Systems, methods, and non-transitory computer-readable storage media for on-demand learning of routing information. The system receives a first packet directed to an endpoint device in a network, the first packet including a destination address associated with the endpoint device. Next, the system determines that the destination address does not have an associated network node identified in a routing table, and lists the destination address in a pending learning cache table to yield a pending learning entry. Based on the pending learning entry, the system then modifies the routing table to identify a network address of the associated network node for the destination address, wherein modifying the routing table is performed in response to receiving a second packet from the endpoint device, and wherein the network address of the associated network node is identified based on the second packet.