Thermo-mechanical fatigue of solder interconnections is one of the primary reliability concerns in power electronics packages subjected to active temperature cycling (aTC). Accurate prediction of accumulated plastic strain and its variability is essential for reliability assessment; however, physics-based finite element simulations are computationally expensive, limiting their use for large-scale uncertainty studies. This work presents an artificial intelligence-driven surrogate modeling and uncertainty quantification framework to efficiently predict accumulated plastic strain across spatial locations in solder joints. A three-dimensional finite element model of an HPSOF8 power package was used to generate nodal response data for 200 uniformly distributed temperature cycling conditions defined by temperature amplitude, cycle timings and starting temperature. Temporal Convolutional Network surrogate models were trained to predict the time evolution of accumulated plastic strain at upper and lower solder interfaces, achieving high predictive accuracy with R² values ranging from 0.917 to 0.996 across spatial groups. Polynomial Chaos Expansion was then applied to quantify the influence of temperature amplitude uncertainty, showing that outer interface regions exhibit between 15 and 18 percent higher relative variability compared to inner and center locations, particularly under specific operating regimes. The proposed hybrid finite element, artificial intelligence, and uncertainty quantification framework enables rapid and accurate probabilistic prediction of fatigue-relevant quantities, providing an efficient approach towards reliability assessment and supporting uncertainty-aware design of power electronics packages.