Abstract: AI thermal model can effectively improve the refrigeration energy efficiency ratio of data center. When applied in data centers, AI model is limited in accuracy and flexibility due to low volume and coverage of training samples collected from real world. A synthetic data enhancement technique based on computational fluid dynamics (Computational Fluid Dynamics, CFD) is introduced in this paper. The synthetic data is used as an augmentation over real-world dataset to train AI model for thermal management. Experimental results show that CFD data enhancement technology can not only improve the accuracy of the model, reduce the prediction error under insufficient data condition, but also improve the performance of the model under the condition that the actual data can not be covered.