This paper investigates the potentials of utilizing artificial intelligence (AI) based edge caching in the next generation of vehicular mobile networks. In recent years, vehicle-to-everything (V2X) has been a research focus, which enables the exchange of information between the vehicles and the outside world. To integrate vehicular networks and cellular radio technology, cellular-V2X (C-V2X) was proposed in 3GPP release 14. Further, mobile edge caching is regarded as an effective technique to allow local data access, which can support the low latency requirement of the V2X use cases. With the advance of AI technologies such as deep learning, there has been increasing demand in inference and learning from big vehicular data. In this paper, we present the detailed architecture of AI-based edge caching in vehicular networks with misbehaving vehicle detection as an illustrative case. Performance results are provided to investigate the benefit of the proposed architecture. Finally, we highlight the potential research directions.