Yam-9 Satellite Spots Targets in Orbit Using Gemma 3, a First for Vision-Language AI
Updated
Updated · TechCrunch · Jun 15
Yam-9 Satellite Spots Targets in Orbit Using Gemma 3, a First for Vision-Language AI
3 articles · Updated · TechCrunch · Jun 15
Summary
April marked the first reported in-orbit use of a vision-language model, with Yam-9 autonomously flagging areas of interest instead of sending all imagery to ground analysts.
Gemma 3, running through NASA JPL’s NAVI-Orbital software on an Nvidia Jetson Orrin AGX, answered natural-language prompts such as finding rail-hub infrastructure or boundaries between nature and development.
That onboard triage could cut the volume of raw Earth-observation data sent to Earth and enable more interactive monitoring, including persistent patrol-style surveillance for suspicious activity.
Loft Orbital launched Yam-9 in fall 2025 as an AI pathfinder and says real-time global coverage would require roughly 50 to 100 similar satellites; it currently operates 12.
The test also points toward a broader shift to orbital AI compute, with Planet Labs researching VLMs and Kepler Communications citing several undisclosed in-space GPU use cases.
With China and the US racing to deploy AI in space, is an automated orbital conflict now inevitable?
Now that AI can analyze Earth from orbit, who ensures this powerful new perspective is used for good?
As commercial chips power satellites, how vulnerable are these new orbital 'brains' to solar flares and sophisticated cyberattacks?
Yam-9’s 2026 Breakthrough: The First Autonomous AI Target Identification in Orbit and Its Impact on the Future of Space Operations
Overview
In April 2026, Loft Orbital's Yam-9 satellite made history by using a sophisticated vision-language model (VLM) to autonomously identify targets while in orbit. This breakthrough allowed Yam-9 to process and understand visual data, enabling real-time, on-board decision-making without constant human input. By introducing logic to satellite operations, Yam-9 could monitor boundaries, report suspicious activities, and even interact directly with other satellites. This achievement marks a major shift in satellite technology, showing how advanced AI can transform space missions by making satellites more independent, efficient, and capable of complex tasks.