Multi-fidelity modeling seeks to establish an efficient framework that combines mathematical models of varying complexities, providing accurate approximations while minimizing computational burden. This approach is especially appealing for balancing prediction accuracy and computational cost. In aerospace engineering, where high-fidelity simulations such as computational fluid dynamics (CFD), finite element analysis (FEA), and coupled aeroelastic computations demand substantial resources, multi-fidelity methods leverage correlations between inexpensive low-fidelity models (e.g., potential flow, panel methods, or reduced-order models) and expensive-to-generate high-fidelity data. This enables efficient surrogate construction for design optimization, uncertainty quantification, and parametric studies. However, classical approaches such as Co-Kriging suffer from cubic scaling in computational cost with respect to the number of training samples, rendering them impractical for high-dimensional problems frequently encountered in aerospace applications. This paper introduces a novel mixture-of-experts–based multi-fidelity framework that addresses the computational bottleneck of traditional Gaussian process-based multi-fidelity methods. The proposed approach decomposes the high-fidelity response into a low-fidelity base model and a residual field, which is subsequently approximated using a mixture of local Gaussian process experts. By partitioning the input space via unsupervised clustering and training independent experts on localized regions, the framework achieves near-linear scaling in training time while maintaining predictive accuracy comparable to CoKriging. The method is validated on a suite of analytical benchmark functions with dimensionalities ranging from 2D to 15D, demonstrating substantial reductions in training time without significant loss in prediction quality. Furthermore, the framework is applied to two realistic aerospace design problems of practical relevance. First, a high-aspect-ratio wing configuration is considered, which poses significant challenges due to strong aerodynamic couplings and a high-dimensional parameter space characteristic of modern aircraft design. Second, a 29-dimensional sonic boom prediction problem is addressed, showcasing the framework’s effectiveness in high-dimensional scenarios where traditional multi-fidelity methods become computationally prohibitive.