Robust segmentation of infant brain MRI across multiple ages, modalities, and sites remains challenging due to the intrinsic heterogeneity caused by different MRI scanners, vendors, or acquisition sequences, as well as varying stages of neurodevelopment. To address this challenge, previous studies have explored domain adaptation (DA) algorithms from various perspectives, including feature alignment, entropy minimization, contrast synthesis (style transfer), and pseudo-labeling. This paper introduces a novel framework called MAPSeg (Masked Autoencoding and Pseudo-labelling Segmentation) to address the challenges of cross-age, cross-modality, and cross-site segmentation of subcortical regions in infant brain MRI. Utilizing 3D masked autoencoding as well as masked pseudo-labeling, the model is able to jointly learn from labeled source domain data and unlabeled target domain data. We evaluated our framework on expert-annotated datasets acquired from different ages and sites. MAPSeg consistently outperformed other methods, including previous state-of-the-art supervised baselines, domain generalization, and domain adaptation frameworks in segmenting subcortical regions regardless of age, modality, or acquisition site. The code and pretrained encoder will be publicly available at https://github.com/XuzheZ/MAPSeg