AI
IBM and NASA Just Released an Open Source AI Built for the Moon
The most interesting AI release this week has nothing to do with writing your emails. IBM and NASA just released their Lunar Foundation Model, an open source AI built to study the Moon, and it is the first open model of its kind trained specifically for lunar science.
The training data is the story. The model learned from more than 30 spatially aligned data layers gathered by nine instruments across four NASA missions, including the Lunar Reconnaissance Orbiter and the GRAIL mission, plus complementary data from Japan's SELENE/Kaguya spacecraft. IBM and NASA also released the unified dataset alongside it: tens of thousands of images and maps showing the Moon's geophysical properties, ready for machine learning. Until now, researchers say, a public dataset bringing all of this into one framework built for modern AI had yet to exist.
What does it do? The model maps lunar ice deposits hiding in permanently shadowed regions, charts craters to pick safe landing sites, and studies volcanic features like irregular mare patches that reveal the Moon's thermal history. In benchmark tests it identified key surface features up to 23 percent more accurately than widely used methods. Tasks that once meant scientists sifting through maps by hand now run at machine scale.
Lunar ice is the prize. Ice means water and oxygen for a future Moon base, and it can be turned into rocket fuel for missions to Mars. NASA's Artemis program aims to return astronauts to the Moon in 2028 and test the technology for a sustained human presence. An AI that finds the ice faster brings that timeline closer.
The model joins IBM and NASA's Prithvi family of open foundation models, which already cover geospatial data, weather, and heliophysics. The Surya model, for example, predicts solar flares from high resolution solar observations. The shared philosophy: instead of building a new algorithm for every scientific question, start from a common foundation and adapt it. The lunar model plugs into the open source TerraTorch toolkit, with a companion paper on Hugging Face so anyone can reproduce and extend the work.
The deeper angle is open science as infrastructure. The model, the dataset, and the benchmarks are all public, built by teams across NASA centers, universities, and industry. When the tools for exploring the Moon belong to everyone, discovery accelerates in ways beyond what any single lab could plan alone.
For anyone watching the road back to the Moon, this release is a quiet milestone. The science of picking where we land, where we build, and where we refuel just got a serious upgrade, and the whole world gets to use it. The next giant leap may be computed before it is walked.
Sources
- Reuters: IBM and NASA launch lunar AI model
- IBM Newsroom: lunar foundation model release
- NASA Science: lunar foundation model
New to crypto? Read the crypto glossary, browse frequent questions, read our story, or explore the story archive.