Hydrogen Beneath our Feet

The Earth naturally produces hydrogen through geological processes in the subsurface, but finding it is the hard part. We built a multimodal AI foundation model that fuses gravity, magnetics, and geology data to learn a rich representation of the subsurface, then predicts where hydrogen-generating rocks are most likely present. Applied to California’s Salton Trough, our model flagged three anomalous zones within an active continental rift characterized by shallow Moho, gabbroic intrusions, and high heat flow. To validate the predictions without expensive drilling ($1M+ per well), we collaborated with the U.S. Geological Survey on a field sampling campaign. Soil gas measurements at model-predicted locations confirmed hydrogen concentrations exceeding 200 ppm, over 400x atmospheric levels and 10–20x local background. This work demonstrates that AI-guided surface geochemistry can serve as a scalable, low-cost pathway to discovering natural hydrogen resources.