Multi-fidelity surrogate models are mathematical approximations that merge sparse, high-accuracy data with abundant, low-accuracy data to predict system performance. They are vital in thermal design optimization, enabling fast and cost-effective evaluation of complex thermal-fluid behaviors.