The future of AI-based 3D point cloud coding has numerous research opportunities and challenges. Key areas include advancing deep learning-based compression techniques to optimize geometry and attribute coding, with a focus on lightweight neural networks to reduce computational overhead and enable large-scale applications. Robust point cloud quality assessment methods are needed, integrating both subjective and objective evaluations to address geometry and attribute quality. Dynamic point cloud coding requires improved temporal context acquisition and computational efficiency for real-time processing. Human and machine perception oriented coding methods demand adaptive compression frameworks to balance bit rates, signal fidelity, and perception performance. Compression artifact removal, particularly for sparse LiDAR point clouds and dynamic scenarios, remains a critical challenge. Standardization efforts must evolve to address dynamic point clouds and integrate low-level and high-level tasks. Implementation systems should leverage hardware acceleration and edge computing for efficient streaming and rendering. Open-source projects and large language models (LLMs) offer promising avenues for innovations, though LLMs require enhanced spatial understanding for effective point cloud coding. Collectively, these advancements aim to enhance point cloud coding efficiency, perceptual quality, low latency, flexibility, and scalability across diverse applications like autonomous driving and virtual reality.