The 5G network not only causes many internet real-time broadband services to become wireless, such as virtual reality and cloud gaming, but also promotes mobile-inherent brand-new emerging services, like in-vehicle real-time video communications. One of the main visions of 5G and beyond 5G is to provide human-centric communications, not only for diverse services, but also better quality of experience (QoE) for users. In the meantime, as with the time variance challenges faced by all wireless networks, the problem of controllability for QoE is still noticeable in 5G. This article presents an AI-enabled QoE predicting and sustainability architecture, cooperating with 5G network data analytic function, network slicing, and MEC technologies to collect cross-layer performance data in real time, as well as adjust network resources accordingly. Thereby, it can support QoE-demanding services, create business innovation, and improve energy efficiency. In addition to illustrating the operation of the proposed architecture with one video streaming scenario, a machine learning QoE predicting model, its performance of the field trial, and research directions are discussed.