In today’s globally distributed and decentralized manufacturing environments, managing inventory efficiently presents significant challenges due to the increasing complexity of demand patterns, lead-time variability, and supply chain uncertainties. Traditional inventory optimization models, which rely on static assumptions and centralized control, often fall short in highly dynamic and geographically dispersed ecosystems. This paper introduces a novel framework for Dynamic Inventory Optimization using Reinforcement Learning (RL), tailored to the needs of decentralized global manufacturing supply chains. From a broader perspective, the study explores the limitations of conventional optimization methods in responding to real-time changes and disruptions, emphasizing the necessity for intelligent, adaptive, and autonomous decision-making systems. Reinforcement Learning is leveraged to create agents capable of learning optimal inventory policies through interaction with the supply environment, dynamically adjusting order quantities and replenishment strategies based on evolving conditions. These agents are embedded within a multi-agent system, enabling decentralized decision-making aligned with local objectives while maintaining global efficiency. The RL framework integrates real-time data streams from IoT-enabled devices and enterprise resource planning systems, ensuring that inventory decisions reflect the most current operational states across distributed nodes. The proposed system is validated through simulation scenarios reflective of real-world supply chain structures in sectors such as automotive and electronics manufacturing. Results indicate substantial improvements in service level performance, inventory holding cost reduction, and adaptability to supply-demand fluctuations compared to baseline heuristics. This work underscores the potential of combining artificial intelligence with decentralized supply chain architectures, offering a transformative approach to inventory optimization that is robust, scalable, and future-ready.