In this paper, we will discuss the capability of a few deep neural networks for linearizing 5G power amplifiers. At first, the digital predistortion (DPD) linearization with traditional neural networks will be discussed. The challenges of the DPD linearization for the ultra-broad band of the 5G power amplifiers will be analyzed after that. Then the possible solution, which is based on some deep neural networks consisted of CNN and LSTM, will be presented. Finally, some primary experimental results will be showed to illustrate the performance of the deep neural networks in linearizing 5G power amplifiers.