Additive manufacturing (AM) is increasingly applied in aerospace for producing lightweight brackets, turbine blades, and propulsion components. A persistent challenge, however, involves the fact that process–property relationships learned from one alloy rarely transfer effectively to another. Aerospace alloys such as Ti-6Al-4V, Inconel 718, and AlSi10Mg each exhibit distinct thermal and absorptive behaviours, making qualification costly and reliant on extensive trial-and-error. Conventional machine learning models provide useful predictions but often act as black boxes, limiting both interpretability and cross-alloy applicability. This study introduces a physics-regularized transfer learning framework that adapts knowledge from a source alloy with extensive data (Ti-6Al-4V) to target alloys with limited datasets (Inconel 718, AlSi10Mg). A baseline neural network is fine-tuned for the target alloys, while the training is guided by physics-informed constraints, including volumetric energy density, thermal diffusivity, and absorptivity scaling. Embedding these terms into the loss function encourages the model to remain physically consistent while capturing alloy-specific features. Preliminary results suggest that the proposed approach improves predictive accuracy for surface roughness and porosity compared with both standalone neural networks and conventional transfer learning. Beyond accuracy, the framework represents a step toward generalizable and interpretable AI tools for aerospace AM. By reducing reliance on trial-and-error, such methods can accelerate alloy qualification, lower experimental costs, and contribute to more sustainable and efficient aircraft and spacecraft design.