Computational Fluid Dynamics (CFD), despite its high accuracy in analyzing subsonic air inlets, restricts iterative design and optimization processes due to its significant computational cost. This study introduces a data-driven hybrid surrogate model capable of predicting static and stagnation pressures, drag force, Mach number, and Velocity Uniformity Index (VUI) with industrial-level precision. The model reduces Reynolds-Averaged Navier-Stokes (RANS) computation time from several hours to less than 5 milliseconds on a Graphics Processing Unit (GPU). The proposed framework relies on a Sparse Linear Model of Coregionalization Gaussian Process Regression (Sparse LMC-GPR with Matern 2.5 and Rational Quadratic kernels, and 15% inducing points) for high-variance quantities, and a boosted Random Forest model (with 500 trees) for low-variance outputs. A deterministic seed search (range 0 to 50) identified an optimal value of 20, reducing the 5-fold nRMSE to 0.11. Group cross-validation across 123 different geometries demonstrated nRMSE<0.13 for drag and nRMSE<0.08 for pressures. External validation on 10 random geometries revealed a Mean Absolute Percentage Error (MAPE) below 0.12% and R2>0.98 for all outputs. Ultimately, by providing calibrated uncertainty estimates, this model offers a reliable alternative to direct CFD solutions, facilitating real-time design and active learning frameworks.