The rapid growth of data-intensive applications and emerging technologies such as 5G, IoT, and edge computing has intensified the demand for efficient traffic management within telecommunications networks. Traditional routing protocols often struggle to adapt dynamically to fluctuating traffic patterns, latency constraints, and Quality of Service (QoS) requirements. This review explores an AI-powered predictive traffic routing framework designed to enhance network performance through real-time analytics and adaptive decision-making. The framework integrates machine learning models, particularly reinforcement learning, deep neural networks, and graph neural networks, to predict congestion trends, optimize routing paths, and balance network loads proactively. By leveraging predictive intelligence and data-driven optimization, the framework minimizes packet loss, reduces latency, and improves throughput across distributed infrastructures. Additionally, it incorporates feedback-driven learning loops and network telemetry for continuous self-optimization. The study reviews current advancements in AI-based routing systems, evaluates their scalability and interoperability in next-generation networks, and highlights implementation challenges such as computational overhead, data privacy, and model interpretability. The findings emphasize the transformative potential of predictive AI in enabling autonomous, resilient, and high-performance telecommunications ecosystems capable of supporting the exponential data demands of modern digital societies.