Mining the minerals to power AI

0:00Hi, everyone. It's Rana. The future of AI is not built in isolation. It is shaped in rooms where ambitious people challenge each other. Masters of Scale Summit is back October 20th through 22nd in San Francisco, bringing together founders and innovators pushing the edge of what is possible. If you're building what comes next, you should be in this room. Apply now at mastersofscale.com slash pioneers. That's mastersofscale.com slash pioneers.

0:35Josh Goldman:On April 4th, 2023, around two in the morning, a man was found stabbed multiple times on a sidewalk in downtown San Francisco. Hey, who did this to you? What happened next turned the story into a political firestorm. Reports have identified the victim as Bob Lee, the founder of Cash App. From Bloomberg Podcasts, this is Foundry, The Killing of Bob Lee, beginning April 16.

1:06An economy that is powered by AI and batteries, by mid-century, we will need to mine more copper. over the next 25 years than has been mined in all of human history. Why can't other exploration companies take a similar approach to what cobalt is doing? The amount of capital that you have to deploy, or that the industry is deploying on average to make a discovery, has grown more than tenfold. It used to be that on average, in base metals exploration, about$80 million worth of exploration projects, mostly failures, would produce a winner, a project that eventually is going to become a mine. And today it's well over a billion dollars.

1:48A billion dollars of failures to produce a winner.

1:53That's Josh Goldman, co-founder and president of Kobold Metals. They're using AI to find the critical minerals that will shape our future. And they're backed by some of the biggest names in tech, like Jeff Bezos and Bill Gates. These minerals are essential to electric vehicles, robotics, semiconductors, and the infrastructure behind the AI economy. They're also at the center of geopolitical competition, as countries race to secure the resources that will define the next era of power and progress. In this episode, Josh and I talk about how AI is transforming one of the oldest industries in the world.

2:34Why critical minerals have become so strategically important? And whether mining can be done in a more ethical, more sustainable way that is better for people and for the planet? And on Earth Day, that last question feels especially important. So let's dig in. I'm Rana El-Khalyubi, and this is Pioneers of AI, a podcast taking you behind the scenes of the AI revolution.

3:06Thanks for joining us on Pioneers of AI. Pleasure to be here. Thank you. So a few months ago, a common friend of ours, Chris Schoder, texted me and he said, you got to meet this guy. And he was right. So shout out to Chris. And full disclosure, I'm an investor in Cobalt. I love what you guys are doing and very proud to be an investor and support you on this journey. Very glad to have you as a partner. Yeah. So I want to start with your origin story. You have a PhD in physics and you spent some time in private equity in the oil and gas industry and then decided to start an AI company or an AI and data company in the mining and exploration space.

3:42How did that come about? I'll start with my background as a scientist. I studied physics just because I thought it was the most interesting thing I could work on. It had the hardest problems and I got to think about the very big, like the origin of the universe, and the very small, like atoms, and the things that make up all the things around us. You can't hang up your hat as a physicist just by doing an undergraduate degree. You don't even reach the frontier of science. You learn mostly what happened up through about 100 years ago in an undergraduate degree. So I went on to graduate school. I actually went to the UK for a couple of years first.

4:15I did a master's program in applied math and theoretical physics and then another master's program in history of science. I had Cambridge University and then Imperial College. I was very fortunate. Did we overlap with Cambridge? 2002-3 is when I was there. Yes, I was there 2000-2005 for my PhD. Really? Yes. Yes. Go figure. We were inhabiting the same place at the time. I had a Marshall scholarship, so I had two years of funding to go study in the UK. And I came back and did a PhD, and I worked on quantum computing. And just because the science is really interesting, and that's still the case. Nobody does a PhD in the sciences just for some kind of careerist reason.

4:54You do it because you're just interested in the material. You're curious. Yeah, you're curious. And that is, throughout my career, one of the things I've always been solving for among the most important things to me is just satisfying my own intellectual curiosity and learning, and learning lots of things, but learning science is one of the most satisfying things one can do. And I knew I wanted to focus on science problems that were really relevant to modern society, and that was focused on energy and sustainable energy. I met my co-founder, Kurt House, as a graduate student. He'd started a reading group for energy nerds at Harvard, and then we did things like organized trips to power plants and coal mines and things in our spare time.

5:32How fun. Exactly. This is what nerds do in graduate school and in adult life. And I think deep grounding in the sciences just makes you a really good thinker and a really good problem solver. And so much of what we're doing at Kobold is applying first principles thinking to really challenging problems. Okay. So you went on to have a career in oil and gas and then at At some point, you were like, enough of that. I wanted to get to know the incumbent energy industry really well. I worked as a consultant. I went to McKinsey in Houston, and I worked with power companies and oil and gas exploration and production companies.

