Recently, the launch and operation of high-resolution satellite data, including KOMPSAT-series satellites, microsatellites, and KOMPSAT-7, have been increasing. In particular, high-resolution satellites can acquire stereo imagery through overlapping observations, which is expected to provide significant potential for the acquisition of three-dimensional geospatial information. The extraction of 3D terrain information from high-resolution satellite imagery is performed through automatic matching between stereo image pairs, and various matching techniques have been developed for this purpose. With the recent advancement of artificial intelligence technologies, a variety of deep learning–based approaches have also been introduced in the field of stereo image matching. Among them, FoundationStereo, developed by NVIDIA, has been reported to demonstrate very strong performance. Therefore, this study applies the FoundationStereo method to high-resolution stereo satellite data, including CAS-500 (Korea Land Satellite) and KOMPSAT-series satellite data, in order to evaluate its performance. In the experiments, the quality of the extracted Digital Surface Model (DSM) obtained through stereo matching was compared and evaluated against the results produced by conventional template-based matching methods. In particular, the quality of the DSM was analyzed with a focus on elevation accuracy and shape completeness.