Significant amount of machine learning (ML) applications on scientific data have hinted at the remarkable versatility of artificial intelligence (AI) in the era of exa-scale computation and data-driven modeling. However, unique challenges in Cosmology require a transition from heuristic applications of ML frameworks to domain-aware AI systems. A subset of these crucial issues will be discussed in this talk, with an emphasis on the adaptation of numerical simulations, explainability of AI models, uncertainty quantification, and scalability with state-of-the-art computational architectures. Bayesian inference of cosmological parameters using surrogate modeling, redshift estimation of galaxies, strong lensing studies and other key topics at the intersection of computation cosmology and scientific machine learning will be explored.