For the first time, an Earth observation satellite has identified what it was looking for entirely on its own, without any input from human analysts on the ground. The breakthrough, which took place in April, marks the first reported use of a vision-language model (VLM) in orbit and signals a fundamental shift in how space-based sensors could operate in future.
The spacecraft, named Yam-9 and built by space infrastructure company Loft Orbital, carried a software package developed by NASA's Jet Propulsion Laboratory. This package harnessed Google DeepMind's Gemma 3, a VLM designed for edge computing — meaning it can run on compact hardware far from a data centre. VLMs combine the language understanding of large language models with the ability to analyse images. In the demonstration, the model was asked to classify sensor data where natural environments meet human development, and to identify infrastructure around railway hubs. It succeeded at both tasks.
Paul Lasserre, head of AI at Loft Orbital, told TechCrunch: 'It opens the door to always-on, patrol layers in space. If you have a VLM, you can have logic — like “monitor this border for me, and let me know when something is suspicious,” and interact back and forth with the satellites.' The achievement has two major implications. In the near term, it could make space sensors far more useful by performing initial data triage in orbit, dramatically cutting the volume of raw data that analysts currently have to sift through. Longer term, it is a proof of concept for running larger-scale AI infrastructure in space.
For UK businesses and consumers, the development could mean faster, cheaper access to high-quality satellite data. Industries such as agriculture, logistics, insurance, and environmental monitoring could benefit from near-real-time intelligence without relying on ground-based analysis. The UK's Information Commissioner's Office (ICO) has yet to issue specific guidance on autonomous AI in space, but the technology will likely fall under existing data protection and AI governance frameworks. Meanwhile, the EU AI Act, which imposes risk-based rules on AI systems, could influence how British firms deploy such models if they operate in European markets.
Experts caution that while the demonstration is promising, significant hurdles remain. Power and memory management in space are major constraints, and scaling up from a single satellite to a constellation of 50 to 100 — which Loft Orbital estimates would be needed for continuous global coverage — will require further innovation. Juan Delfa Victoria, a technical leader at NASA JPL's AI group, led the development of the software harness for the VLM, noting that engineers had to streamline libraries and reduce memory usage to make the model fit for orbit.
Other space companies are already moving in a similar direction. Planet Labs flies satellites with Nvidia Jetson Orin processors, currently used for simpler object detection, but a spokesperson confirmed research is underway on VLMs. Kepler Communications, which operates the largest group of GPUs in space, declined to comment on specific deployments due to non-disclosure agreements but noted there have been 'several undisclosed use cases of our compute environment' since its spacecraft launched in January. Source: TechCrunch