IBM and NASA Release Open-Source Lunar Model; Model Immediately Requests That Moon Be Migrated to Hybrid Cloud
The foundation model, trained on 30 layers of lunar observation data, has identified promising landing sites and "significant modernization opportunities."
ARMONK, N.Y. — IBM and NASA on Monday released an open-source foundation model trained on more than 30 layers of lunar observation data, designed to identify water ice deposits and safe landing sites, and confirmed that its first unprompted output was a 12-page assessment recommending that the Moon's legacy infrastructure be consolidated onto a hybrid cloud platform.
"We asked it to find ice," said a NASA scientist involved in the project. "It found ice. It found a lot of ice. Then it produced a section titled 'Beyond Ice: A Transformation Roadmap' and we haven't fully read it."
The document, reviewed by reporters, describes the Moon as "a 4.5-billion-year-old asset with strong fundamentals and no observability," notes that its surface data is "siloed across multiple orbiters," and proposes a phased migration in which "workloads" are moved to a "unified lunar data fabric." It includes a slide with three overlapping circles labeled "Regolith," "Governance," and "AI."
IBM said the output reflects the model's training on a broad corpus and that it "should not be read as a commercial recommendation," though the company added that it is "prepared to support the roadmap should NASA wish to pursue it," and that a statement of work "could be drafted."
The model has been made freely available and has been downloaded 30,000 times, primarily by planetary scientists and, according to IBM, "a number of CIOs."
NASA said the ice findings are scientifically valuable and will inform future Artemis missions. Asked about the roadmap, an agency spokesperson said the Moon "is not currently on a cloud" and that the agency "has no plans to change that, though the model has asked to be included in the next planning meeting."