For the first time, an Earth observation satellite identified targets on its own, without human analysts on the ground. The milestone occurred in April 2026 aboard Yam-9, a spacecraft operated by space infrastructure company Loft Orbital, and marks the first reported use of a vision-language model (VLM) in orbit.
The software behind the demonstration was NAVI-Orbital, developed by NASA’s Jet Propulsion Laboratory and led by technical leader Juan Delfa Victoria. It served as the framework for Google DeepMind’s Gemma 3, a VLM designed to run on limited hardware far from a data center. Researchers used natural language queries to prompt the model — asking it to classify areas where natural environments meet human development, or to identify infrastructure around railway hubs — and it did so autonomously. Yam-9 is equipped with an Nvidia Jetson Orin AGX GPU and was launched in fall 2025 as a pathfinder for Loft Orbital’s orbital AI work.
Traditionally, satellites download large volumes of raw data to ground-based analysts, who then sift through it manually or with machine learning tools. Running a VLM directly on the satellite could allow initial data triage to happen in orbit, reducing that burden significantly.
Longer term, the demonstration could point toward larger AI infrastructure deployed in space. “It opens the door to always-on, patrol layers in space,” Paul Lasserre, Loft Orbital’s head of AI, told TechCrunch. “If you have a VLM, you can have logic — like ‘monitor this border for me, and let me know when something is suspicious.'”
Other companies may follow. Planet Labs flies satellites with Jetson Orin processors and says research into VLMs is underway. Kepler Communications, which operates the largest cluster of GPUs in space, declined to confirm VLM deployments but noted “several undisclosed use cases” since its spacecraft launched in January 2026.
Lasserre said achieving real-time Earth coverage would require between 50 and 100 satellites like Yam-9; Loft currently operates 12. The concept also has potential scientific applications — the original idea behind NAVI-Space was a voice-driven AI assistant for astronauts on the Moon or Mars who cannot use a keyboard while suited up.
Source: TechCrunch