Artificial Intelligence (AI) methods can complement traditional downscaling techniques such as dynamical downscaling, reducing the computational cost and facilitating the downscaling of large ensembles of Global Climate Models (GCM). To this end, we train, test and evaluate multiple super-resolution convolutional neural networks (SRCNN) using 40 years of daily precipitation data from a coarse resolution reanalysis and a high-resolution dynamically downscaled counterpart. The dynamically downscaled simulations are constrained using spectral nudging, which allows for the reproduction of historical events at a finer spatial resolution. This unique nature of the training dataset enables the AI-downscaling method to effectively emulate dynamical downscaling. Additionally, incorporating elevation data and bias-correction inputs in the model leads to overall improvement. Furthermore, using custom loss functions based on exponential and quantiles improves the simulation of extremes. We also demonstrate the effectiveness of SRCNN models in efficiently downscaling precipitation from six Coupled Model Intercomparison Project Phase 6 (CMIP6) GCMs over a 40-year historical period. Future work will build on this methodology to downscale additional variables using climate projections for future periods, which will be used in subsequent climate impact research. Overall, this study aims to advance research efforts to make AI downscaling techniques practical for the climate research and impact assessment communities.