Location prediction is an important aspect of mobility modeling. However, predicting person’s next location is challenging when they move to a different place. In this context, the cross-city next POI (Point of Interest) prediction task involves predicting which POI a user is likely to visit next while traveling from one city to another, for example, from City A to City B. However, the data on cross-city visits is usually very limited and often does not have past visit records in the target city. Given the data being sparse, we hypothesize that understanding the causal features of mobility will help predict the location of a user in a new city. In this work, we propose a novel method for spatiotemporal cross-city POI prediction using generative learning and causal perspective. We employ the interventional aspect of causality to identify the causal features and apply the causal features to cross-city next-location prediction. Our method exploits a Conditional Variational Autoencoder based generative model. We use large real-world LBSN (Location-based Social Network) data. In our experiments, we evaluate the causal sensitivity of different features related to the cross-city next POI prediction. Our proposed method demonstrates competitive performance in predicting the next POI influenced by causal features across cities.