Semi-competing risks data arise where a non-terminal event (e.g., lung cancer) is censored by a terminal event (e.g., death). In some applications, semi-competing risks data are arranged in clusters such as geographic regions. Incorporating the cluster effect on the risk of events not only improves the accuracy and efficiency of parameter estimation, but also investigate spatial pattern of events over the study period and identify high-risk areas. The commonly used spatial-survival models are mostly restricted to single-event or competing risks settings. This work proposes a spatial semi-competing risk model in a Bayesian setting that allows for spatial variation while estimating risks of terminal and non-terminal events. The performance of the proposed model is evaluated in a simulation study, and a comparison of models with or without spatial effect is provided to investigate the cost of ignoring spatial variation. We also illustrate the proposed model in real data examples.