The rapid proliferation of cloud-native architectures has introduced unprecedented complexity in resource management, leading to significant financial waste and operational inefficiencies. This review examines the evolution of AI-driven cloud resource optimization, focusing on how machine learning (ML) models—ranging from predictive analytics to reinforcement learning—have become essential for modern enterprise infrastructure. By analyzing the shift from reactive monitoring to proactive, automated orchestration, we explore the integration of AI within the FinOps framework to achieve \"Inference Economics.\" The article investigates key methodologies such as predictive auto-scaling, intelligent rightsizing, and carbon-aware scheduling. Furthermore, it addresses the challenges of algorithmic bias, data privacy, and the computational overhead of AI models themselves. Ultimately, this review provides a comprehensive overview of how AI-driven optimization not only reduces Total Cost of Ownership (TCO) but also aligns cloud consumption with sustainability goals, offering a roadmap for future research in autonomous cloud environments.