NASA announced that its open-source Prithvi Geospatial AI model has become the first geospatial foundation model successfully deployed and tested in orbit. Developed through a collaboration between NASA and IBM, the model was uploaded to two space platforms — the Kanyini satellite and the IMAGIN-e payload aboard the International Space Station. The achievement marks a major milestone in bringing advanced artificial intelligence directly into space-based operations.
Prithvi was trained on 13 years of Earth observation data collected from NASA Landsat and ESA Sentinel-2 satellites. As a geospatial foundation model, it can perform a wide range of Earth-monitoring tasks, including flood mapping, wildfire detection, cloud analysis, crop monitoring, and disaster tracking. Researchers demonstrated that compressed versions of the model could successfully operate within the limited computing and power environments available on orbiting spacecraft.
A key advantage of deploying AI directly in orbit is that satellites can analyze data before transmitting it back to Earth. This reduces bandwidth requirements and enables faster responses to time-sensitive events such as natural disasters, wildfires, or severe weather conditions. NASA researchers explained that instead of uploading entirely new AI systems for different tasks, operators can simply send smaller decoder updates, making future space-based AI systems more flexible and efficient.
The project also reflects NASA’s broader push into AI-driven scientific research. Alongside Prithvi, NASA has released additional open-source AI foundation models such as Surya for heliophysics and solar storm prediction. Experts believe these technologies could eventually allow astronauts and mission operators to interact with satellites using natural language while enabling spacecraft to make autonomous scientific decisions in deep space. The success of Prithvi’s orbital deployment demonstrates how AI may become a core part of future Earth observation, climate science, and space exploration systems.
