Uncertainty quantification in scientific machine learning, particularly for neural network-based methods, has garnered significant attention. However, current inference techniques often face challenges such as high computational costs for high-dimensional posterior inference or inadequate uncertainty estimates. In this talk, we introduce an efficient uncertainty quantification framework for physics-informed neural networks and deep operator learning using Ensemble Kalman Inversion (EKI). Our findings demonstrate that the proposed method delivers uncertainty estimates that are as informative as those obtained through Hamiltonian Monte Carlo (HMC)-based approaches, but with significantly lower computational costs. Additionally, we shall discuss the promising ways of extending this approach to larger-scale networks.