Terranox AI Is Using 70 Years of Exploration Data to Find Uranium Deposits That Humans Missed

The Macro: Nuclear Energy Needs Fuel and Nobody Is Finding It Fast Enough

The nuclear energy renaissance is real. New reactor designs from companies like Kairos Power, X-energy, and TerraPower are progressing through regulatory approval. Data centers are signing power purchase agreements with nuclear plants. Countries are including nuclear in their clean energy strategies. The demand for uranium is growing.

But uranium exploration has barely changed in decades. Geologists still analyze drill core samples, study geological maps, and use geophysical surveys to identify potential deposits. The process takes years and costs millions per prospect. Most exploration programs fail to find economically viable deposits. The hit rate is low because the geological signals are complex, subtle, and scattered across disparate data sources.

The irony is that 70 years of uranium exploration in North America has produced an enormous amount of geological data: drill logs, geochemical assays, geophysical surveys, and exploration outcomes. Most of this data sits in government archives and company files, never analyzed comprehensively. A geologist working on a new prospect might reference a few dozen historical data points. The full dataset contains millions.

Terranox AI, backed by Y Combinator, is the first company to apply machine learning to this historical treasure trove. They are training AI on seven decades of exploration outcomes to predict where high-grade uranium deposits are hiding.

The Micro: PhDs Who Know Both Rocks and Code

Jade Checlair (CEO) holds a PhD in geophysics from the University of Chicago, conducted research at NASA Ames, and worked at BCG in mining and energy strategy. Leeav Lipton (CTO) spent 8+ years as an AI/ML scientist at Borealis AI, worked at NASA JPL on remote sensing, and has a background in astrophysics. Two founders who can bridge geoscience and machine learning is exactly the team this company needs.

The “vertically integrated” descriptor is important. Terranox is not selling AI tools to mining companies. They are an exploration company that uses AI to identify targets, then acquires mineral rights and conducts exploration themselves. This means they capture the full value chain from discovery to development rather than being a software vendor to an industry that is slow to adopt new technology.

The AI is trained on geological, geochemical, and geophysical data from historical exploration programs across North America. The model identifies patterns in subsurface data that correlate with high-grade uranium deposits. Patterns that might be invisible to a human geologist looking at one prospect at a time become apparent when analyzed across thousands of historical data points.

Competitors include traditional uranium exploration companies like Cameco, NexGen Energy, and Denison Mines, as well as AI-in-mining plays like KoBold Metals (which focuses on battery metals). KoBold has validated the AI-powered exploration model with backing from major investors, which creates a favorable market narrative for Terranox.

The North American focus makes strategic sense. The US and Canada have extensive historical exploration data, established mining regulations, and growing domestic demand for nuclear fuel. Geopolitical tensions around uranium supply chains make domestic production a priority.

The Verdict

Terranox AI is applying proven AI techniques to a high-value, data-rich problem in an industry that is structurally underinvesting in exploration.

At 30 days: has Terranox identified specific exploration targets using their AI models, and have they acquired mineral rights?

At 60 days: are the AI-identified targets showing geological indicators consistent with the model’s predictions? Early drilling results are the ultimate validation.

At 90 days: is Terranox raising capital for exploration programs based on AI-identified targets? The path from prediction to drilling requires significant capital.

I think Terranox is building in a strong tailwind. Nuclear energy demand is growing. Uranium supply is constrained. And the historical exploration data exists to train meaningful models. If their AI can identify deposits that traditional methods missed, the value creation is enormous. The vertical integration model means they do not need to convince skeptical mining executives to buy software. They just need to find uranium.