Traditional model-based and mass-compensated leak detection systems for supercritical CO2 pipelines face challenges due to inaccurate fluid density predictions, leading to frequent false alarms and potential missed leaks from system detuning. Negative pressure wave methods also struggle to balance sensitivity with false positives, exacerbated by high signal attenuation in CO2 fluids. This paper presents the successful design, implementation, and testing of an internal pressure pulse-based leak detection system on an operational 12-mile supercritical CO2 pipeline segment in South Texas, USA. The system integrates edge-based signal processing with a range of near-real-time, server-based machine learning techniques to deliver highly sensitive leak detection with minimal false positives, achieving detection-to-notification times of typically 2–5 minutes. Accurate leak location is enabled by high-frequency pressure sampling (1 kHz), GPS-synchronized timing, and real-time updates to signal propagation rates (speed of sound). The technology’s adaptability to both new and existing pipelines is discussed, requiring only distributed pressure sensors spaced 10–20 miles apart. The system’s ability to complement existing leak detection methods, enhancing sensitivity in high-consequence areas, is also highlighted.