NASA and IBM have jointly released an open-source AI model of the Moon, designed to help researchers make new discoveries about Earth's natural satellite. The NASA-IBM Lunar Foundation Model, now available on Hugging Face, is described as the first AI model to combine lunar observations captured in multiple data formats, viewing angles and spatial scales, allowing scientists to analyse geographic features without manually sifting through maps and low-resolution imagery.

The tool is intended to help researchers locate previously unidentified lunar ice deposits, which could support future crewed missions by indicating the presence of water and oxygen, as well as analyse volcanic features called Irregular Mare Patches and classify craters. It draws on an accompanying open-source dataset of over 30 spatially aligned layers, built from data gathered by nine instruments across four missions. The release extends IBM and NASA's Prithvi family of models, following earlier collaborations including a satellite imagery model in 2023, a climate model in 2024, and a solar flare prediction model named Surya last year. Neither organisation has disclosed the model's parameter count, though IBM said fine-tuning was largely done on Nvidia A100 GPUs, while lighter inference tasks may run on more modest hardware.