Navigating uncertainty

Fifteen years. That is the average amount of time it takes to go from a mineral discovery to an operational mine.

“Our goal is to bring that down to under a year,” says John Mern, MS ’18, PhD ’21, cofounder and CEO of Terra AI, a startup that is accelerating the underground modeling process for minerals, geothermal energy, and carbon storage. “We want to do for minerals and exploration what AI is doing for drug discovery, and become the default standard for the industry.”

Developing a subsurface resource usually means collecting data from several sources, including soil and rock sampling, fluid flow measurements, as well as geophysical and geochemical surveys. A company searching for copper, for instance, might run an electrical current through the ground to observe its conductivity and compare that with other data to estimate the metal’s location.

Geologists use this trove of information to make an educated assessment about where to drill next in order to further model the geology and access the hidden resources below. However, the process is lengthy, requiring significant capital and trial and error.

Terra AI uses these traditional inputs to produce up to hundreds of thousands of potential 3D subsurface models that match the possibilities in the ground. They then advise clients on which models have the greatest probability of depicting the subsurface as it really is. The approach still has uncertainty, but it is quantified, better-informed—and more than twice as fast.

In a client project last year, the startup’s platform produced resource models that were two to three times more accurate than the client’s usual approach—translating into 60 percent fewer drilling meters throughout the exploration process and lower associated costs.

“It’s sort of counterintuitive that exploration happens after discovery,” says Mern, but he explains that a discovery drillhole is only 4 inches in diameter.

“It’s very, very small when you’re trying to measure something that’s 500 meters in each direction and you have to know exactly where it is in three dimensions, and how consistent it is, and where the edges are.”

From renewable energy and electric vehicles to the ever-expanding domain of AI and its appetite for electricity, minerals are critical to wiring an electric future. Mineral shortages could create bottlenecks in these technology transitions, whether that’s due to actual mining deficits or geopolitical jockeying.

For example, the S&P Global projects a potential shortfall of 10 million metric tons of copper by 2040 without expanded supplies. So, the hunt continues to find new sources for copper, lithium, nickel, and cobalt, as well as graphite, manganese, and a sweep of magnetically useful “rare-earth” minerals.

“A lot of the work we’ve done at Terra AI has been inferring what is underground, inferring things we cannot see based on things we can measure,” says Mern of how they’re partnering with companies trying to close the gap. “Ultimately, the value we unlock is the ability to do that rigorously in a very difficult to observe environment: the subsurface.”

Self-driving to mining

The Terra AI cofounders met in graduate school in 2016, at a time when Mern says it felt like there was a new self-driving startup every week. He and Anthony Corso, MS ’16, PhD ’21, were members of the Stanford Intelligent Systems Laboratory (SISL), run by Associate Professor Mykel Kochenderfer. 

Mern says, “I had started out, like a lot of folks in our lab, focusing on autonomy and self-driving vehicles.”

Both of their PhDs were related to decision-making in autonomous systems. Mern studied Monte Carlo planning, an AI method that uses randomized simulations to anticipate future scenarios on a large scale. Corso researched safety validation for robotics, aviation, and self-driving cars. 

Calculating risk was the throughline from driving a car to aiming a drill-bit. In an autonomous car, the navigation system must do more than estimate where other vehicles are located on the highway—it must also calculate an envelope of uncertainty.

“Here’s where it might be. Here’s how fast the car might be going. Here’s the range. Here’s what it might do in the next five seconds, some outliers and things like that,” says Mern.

Which is hard enough to do on the road, but could it work below the soil?

“When you’re looking at an ore body that is entirely under tens to hundreds of meters of rock, that uncertainty envelope is much harder.”

And yet, the methodologies matched up.

Markus Zechner, PhD ’19, who is now the general manager of reservoirs for Terra AI, was the first to tip them off to this connection. Zechner was studying energy resource engineering at Stanford, and he pitched a collaboration with the SISL team to apply their approaches to reservoir field development.

“That was my first foray into natural resources,” Mern says. “The more I read about the implications and eventualities with the copper supply and lithium supply, the more I thought, oh, this is a problem with a capital P.”

Step-function change

“Starting a company around this concept felt necessary in order to take the technology we were building out of the lab and ready to deal with real-world complexity,” says Corso, who is cofounder and chief technology officer.

With an Innovation Transfer Grant from the TomKat Center, the cofounders developed their proof of concept to where they could connect with industry partners, gaining access to decades of mining data.

“Back-testing was really valuable to us,” says Mern. “Companies have decades worth of data per project that we could go and evaluate against and compare us in a blind test setting. It allowed us to basically accumulate decades’ worth of experience without spending decades in the field.”

Their early results pointed to the potential for 20 percent to 30 percent greater efficiency, which meant fewer meters drilled to arrive at the same conclusions.

In a year, that number had doubled.

“That’s massive when you talk about timelines that are measured in years per mine.”

When Terra AI incorporated in 2023, the cofounders charted a course for growth. “Early on, we had conviction that we didn’t want to take an incremental approach.”

In less than three years, the startup announced a $20 million Series A round led by Khosla Ventures and joined by the venture arm of Australian mining giant BHP, with hopes of bringing new copper resources to market in the coming months.

“The TomKat Center grant allowed us to have more freedom to explore a variety of different technical approaches,” Mern says. “Not having the grant would have, I think, drastically slowed down the pace of innovation and the overall scale of impact that we were able to achieve to date.”

This article is part of the TomKat Center Spotlight series designed to highlight the impact and trajectory of the work of faculty and students who received funding through our Innovation Transfer Program, TomKat Solutions, and Graduate Fellowships. Stanford University does not endorse any non-Stanford entities, programs, products, or services listed in the article.