Coupled thermo-mechanical part-scale simulations in metal additive manufacturing via laser powder bed fusion (LPBF) enable the prediction of residual stresses and thermal distortion without physically producing the part. Many existing part-scale simulations rely on heuristic models to capture the laser heat source, which often require calibration for specific process parameters and lack the capability for generalization. In contrast, scan-resolved simulations based on first principles offer the potential to apply the same computational model across a wide range of processing conditions, allowing for the assessment of how different factors influence the quantities of interest (QOIs) and even the exploration of novel process strategies without costly experimental trial and error. However, these high-fidelity simulations entail significant computational costs, requiring millions of time steps to accurately capture the multi-scale physics in LPBF. Previous work has demonstrated efficient computational strategies for purely thermal simulations [1] and for the coupled thermo-microstructure evolution [2], leveraging high-performance computing (HPC) based on modern computer architectures. The present contribution advances scan-resolved, coupled thermo-mechanical simulations on part-scale by presenting efficient solution strategies for the resulting linear systems, thus reducing the computational effort for accurately predicting QOIs such as residual stresses without over-resolving due to unnecessarily strict tolerances. Additionally, key aspects of the HPC implementation, such as the parallelism of modern computer architectures and the trade-offs between memory bandwidth and floating-point performance, are highlighted. These methods enable scan-resolved simulations of hundreds of layers and represent a significant step toward scalable, fully coupled simulations at the scale of real parts.