aerodynamically plausible regions, thereby eliminating trial-and-error tuning and ensuring greater robustness to initialization and optimizer choices. We will outline algorithmic components, integration with CFD and adjoint workflows, and representative transonic aerodynamic results that illustrate improved robustness compared to conventional approaches.