NASA and IBM have released an open-source artificial intelligence model that can study the Moon’s surface, identify possible ice deposits, map craters, and examine volcanic formations. Announced on September 10, 2026, the NASA-IBM Lunar Foundation Model shows how specialized AI could accelerate scientific discovery and help researchers prepare for future human missions.
Unlike general-purpose chatbots, this model was developed specifically to understand complex lunar information. It combines decades of observations collected by different instruments, missions, and sensors into one reusable AI system.
What Is the NASA-IBM Lunar Foundation Model?
The NASA-IBM Lunar Foundation Model is an artificial intelligence system trained to analyze scientific data about the Moon.
According to IBM Research, the model combines tens of thousands of images and maps representing more than 30 aligned data layers. The information came from nine instruments associated with four lunar missions.
These sources include NASA’s Lunar Reconnaissance Orbiter and Gravity Recovery and Interior Laboratory mission. Data from Japan’s SELENE lunar mission was also included.
Traditional machine-learning systems are commonly trained for one narrow task. A foundation model learns broader patterns first and can then be adapted for several related purposes. Scientists can fine-tune the lunar model for research tasks without building a completely new system every time.
What Can the Lunar AI Model Do?
NASA and IBM initially focused on three important applications.
Detect Possible Ice Deposits
Permanently shadowed craters near the Moon’s poles may contain frozen water. These regions are difficult to study because sunlight rarely or never reaches them.
The AI evaluates several types of information together, including temperature, terrain slope, surface direction, maximum temperature, and estimated ice stability. In IBM’s testing, the model reduced the error rate for identifying promising ice locations by 22 percent compared with a specialized SwinV2 model.
Finding lunar ice matters because it could potentially provide water for astronauts. Its components could also contribute to oxygen production and rocket fuel. However, an AI prediction does not confirm that usable ice is present. Physical observations and future missions will still be necessary.
Map Lunar Craters
Accurate crater maps can reveal the Moon’s geological history and help mission planners evaluate potential landing areas.
The model matched a specialized crater-detection system when examining images at one-meter-per-pixel resolution. At the broader 100-meter context scale, it performed nearly 19 percent better while using half as much training data, according to IBM.
Better crater identification could help scientists find previously uncatalogued features, monitor new impacts, and locate hazardous terrain. It may also support the planning of roads, research stations, solar installations, and other future infrastructure.
Study Volcanic Features
The Moon contains unusual formations known as irregular mare patches. Scientists continue to debate when these volcanic features formed and what they reveal about lunar activity.
IBM reports that the foundation model performed 3 percent better than a task-specific comparison model when mapping these formations. This may appear like a modest improvement, but more reliable maps can help researchers decide where to focus limited mission resources.
Why This AI Model Is Different
The Moon has been observed through cameras, radar, spectroscopy, gravity measurements, and other scientific instruments. Each source produces information at different scales and in different formats.
The lunar model brings these measurements into a shared representation. It can search for relationships that may be difficult to identify when each dataset is studied separately.
Researchers adapted an AI architecture related to TerraMind, which IBM and the European Space Agency developed for Earth observation. They also used low-rank adaptation, commonly called LoRA, during fine-tuning. This method left 90 percent of the base model’s weights unchanged, reducing the computing required to adapt the system.
A 2025 IBM Research presentation described the project’s planned uses, including geological mapping, water detection, landing-site risk assessment, and research into young volcanic features.
Why Open-Source Access Matters
IBM and NASA have made the model, technical information, and benchmarking datasets publicly accessible. This gives universities, independent researchers, startups, and international scientific teams an opportunity to examine and extend the work.
Open access also supports transparency. Researchers can test the model on additional data, identify weaknesses, compare results, and develop new scientific applications.
The announcement does not mean AI can independently choose landing sites or confirm resources on the Moon. The results remain predictions that scientists must evaluate. Still, the system could reduce the time spent manually reviewing enormous collections of lunar imagery.
Reuters reported that the model identified important lunar features up to 23 percent more accurately than widely used methods in benchmark testing.
What This Development Means for the Future of AI
This project demonstrates that the future of artificial intelligence is not limited to text generation. Foundation models are increasingly being designed for specific scientific fields.
Similar systems could help researchers study weather, climate, oceans, planets, and the Sun. The greatest value may come from combining large volumes of previously disconnected data and directing scientists toward patterns worth investigating.
It is reasonable to expect more domain-specific AI models as scientific organizations make their datasets easier to use. That is an inference based on the growing use of foundation models in research, not a confirmed NASA timeline.
Frequently Asked Questions
Is the NASA-IBM lunar model open source?
Yes. IBM and NASA have made the model and related resources publicly available for scientific research and development.
Can the AI model confirm that water exists in a crater?
No. It can identify areas with conditions that suggest ice may be present. Additional measurements and physical exploration are required for confirmation.
How accurate is the lunar AI model?
Performance depends on the task. IBM reported a 22 percent reduction in error for ice prospecting and nearly 19 percent better crater detection at a broader mapping scale compared with selected benchmark models.
Will AI replace lunar scientists?
No. The system helps scientists process data, generate leads, and prioritize research. Experts remain responsible for validating results and making mission decisions.
Conclusion
The NASA-IBM Lunar Foundation Model is an important example of AI being used for focused scientific discovery. By combining decades of observations, it could help researchers detect possible ice, understand volcanic history, and create better maps for future exploration.
Follow Tech Ketchups for more clear and practical updates about artificial intelligence, machine learning, robotics, gaming, education, and emerging technology.
