With advancements in deep learning, there has been a burgeoning interest in the exploration of pre-training techniques and the deployment of large models with billions of learning parameters. Self-supervised pre-training addresses the challenges associated with supervised learning, particularly the need for large amounts of labeled data, making it possible to leverage vast amounts of readily available data without annotations. Besides, it also catalyzes the emergence of large models that benefit from having more parameters to capture the variability and complexity of large-scale data. This chapter aims to provide a concise yet comprehensive overview of these domains, starting with an introduction to the emergences and foundational concepts of pre-training techniques and large models. Subsequently, we delve into the specific realm of point cloud data, demystifying the associated method designs related to pre-trained models and large models, which furnishes readers with a thorough understanding of these cutting-edge technologies.