The increasing encryption of internet traffic challenges traditional network analysis tools, driving the need for innovative approaches. This poster outlines our research objectives in developing performant, AI-driven solutions for encrypted traffic analysis. We aim to leverage distributed, in-network AI inference by deploying decomposed AI models across programmable network devices, enabling real-time processing with minimal performance trade-offs. Focusing on the QUIC protocol as a representative case, we will address challenges related to resource constraints, compatibility with network operating systems, and energy efficiency. This work establishes a foundation for our future research, paving the way for high-performance network monitoring solutions.