Generative Adversarial Network (GAN) has drawn increasing attention to augment training data to improve palm vein authentication performance. However, existing GAN-based data augmentation methods focus on generating only intra-class or inter-class samples. In this paper, we propose generating intra-class and inter-class samples by the same GAN, leading to an inter-intra collaborative data augmentation method for palm vein authentication. Furthermore, to effectively exploit the generative data for training, we design a domain adaption network to alleviate the domain discrepancy between the real data and synthetic data. Experiments conducted on the SCUT_PV_V1, PolyU_NIR, and CASIA-MS-PalmprintV1 databases have demonstrated the effectiveness of our method that performs competitive performance for palm vein authentication tasks.