Three-dimensional point cloud data (PCD) play an increasingly essential role in virtual reality, autonomous driving, gaming, digital manufacturing, and many other interactive applications. With the rapid development of deep learning, generative artificial intelligence has become a powerful solution for synthesizing high-quality 3D shapes. However, point cloud generation remains a challenging task due to the unordered, sparse, and noisy characteristics of PCD. This paper reviews recent advances in generative models for 3D point cloud synthesis, with a special focus on diffusion models and transformer-based architectures. We analyze the underlying mechanisms of text-to-3D generation, discuss the role of multimodal prompt embedding, and evaluate state-of-the-art models. We further summarize existing limitations, including computational cost and dataset scarcity, and explore future research directions that may improve the performance and generalization of 3D generative AI.