6:05I worked with equipment manufacturers who make power equipment, everything from wind turbines to subsea oil field equipment and things in between. And I worked on strategy questions and advanced analytics questions. and then started working on private equity investment in oil and gas with my co-founders of Cobalt. And after doing that for a short while, we decided we're going to stop working on this. We want to work on the materials of the future. There's much better business opportunities there, and that's more personally motivating. And we thought from first principles about what we should be working on and thought about, well, what are the raw materials that the future economy is going to need.

6:44And those are materials like copper and lithium. Now, why do we need lithium? You need lithium for batteries. If you want to make a device that you want to pick up and move around, whether that's a car or a truck or a drone or a robot or an airplane, you want a vehicle that is durable and has long range. And so lithium wins. It is the lightest and the most electropositive metal. And we have the periodic table we have. We're not going to invent a new element. And so if we're going to have an economy that's powered by these battery-powered devices, by mid-century, we need lithium production rates to be 10 times what they are today.

7:20Wow. There's a very different material profile. If you recycle an internal combustion engine car, you can't just turn it into an electric vehicle because it's not made out of the same materials. You need something like five kilograms of lithium in a typical EV. There's no lithium at all in a conventional combustion engine vehicle. And so we need new materials. And the sort of big trends of materials like copper, so if you want to move electrons around, copper is the metal of choice. And again, we have the periodic table we have. We're not going to invent a new element. We're not likely to invent a plant that is cheaper and more electrically conductive than copper.

7:54And so we need this obviously for broader electrification, but also for AI. All of these data centers need power, and we need the power transmitted to the data centers, and we need the data centers themselves built. All of this drives demand for copper. And so an economy that is powered by AI and batteries, by mid-century, we will need to mine more copper over the next 25 years than has been mined in all of human history. Wow. So these are incredible demand tailwinds. These are the things that we need more of. And well, where are they going to come from? And we started to work on that problem and started pulling on the threads that led to us developing cobalt.

8:32Very cool. So for the people who are not geologists or in the mining industry like me, I want to paint a picture to our audience where these resources are found. And let's dig into that a little bit. And I actually have a visual prop that I want to bring up for my next question. Okay. You've compared mineral deposits to a certain bakery item. Yes. So we are going to bring that in. I love it. Thank you, Steph. All right. So tell us, what are we looking at here? What is this? Okay, we're looking at raisin bread. Okay. Okay, so what Ron is referring to here, so there's one of these myths is about sort of scarcity.

9:08Like, okay, is there enough copper? Is there enough lithium? The vast majority of the ore deposits that contain these metals are still out there to be found. How do we know this? We know this because we know how these ore deposits get made, or we know enough about them to know, we know the pressures and the temperatures that they format. You can actually, you can determine those things by looking at the minerals themselves. You need some microscopes and things. Those are pressures and temperatures that are formed deep under the earth, thousands of meters below the surface. But the vast majority of the ore deposits that we've found are right at the surface, or they're really close.

9:42They're literally sticking out of the ground, okay? And so the way to think about this is like an ore deposit's like a raisin, and the eroded surface of the earth is like slicing off the end of the loaf of raisin bread, okay? Because so the ore deposits are at different depths, but the surface of the earth is an eroded surface it's like that exactly that's right so you sliced off the top of the loaf and you can only see the raisins on the surface where you just cut but if you know the recipe you know there's many many more that you can't see and so the rest of them are out there to be discovered so another misconception so copper and lithium are not rare earths rare earths are a suite of elements if you're looking at the periodic table like mostly the lanthanides and the actinides, people commonly think of, oh, well, rare must be valuable.

10:29It's like, well, no, rare earths is just a chemical name for this set of important elements. Ore deposits themselves are all very rare, and the earth's crust is very, very big. So there are many out there, but they occupy a very small fraction of the earth's crust. So talk us through, like, historically, how do you mine these ore deposits? Yeah, so before you mine it, you have to find it. And so the finding is the hardest part. And so it's not the case that we just know where the minerals are, and it's a matter of digging them up. Mostly, we just don't know where the minerals are. And historically, the single most important method for finding ore deposits has been humans walking around looking for strange-looking rocks, like actually prospecting, like walking across the surface of the Earth.

11:11And so you can think of copper minerals, wind and water expose the rocks. The water and oxygen turns them blue and green. Think of the patina of the Statue of Liberty. And so in places where there's a lot of copper, like in the Central African Copper Belt, there's sort of blue and green rocks in different places. And you can see them. But then you start to deplete this inventory of things that stick out of the ground. And certainly for copper deposits, anywhere in the world that's known to host copper, if you can just find it sticking out of the ground, like somebody found it already. Now, there could be deposits that people have never looked for before that could still be sticking out of the ground.

