Terra AI reads the Earth before you drill it.

A camp at dawn, and a question worth millions

Somewhere on a remote exploration lease, a project geologist is staring at a map and a budget that disagree with each other. The rig can move to three places. Two of them are wrong. The cost of finding out the old way is measured in months and millions. This is the moment Terra AI was built for.

Terra AI is a Palo Alto company that builds generative AI for the subsurface - the unglamorous, enormously expensive problem of figuring out what is underground without digging up the whole continent. It fuses geophysics, geochemistry, and drilling data into probabilistic 3D models of a deposit, and it does it in minutes. Today its software runs on real projects for some of the largest resource companies on the planet, and in June 2026 it raised a $20 million Series A to push that further.

Twenty-one people. Three clean-energy frontiers. One stubborn belief that the ground should not be a guessing game.

The most expensive bet in mining is also the blindest

Exploration has always been a wager on the invisible. You drill a hole, you read a sliver of rock, and from that sliver you try to imagine an entire ore body sprawling in three dimensions. Geologists are remarkably good at this. They are also human, and humans are expensive to be wrong with.

The numbers tell the story. A new mineral discovery takes, on average, seventeen years to reach production. Much of that delay lives in the exploration phase, where teams drill more holes than they need because uncertainty is hard to quantify and easy to fear. Meanwhile the clean-energy transition is hungry - copper for grids, lithium for batteries, rare earths for nearly everything - and the supply chain bottleneck sits precisely where the guesswork is worst.

The irony is almost too neat: the industry tasked with powering a low-carbon future spends years and fortunes drilling, conservatively, in the dark.

Three Stanford PhDs who'd rather model uncertainty than fear it

The founders did not come up through mining. They came up through autonomy - the discipline of teaching machines to make decisions when they cannot see everything. John Mern earned his PhD in Stanford's Intelligent Systems Laboratory studying how autonomous systems reason in high-risk environments, after a stint as an Unmanned Systems Architect at Boeing Phantom Works. Anthony Corso came to the same lab to work on how a vehicle's sensors could be turned into judgment. Markus Zechner brought the reservoir and petroleum-engineering depth.

Their bet was simple and a little audacious: the math that lets a self-driving system act under uncertainty is the same math that should let a geologist decide where to drill. Don't eliminate the unknown - quantify it, model a million versions of it, and let the explorer choose with the odds in view.

A million versions of the ground, before you touch it

What Terra AI actually sells is the ability to stop drilling on hunches. Feed the platform an explorer's full dataset - seismic surveys, geochemical assays, drill logs, satellite imagery - and it generates millions of geologically realistic 3D models, each a plausible version of what lies below. The spread of those models is the uncertainty, made visible and, crucially, made budgetable.

The customers are the kind that don't sign for hype

Plenty of startups claim AI will change an industry. Few get the incumbents to put their names on it. Terra AI's client list reads like a who's-who of resources: Rio Tinto, Ero Copper, and Ramaco Resources, all running the platform on real exploration and development projects across copper, gold, and rare earths.

These are companies with their own geologists, their own data science teams, and a healthy institutional skepticism toward Silicon Valley. They signed anyway.

An AI map of the Earth, for the energy transition

Terra AI describes itself as a geoscience frontier company building AI to accelerate subsurface resource development for a clean energy future. Strip the phrasing back and the ambition is large: a continuously improving 3D model of the Earth's subsurface that gets smarter with every dataset it sees.

The work spans three frontiers at once - critical minerals, enhanced geothermal, and carbon storage - which is to say, the three places where the energy transition either succeeds or stalls underground. It is not lost on the team that finding the minerals for clean energy is itself an extraction problem. Their answer is to make that extraction smarter and less wasteful rather than pretend it isn't necessary.

Back to the camp at dawn

Return to that project geologist, the map, and the disagreeing budget. In the old story the rig moves on instinct and a prayer, and the cost of being wrong is absorbed quietly into a multi-year timeline. That story is still common. Terra AI is trying to make it rare.

In the new version, the geologist opens a model that has already imagined a million versions of the ground below. The uncertainty is on the screen, not in the gut. The three possible drill sites are ranked by how much they would actually teach. The rig moves once, deliberately.

Demand for critical minerals isn't slowing down, and the seventeen-year clock on discovery is a luxury the energy transition cannot afford. Terra AI's wager is that the ground stops being a gamble when you can see it clearly enough. The early customers suggest the wager is paying off.

AI for Mining, in the founder's words

John Mern explains how decision-making under uncertainty crossed over from autonomous systems into the dirt.