NASA and IBM Built a Free AI Model to Read 17 Years of Moon Data

Key points

• NASA and IBM announced the NASA-IBM Lunar Foundation Model on 10 September 2025.
• The free, open-source model is available on Hugging Face, a public platform for sharing AI tools.
• It was trained primarily on data from NASA's Lunar Reconnaissance Orbiter, which has been in lunar orbit for 17 years.
• The orbiter's dataset alone is larger than all of NASA's other planetary mission datasets combined.
• The model is the third in a series: earlier versions covered Earth observation and solar weather.

NASA has a moon-data problem. Seventeen years of imaging from the Lunar Reconnaissance Orbiter, a spacecraft that maps the moon's surface in high definition, have produced more data than every other NASA planetary mission put together. Reading all of it would take human researchers lifetimes.

So NASA and IBM built a tool to help. The NASA-IBM Lunar Foundation Model, a specialised AI trained to recognise patterns in lunar imagery the way a doctor learns to read an X-ray, launched on 10 September. It is open-source, meaning anyone can inspect how it works, and it is free to download from Hugging Face, the repository where researchers share AI tools.

The model is not a chatbot. Feed it an image of the lunar surface and it can flag craters, pick out features that suggest recent volcanic activity, or highlight regions where water ice might sit below the soil. NASA's chief science data officer Kevin Murphy said in the announcement that "collecting data is only part of the job" and that the agency needs to make that data easier to explore.

This is the third model in a steady NASA-IBM collaboration. An earlier version read satellite data about Earth's surface and atmosphere; a second focused on heliophysics, the study of the sun and the space weather it sends toward Earth. The pattern is consistent: take a volume of scientific data researchers cannot process fast enough, train an AI to do the first pass.

IBM has been on an active run lately. We reported on IBM's zero-shot AI forecasting model on 9 September, the day before this announcement, making the lunar model the latest in a string of IBM open-source releases we've tracked since 14 July.

For most people the practical effect is indirect but real. Faster crater-mapping means faster answers about where to land future missions safely. Better water-ice mapping directly shapes where astronauts might one day set up a base. Science that used to wait years for a team to finish annotating images could move on a much shorter timetable.

Murphy put it plainly: "That's a real opportunity we see with AI: turning large-scale data into new discoveries." The model is available now.

What does this mean for everyday people?

Nothing you need to act on today. What it signals is that publicly funded science is starting to move faster because AI can pre-process data that would otherwise sit in queues for years.

Common questions

Can anyone actually use this model?

Yes. Because it is open-source and hosted on Hugging Face at no cost, any researcher, university student or interested developer can download and run it. You don't need a NASA account or a special licence.

Is this connected to any crewed moon mission?

Not directly, but the science it produces, particularly better maps of water ice near the lunar poles, feeds directly into planning for NASA's Artemis programme, which aims to return astronauts to the moon's surface.