11:47So lithium, for example, people haven't been looking for lithium until very, very recently because we didn't need big chunks of it. But for copper deposits, what's left to find is still harder to find. It's concealed somehow. It's either got, you know, maybe several layers of rock above it, potentially hundreds of meters of rock. In the case of our Mingomba project in Zambia, more than a thousand meters of overlying rock. Key thing is they're not easy to detect with one data set. You can't just look at a satellite image and see a kind of bullseye pattern of how the rocks have been altered by the flow of fluids that carry the copper in.

12:22You can't just take a bunch of soil samples over a grid and make a plot of the copper in the soil and drill the things that are hot. You can't just fly an airborne survey measuring magnetics or gravity and be like, ah, here's a red spot, I'm going to stick a hole in that. The data is very high dimensional and you need a combination of signals in all of these different data types together in order to predict where you're going to find ore. And that's a hard science problem. And that's the kind of thing that we work on. One of the things that I when I first kind of learned about COBOL that I really loved and it's super aligned with our investment thesis at Blue Teal Adventures.

13:01It's basically this idea of a trifecta of sensors, multimodal data and then both generative and predictive AI. Yeah, I love the way you frame that. The technology that we use, there's no one product that is a black box that predicts ore deposits. There's no one product at all. What we're trying to do is make better predictions of all of the composition of the rocks. Performantly, we're collecting many terabytes a day of data. And so you need to be able to interact with that data. Index it, query it. That's right. And then you've got to be able to put it into models and combine it with other kinds of data.

13:34and then we use that. There are many models that are trained on that data where we then go make predictions. And so we will take data like that and we'll use a combination of historic archive data. We can talk more about sort of old paper records and public data. I would love to do that. I'm fascinated by archival data. Yeah, it's so fun. So we're taking models that are trained on that data and then we're using them to make predictions on a given project. Where should we go next on this project? but also to make predictions in other continents. And data that we've used in Canada has helped us find the right rocks for hosting lithium deposits in Australia, because the features of the ore deposits are in common.

14:13It's the same geological process. It's the same recipe to cook the raisin bread. And we can see when we see those processes at work in one place, then we can use that. The model learns how to recognize those features and can recognize it on new data in totally unrelated domains. And this is the whole point of the company, which is that anything that we do doesn't just help us execute this project. We have better models. We have new technologies that allow us to execute the next program better and better and better. So each thing that we do makes the whole system more and more capable. Let's take a short break, and we'll be back to dive more into the sources of Kobold's data.

15:27Transcription by CastingWords Loom, and now AI agents in Rovo, which connects the dots across your work so nothing gets lost. It's one AI-powered teamwork platform designed for how modern teams actually build. Learn more at Atlassian.com slash TeamChanger. That's A-T-L-A-S-S-I-A-N dot com slash TeamChanger.

15:55Josh Goldman:In our new podcast, Everybody's Business, we talk about the business news that concerns everybody. From Bloomberg Businessweek, I'm Stacey Vanek-Smith. And I'm Max Chavkin. Each week, we unpack what is happening on Main Street and Wall Street, all the streets. WrestleMania has taken over the U.S. economy. A poetry that executives write on LinkedIn. A little actual magic in our underrated story of the week. The single-grade marketing campaign the music business has ever seen. I decided to ask people how they felt about the penny going away. Listen to Everybody's Business wherever you get your podcasts.

17:02So let's talk about this historic data and the archives. Like, where do you get that data from? What does it look like? How do you digitize it? Yeah. The vast majority of data ever collected about the Earth is in the public domain. But it's incredibly fragmentary. There's, I think, tens or hundreds of thousands of different repositories. In many jurisdictions, the data of prior explorers is all public. Wow. So again, I refer to Quebec. I refer to Finland. We're a big explorer in Finland. You have to write a technical report on what you did every year, like 100-page documents describing what were we looking for and what had people done before us and what activities did we undertake and what were the results of those activities.

17:44We also have to deposit copies of the raw data with the regulator, and then that goes into the public domain. And so there are hundreds of thousands of historic drill holes in these provinces. Sometimes an individual data set might be a folder with a dozen or so files, and it might represent one drill hole. And these things are differently formatted. They're not all in the same language. There are all kinds of different mistakes. A lot of the data is in the text of these reports, and it's in the figures and hand-drawn maps and tables in these things. So since we started the company, we have been both aggregating this information, extracting structured data from it, providing tools to our explorers to work with unstructured data, and putting it all in a system where we can interact with this entire corpus of data at the same time.

18:33So that's very powerful. And the latest developments in frontier models just accelerate this and build on all the infrastructure work. Because it allows you to bring all this unstructured data together. Yeah, that's right. It allows us to work with much more of this data much faster. And some of the data is still in paper form. And so in Zambia, for example, we have our first major discovery is in Zambia. There are historic archives that include maps that are almost 100 years old that are hand-painted on linen. How cool is that? They're so cool. And they're really— And they're just in some governmental office somewhere in Zambia.

