Inventory pooling is a collaborative arrangement in which different agents share their inventories to reduce the total inventory cost. Each agent maintains its own inventory, which are shared periodically through lateral transshipment. Analysis of an inventory pooling system is typified by demand profiles, cost sharing mechanism, level of collaboration, type of transshipment etc. By varying the choice in each dimension, different possible configurations of the system can be defined. Extensive analysis of the possible configurations is required for the agents to agree on a particular configuration. Such an analysis would require a multi-disciplinary approach to study each of the dimensions in isolation. We propose deep reinforcement learning as a single framework to analyze the inventory pooling configurations. Extensive computational experiments illustrate the efficacy of deep reinforcement learning for inventory pooling with generic assumptions on system characteristics, that are intractable using existing models.