The proliferation of AI into everyday devices is a major trend today. This trend combined with the increasing amount of different AI hardware architectures and software frameworks imposes significant challenges when we want to interconnect such AI-based devices into single, large AI-driven distributed system. This paper addresses one key challenge which is around the problem of sharing AI encoded information among components of vastly heterogeneous nature. For that end we propose a new concept called Neuromorphic Data Layer, which can bridge various internal AI data representations in a communication channel-friendly way. The proposed methods are also stress tested in a distributed industrial robotic control & training use-case where all components are state-of-the-art devices, have some form of AI computation and they are interconnected over wireless technologies using the proposed Neuromorphic Data Layer.