19:06That's right. And so like the Geological Survey Department of Zambia has been the caretaker of a bunch of these records. And so they are accessible. You're allowed to go in and see them. You're allowed to sort of – and the – Unroll the map and look at it. Exactly right. Yeah, that's right. And the staff who work there care enormously about the data. They're the conservators of this legacy. And it is data. It's extraordinary data. Relevant still? Absolutely. The rocks haven't moved. Okay. Go there. It's ground truth for models that are trained on hyperpod data and modern airborne geophysics and satellite imagery.

19:40But you need the ground truth. And the ground truth for these maps was collected laboriously through hundreds of thousands of person hours of work of skilled geologists. And you could never go collect this data again. It's too hard. So we're digitizing all of these maps. And in the case of the Zambian Geologic Survey, you can access it publicly. You can go to gsd.gov.zm. This is our partnership with the government of Zambia. And so we're not doing it for ourselves. I mean, we benefit, obviously, because we want to use the data as we want to use every data source in the world. But really, the benefit of the data is for the people of Zambia to stimulate investment.

20:17Why can't other exploration companies take a similar approach to what Cobalt is doing? What's so hard about this? It's a great question. The amount of capital that you have to deploy or that the industry is deploying on average to make a discovery has grown more than tenfold. It used to be that on average in base metals exploration, about$80 million worth of exploration projects, mostly failures, would produce a winner, a project that eventually is going to become a mine. And today it's well over a billion dollars. A billion dollars of failures to produce a winner. Wow. And that's for two reasons.

20:53One is that the problem's getting harder, right? The easy to find things have been found, the ones, the raisins at the surface, right? But the second reason is because the pace of innovation has been too slow. So you talk about mineral exploration as an information problem. And in order to tackle that, you have a chief philosopher on the team. What does a chief philosopher do and why that role? We do have a chief philosopher. And it's funny, I had a conversation with a journalist the other day where he had reached out because he saw that we're a mineral company, the mining company with the chief philosopher.

21:30and he was a little bit disappointed to learn that our chief philosopher is an epistemologist and not an ethicist. Cobalt ethics are also incredibly important for our business. It's right there on the homepage of the company. We start every all-hands meeting with an ethics share, with a vignette about one of the company's ethics. But we need the help on the epistemology because we're trying to understand— So define epistemology for someone who has not heard that term before. Yeah, so this is all about—it's all about thinking about knowledge. It's how do we know what we know? How can we reason about these things that we can't see?

22:05How can we make inferences about these things that we can't directly observe? And so we've written an epistemology of exploration. Really, as we're thinking about it, the core technologies are epistemic in nature. They help us know things better. And so the idea of Kobold's epistemology of exploration is, number one, we have to make really good hypotheses. What we're doing when we're going out in the world is we're deploying some capital to get some information. That information is useful if we're testing an idea, and it needs to be a well-defined idea. We need to make some predictions about what we're going to observe.

22:45We need to reconcile the results of our measurements with what our prior understanding was and use that to guide at the next decision. If you're an explorer at Kobold and you're writing a technical memo to explain your idea and then you're writing an authorization for expenditure and you're saying, you know, Josh and Kurt, I want two and a half million dollars. You've got to get that on record before you get an allocation of resources to go pursue an idea. So that's first, is that you need good hypotheses, and there's what characterizes a good hypothesis. And then the second is that you've got to have alternative hypotheses.

23:18Now, you can't just have one view of what the world is, and then the alternative is, well, not that. There are many possibilities. So we enlisted the help of our chief philosopher, who's written this wonderful book called The Knowledge Machine, about the nature of what is science. What is science? Why is it powerful? How is it different from other forms of inquiry? to help us think through these kinds of problems and communicate them clearly. Very cool. Okay, let's talk about your ethics framework. What does that look like? What are some of the ethical principles underlying the company? And give us some examples of how you open up your kind of team meetings with an ethics share.

23:56That's super cool. Yeah, great question. So this is important because we want to build a great company. And part of being a great company is leading in science. Part of being a great company is doing something good for the world and obviously providing the raw materials that we need for the future economy. That's really valuable. But mining has local impacts and it's not sufficient to say there are global benefits. Therefore, we accept all of these local impacts. Actually, we need to work to minimize those things and to do benefit to the communities where we operate. The ethics, cobalt's ethics include protecting the environment.

24:35and protecting the safety of our workers and of everyone else around us, of anti-discrimination in all of our practices, and of anti-corruption, for example. These are all very important values. Some of the geographies where you operate, that's real. Yeah, of course. And so we're committed to working at the highest international standards wherever we operate, and we can do that in the places that we're operating. and we've built some extraordinary operations. In Zambia, we're led by our CEO of Cobalt Africa, Mfikei Makai. She's an amazing engineer and leader. She was on the Time AI 100 list last year, which is super awesome.