Catalytic performance of single-atom catalysts (SACs) relies fundamentally on the electronic nature and local coordination environment of the active site. Here, based on a machine-learning (ML)-aided density functional theory (DFT) method, we reveal that the intrinsic dipole in Janus materials has a significant impact on the catalytic activity of SACs, using 2D γ-phosphorus carbide (γ-PC) as a model system. Specifically, a local dipole around the active site is a key degree to tune the catalytic activity and can be used as an important descriptor with a high feature importance of 17.1% in predicting the difference of adsorption free energy (ΔGO* - ΔGOH*) to assess the activity of the oxygen evolution reaction. As a result, the catalytic performance of SACs can be tuned by an intrinsic dipole, in stark contrast to those external stimuli strategies previously used. These results suggest that dipole engineering and the revolutionary DFT-ML hybrid scheme are novel approaches for designing high-performance catalysts.