The global transition to renewable energy demands not only the large-scale deployment of photovoltaic (PV) systems but also their seamless integration into smart grid infrastructures. Conventional PV optimization strategies face significant challenges, including environmental variability, partial shading, and decision-making latency. This paper introduces a novel Edge-AI-driven optimization framework for PV systems, enabling real-time, decentralized intelligence for maximum power point tracking (MPPT), predictive load management, and fault detection. Mathematical models of PV cells are developed to highlight inherent nonlinearities, followed by a comparative review of classical MPPT and metaheuristic algorithms. An Edge-AI architecture is then proposed, integrating reinforcement learning, lightweight deep learning models, and IoT protocols for efficient grid interaction. Simulation results demonstrate the superior performance of the Edge-AI approach compared to conventional methods in terms of tracking efficiency, decision latency, and fault detection accuracy. Finally, the paper discusses future perspectives on federated learning, distributed optimization, and cyber-resilience, positioning this framework as a key enabler for next-generation renewable network infrastructures.