The integration of renewable energy sources into smart grids has accelerated their evolution towards greater intelligence. However, as grids expand, traditional terrestrial communication networks face significant challenges, especially in remote areas with poor coverage. Satellite communication offers a potential solution but is limited by bandwidth and real-world channel conditions, such as Multiple-Input Multiple-Output (MIMO) channels. To address these issues, we propose a Satellite Semantic Communication MIMO system for Smart Grids (SSC-MIMO-SG). This system leverages the BeiDou satellite’s Regional Short Message Communication (RSMC) and a DNN-based semantic encoder to efficiently compress image data, reducing bandwidth usage. The system’s end-to-end training adapts semantic features to BD satellite MIMO channels, improving transmission reliability. Experimental results demonstrate that, compared to traditional methods, SSC-MIMO-SG significantly enhances PSNR performance under low SNR conditions, while both methods exhibit similar PSNR performance under high SNR conditions; MS-SSIM is improved by 28.53%; data transmission volume is increased by 71.43%; and communication latency is reduced by 88.47%.