In modern aerospace design, aerodynamic data are collected at multiple fidelity levels, such as Computational Fluid Dynamics (CFD) simulations and Wind Tunnel (WT) experiments, each characterized by different trade-offs in cost, accuracy, and spatial coverage [6, 4]. Low-fidelity CFD enables extensive exploration of the design space at relatively low cost during early design phases, whereas high-fidelity WT measurements provide more reliable but expensive information limited in availability. Efficiently combining these heterogeneous data sources within an Adaptive Design of Experiments (ADoE) framework is therefore essential for surrogate modeling, optimization, and uncertainty quantification in aerodynamic applications [5]. Several surrogate Gaussian Process (GP) modeling approaches have been proposed to address these challenges, including multi-output GPs [8], Hierarchical Kriging [1], and co-Kriging-based multi-fidelity models [2]. These methods exploit correlations between outputs or fidelity levels to improve predictive accuracy and sample efficiency. However, standard GP-based formulations remain computationally demanding and become difficult to apply efficiently when large and heterogeneous datasets are involved. To address these limitations, this work proposes a scalable multi-fidelity ADoE. The approach combines the autoregressive formulation of multi-fidelity modeling [2] with sparse GP approximations [7], enabling efficient inference for large-scale datasets. The proposed framework naturally accommodates flexible sampling strategies, removing the requirement of nested DoEs, which are commonly assumed in multi-fidelity Kriging frameworks [3], while efficiently exploiting cross-fidelity correlations. The resulting methodology enables scalable surrogate construction, efficient uncertainty quantification, and adaptive data acquisition across fidelity levels. Its effectiveness is demonstrated on analytical benchmark problems as well as real aerodynamic applications involving large multi-fidelity datasets combining CFD simulations and WT experiments, showing promising results in terms of accuracy, scalability, and experimental efficiency.