A comprehensive review on AI-driven beamforming techniques, trends, and challenges

Beamforming is a key enabler of 5G and beyond, particularly in millimeter-wave (mmWave) and terahertz (THz) bands, where severe path loss, blockage, and fast-varying channels challenge conventional approaches. Analog, digital, and hybrid beamforming architectures offer trade-offs between hardware cost, spectral efficiency, and flexibility but remain constrained by reliance on exhaustive beam search, accurate channel state information (CSI), and computationally intensive optimization. Artificial intelligence (AI) has emerged as a transformative solution, embedding predictive and adaptive capabilities into the beamforming process. Machine learning and deep learning enable CSI prediction and beam selection, reinforcement learning supports online policy adaptation in dynamic environments, and federated learning facilitates distributed optimization without raw data sharing. Generative AI, intelligent reflecting surfaces (IRS), and integrated sensing and communication (ISAC) further enhance coverage, adaptability, and dual functionality. This paper reviews and compares these approaches in terms of complexity, scalability, and robustness, and outlines key challenges in lightweight model design, dataset generalization, security, and energy efficiency. The findings highlight AI-native beamforming as a cornerstone for realizing intelligent, scalable, and energy-efficient 6G networks.