<p>Adaptive Optics (AO) has become a cornerstone technology for mitigating atmospheric turbulence in Free Space Optical (FSO) communications, astronomical imaging, and quantum networks. Despite decades of progress, conventional AO architectures remain constrained by latency, scalability, and hardware cost, limiting their suitability for emerging 6G and quantum-secure infrastructures. This survey provides a comprehensive and structured review of recent AO advances across three interconnected research pillars: hardware-centric optimization, Machine Learning (ML)-driven intelligence, and model-free correction algorithms. By systematically comparing these approaches and showing how each addresses limitations of the others, the survey identifies several critical insights. High-density deformable mirrors (DMs) enhance correction fidelity but increase control complexity and cost. ML introduces predictive, low-latency control, yet depends heavily on realistic training datasets. Model-free strategies such as Stochastic Parallel Gradient Descent (SPGD) and Gerchberg–Saxton improve robustness in strong scintillation regimes, though at the expense of slower convergence and stability challenges. These findings underscore that no single AO paradigm is universally optimal; instead, hybrid frameworks combining hardware, AI, and blind correction represent the most resilient pathway forward. Ultimately, this survey integrates fragmented research into a unified perspective and outlines a roadmap for AO as an enabling technology for next-generation terrestrial FSO links, satellite-based internet, and global-scale quantum communication networks.</p>