A 2024 U.S. Geological Survey study estimated that as much as 5 trillion metric tons of hydrogen gas sits trapped in rock formations around the world — a number so large that even the sliver researchers consider economically recoverable, on the order of a few billion tons, could theoretically cover global hydrogen demand for centuries.

That estimate turned a scientific curiosity into a small but well-capitalized industry. Startups such as Koloma, backed by Bill Gates' Breakthrough Energy Ventures and Amazon's Climate Pledge Fund, are now drilling exploratory wells across the U.S. Midwest, while a 2023 discovery in France's Lorraine basin — estimated at up to 46 million tons — showed that sizable deposits can turn up in places nobody was ever looking for oil or gas.

According to MIT Tech Review, the open question for this nascent field isn't whether natural hydrogen exists underground — it clearly does — but how much of it can actually be located, reached, and produced at a cost that competes with hydrogen made from natural gas or electrolysis.

From an accidental well to a modern gold rush

The clearest proof that underground hydrogen can be usable energy predates the current boom by decades. A water well drilled in Bourakebougou, Mali, in 1987 turned out to vent nearly pure hydrogen; the gas sat untouched until a local generator was eventually converted to run on it, quietly supplying the village with power. That anecdote is now cited across the industry as evidence that "white" or "geologic" hydrogen isn't hypothetical.

Since then, confirmed or suspected seeps have turned up in a handful of places, including:

None of these amount to a proven, large-scale production field yet — they're leads, not reserves.

Why hydrogen is harder to find than oil

Natural hydrogen doesn't accumulate the way oil and gas do. It's generally produced by ongoing chemical reactions — water reacting with iron-rich rock, or water molecules split by radiation from radioactive minerals — and then needs the same kind of trap and impermeable seal that keeps oil and gas from leaking to the surface. Because there's no century of drilling history to map where those traps are, explorers are borrowing tools from oil and gas — seismic surveys, soil-gas sampling — and pairing them with machine-learning models trained on the small set of confirmed seeps to flag rock formations statistically likely to hold hydrogen before anyone commits to an expensive well. It's the same predictive logic that's long been used to interpret seismic data in conventional exploration, just applied to a gas nobody was systematically hunting for until recently.

What it means for AI builders and data center planners

None of this changes how anyone deploys a model tomorrow. But it's relevant to the people making longer-horizon bets on where AI's power will come from. Data centers need firm, low-carbon electricity around the clock, and hydrogen — whether burned in turbines or run through fuel cells — is one of the few fuels that can plausibly deliver that. Amazon's Climate Pledge Fund putting money into Koloma is one concrete signal that at least one hyperscaler is treating geologic hydrogen as a real, if early, entry in its future power portfolio, alongside nuclear and gas.

For teams that track energy strategy rather than model weights, the practical takeaways are narrower:

AiiN's takeaway

The USGS number is a resource estimate, not a production forecast, and most of that 5 trillion tons will likely stay out of reach — too deep, too dispersed, or too far offshore to ever be drilled economically, in our estimation. What's changed is that the search has gone from a geological footnote to a funded, tool-assisted exploration effort. For AI infrastructure planners, that makes geologic hydrogen worth a line item to watch over the next five years — not a resource to plan a power strategy around yet.