Constellation Space, a YC W26 startup, has built an AI system that forecasts satellite link failures minutes before they occur. The founding team brings experience from SpaceX, Blue Origin, and NASA, where they repeatedly saw operators lose data because degradation was only noticed reactively. Their platform ingests telemetry from satellites, weather, and space weather monitors to model the complex physics of RF links.

Traditional satellite operations rely on dashboards and manual rerouting once a signal-to-noise ratio drops. With constellations growing to over 70,000 satellites by 2030, this approach is unscalable. Constellation Space's system combines real-time physics-based models with machine learning trained on billions of data points. It uses federated learning to share patterns across operators without exposing raw telemetry, enabling transfer learning across different orbits and frequencies.

The system can predict most failures 3-5 minutes ahead with over 90% accuracy, providing time for automated rerouting. It's containerized for deployment on-premise, in government clouds, or commercial environments. While longer-range predictions remain a challenge, the startup is currently testing with defense and commercial partners. They expose APIs and even an LLM chat endpoint for troubleshooting, seeking feedback on making the tool production-ready for network operations centers.