The ability to understand, reason about, and generate 3D structures is fundamental for intelligent agents operating in the real world. Recent advances in 3D generative AI have opened new frontiers in spatial intelligence, enabling breakthroughs in areas such as scene reconstruction, object synthesis, robotic perception, and embodied AI. In this talk, I will present the latest research from my lab at UCSD over the recent years, focusing on novel generative models that enhance spatial understanding and interaction. I will discuss key innovations in neural implicit representations, diffusion models, and multi-modal learning paradigms that bridge 3D perception and generation. Additionally, I will highlight how these advancements contribute to practical applications in robotics, simulation, and augmented reality. Through this talk, I aim to provide insights into the evolving landscape of 3D generative AI and its implications for the future of intelligent systems.