In this paper, we propose an artificial intelligence (AI) powered data-driven computing method (DDCM) for shape memory alloy (SMA) behavior modeling. DDCM can directly use material behavior data to simulate thermomechanical behavior, bypassing complex constitutive modeling. The recurrent neural network (RNN) model are utilized to expand existing data collected from numerical simulations with simple loading and unloading conditions. The combination of DDCM and AI provides an alternative way for material development with less prior knowledge and good problem migration.