Services for Cyber-Physical Systems based on Artificial Intelligence and Machine Learning require a virtual representation of the physical. To reduce modeling efforts and to synchronize results, for each system, a common and unique virtual representation used by all services during the whole system life-cycle is needed, i.e. a DigitalTwin. In this paper such a DigitalTwin, namely the AI reference model AITwin, is defined. This reference model is verified by using a running example from process industry and by analyzing the work done in recent projects.