The transition toward circular supply chains requires a fundamental rethinking of inventory management strategies to accommodate not only product flows but also reverse logistics, resource recovery, and environmental constraints. Traditional inventory models, typically optimized for linear throughput and cost efficiency, are ill-equipped to manage the complexities of circular systems, where material reuse, remanufacturing, and recycling must be balanced with customer service levels and sustainability goals. This article presents an AI-driven predictive inventory management framework tailored for circular supply chains. By leveraging machine learning algorithms and deep neural networks, the model forecasts both forward and reverse flows of goods, enabling accurate predictions of product returns, component availability, and secondary material inputs. These predictive insights are then used to inform dynamic inventory policies that optimize service level attainment while minimizing overstock, waste, and environmental impact. The proposed system integrates environmental performance indicators, such as carbon emissions and energy use, directly into the inventory decision process. It also accommodates uncertainty in return rates, reprocessing lead times, and variable demand for refurbished goods. Reinforcement Learning is employed to continuously adapt policies through simulation environments reflective of real-world circular operations. Case studies across the electronics and apparel sectors demonstrate that the model improves the synchronization of take-back systems, inventory replenishment, and production planning, resulting in enhanced resource efficiency and lower ecological footprints. The findings support a shift toward predictive, closed-loop inventory systems capable of sustaining service levels in environmentally constrained supply ecosystems.