AI-Driven Cloud Cost Optimization: A Review of Machine Learning and Large Language Model Approaches and a Proposed Hybrid Architecture

Cloud computing provides organizations with flexible and scalable infrastructure, but this flexibility can also make cloud spending difficult to control. This paper reviews how machine learning and large language models are being applied to cloud cost optimization, including workload forecasting, anomaly detection, rightsizing, natural language cost analysis, and emerging agentic systems. Based on the reviewed literature, the paper proposes a conceptual eleven-layer hybrid architecture that combines machine learning analytics, large language model reasoning, FOCUS-based billing normalization, organizational context, risk assessment, human governance, and post-execution feedback. The review also discusses the limitations of current approaches, including dependence on data quality, limited production validation, lack of organizational context, and the security risks associated with automated infrastructure changes. A four-tier governance framework and a proposed evaluation approach are presented to support safer future development of AI-driven FinOps systems. The paper is intended as an undergraduate research exploration and conceptual framework for future empirical evaluation.