Deep Semantic Communication (DSC) has emerged as a promising paradigm within the rapidly evolving landscape of digital communication. However, its potential in impulsive noise environments—a significant challenge to transmission reliability—has never been explored before. This letter addresses this gap by proposing IN-DeepSC, a DSC architecture specifically designed to mitigate impulsive noise. We introduce a novel empirical approach to evaluate the mutual information (MI) conveyed by IN-DeepSC, demonstrating its superior performance compared to traditional bit-based communication systems and the original popular DeepSC model. Our findings highlight DSC’s ability to exceed the Shannon capacity limit in challenging noise conditions, positioning it as a transformative technology for future communication systems.