They drilled where nothing on the surface hinted at treasure — and hit heat.
Zanskar Geothermal & Minerals, a Utah startup, announced this month that artificial intelligence helped it locate a 250°F geothermal reservoir beneath a scrubby stretch of western Nevada. The site, nicknamed “Big Blind,” is a classic blind system: no hot springs, no steam vents, no surface drama. Just heat, quietly stored thousands of feet below ground.
How the machine saw what humans missed
Finding geothermal that’s ready for power production has long been a needle-in-a-haystack problem. In the 1970s and 1980s oil companies spent heavily searching for hidden systems, but the high costs and low hit rates drove many away. Zanskar’s co-founders say AI has changed the odds.
Their models were trained on historical examples — many blind systems were discovered accidentally over the past century by oil and gas drilling — and then fed vast, varied datasets: rock composition, magnetic surveys, gravity anomalies and other geophysical fingerprints. The result is a pattern-recognition engine that teases signal out of noise and points to likely reservoirs.
Drilling this summer to about 2,700 feet confirmed porous rock at roughly 250°F, enough heat to supply utility-scale turbines. Zanskar says the field is at least large enough to support a power plant and could produce electricity in three to five years if permitting, financing and grid connections fall into place.
The company’s approach is part geological detective work, part data science. As machine learning tools become more capable — and as the industry borrows techniques from the oil and gas world — finding concealed resources is getting cheaper and faster. These developments echo broader advances in AI tooling and agentic systems that are reshaping how complex, multi-step tasks get automated how agentic AI is expanding in consumer tools. The same surge in model capability that lets systems analyze reams of geophysical data also powers new search and research features in productivity tools like Gemini Deep Research.
Why this matters (and what could slow it down)
Geothermal has long been an attractive renewable: continuous baseload power, minimal direct emissions, small footprint once a field is operating. But the stumbling blocks have been discovery risk and upfront drilling cost. If blind systems account for the bulk of undiscovered resources — estimates cited by geoscientists suggest more than three-quarters in the U.S. — better ways to find them could unlock tens or even hundreds of gigawatts, particularly across the Western states.
Experts temper enthusiasm with realism. Drilling confirms heat; it doesn’t automatically equal an economic, long-lived power plant. Developers must map the reservoir, estimate flow rates, secure water rights where relevant, thread through permitting, and build transmission. There are also environmental and regulatory conversations to have — for instance, next‑generation methods that enhance or create permeability (sometimes compared to techniques used in oil and gas) raise questions about induced seismicity and water use.
Still, the timing is favorable. Political and market attention to geothermal has grown even as other technologies grapple with policy shifts. And demand is only rising: data centers, which underpin the modern AI boom, gobble electricity and prize reliability — a perfect fit for baseload geothermal.
Zanskar isn’t relying on exotic engineering. Its discovery suggests that “conventional” geothermal — finding naturally occurring hot reservoirs and tapping them — still holds large potential. At the same time, the industry is experimenting with engineered geothermal systems that drill deeper and create permeability where nature didn’t provide it. Both paths could expand supply if costs and environmental hurdles are managed.
What to watch next
- Replication: Can Zanskar’s AI reliably locate multiple blind systems with the same success? The company says it has already flagged several hotspots across the West.
- Economics: Will drilling and development costs stay low enough to attract utility-scale investment? If they do, a cascade of projects could follow.
- Grid integration and permitting: Even a ready field needs transmission access and regulatory green lights; those often take years to align.
This discovery sits at a crossroads of old and new: decades of geological knowledge meet modern machine learning. If the pattern holds, deserts that look empty from above could soon be sending steady electricity to grids — quietly, reliably, and on a timetable faster than many expected. Advances in model architectures and tooling — the same sort of progress that has produced new image and multimodal AI systems — continue to lower the friction for applied problems like this one see how new models are changing creative tooling.
Zanskar’s Big Blind is a reminder that sometimes the next big breakthrough in energy isn’t a shiny new machine; it’s the ability to see what’s already been there, hidden in plain data.