The rapid expansion of satellite constellations has fundamentally transformed space operations, resulting in unprecedented volumes of onboard data and increasing dependence on limited downlink capacity and ground-based processing. This traditional ground-centric paradigm introduces latency, scalability challenges, and operational inefficiencies, particularly for time-critical applications such as anomaly detection, Earth observation, and constellation-level coordination. As the number of satellites continues to grow, these limitations pose significant risks to the sustainability and responsiveness of future space systems.

This paper proposes an edge AI-enabled framework for on-orbit intelligence that shifts key decision-making processes from ground stations to the satellites themselves. Lightweight machine learning models are deployed onboard to perform real-time data filtering, prioritization, and anomaly detection, allowing only high-value information to be transmitted to Earth. By reducing unnecessary data transmission, the proposed approach alleviates communication bottlenecks while improving operational agility.

A conceptual system architecture is presented to illustrate how edge AI can be integrated into resource-constrained space environments. Design considerations such as power limitations, radiation exposure, fault tolerance, and model update strategies are discussed to highlight practical deployment challenges. The framework emphasizes scalability and resilience, making it suitable for large and heterogeneous satellite constellations.

This study demonstrates how relocating intelligence from ground control to orbit can enhance system autonomy, reduce downlink dependency, and support more sustainable space operations. By advancing the digitalization and autonomy of space systems, this work contributes to the development of next-generation aerospace architectures and provides actionable insights for AI-enabled satellite constellation design.