0:05

Hi, listeners, and welcome back to

0:07

KnowPriors. Today, we're speaking with Josh

0:09

Goldman, co -founder of Cobalt Metals. Cobalt

0:11

is building the world's largest collection of

0:13

geoscience data and using their AI

0:15

tools to better identify mineral deposits like

0:17

lithium and copper to be a

0:19

better explorer. Cobalt invests over

0:21

$100 million annually across 70

0:24

projects on four continents today. Josh,

0:26

welcome to No Priors. It's

0:28

a pleasure. Thanks so much for having me. This is

0:30

a super interesting real -world business. You

0:33

run an intelligent mining company. What

0:35

does that mean? What does cobalt

0:37

do? We explore for

0:39

minerals. We're looking for lithium

0:41

and copper and the other metals

0:43

that we need. to build other businesses

0:45

that are powered by batteries and

0:47

AI. And we develop AI technologies and

0:49

we combine AI with human intelligence

0:52

to be better explorers, more successful at

0:54

finding the sources of minerals that

0:56

we need for these businesses. Are you

0:58

both finding them as well as

1:00

actually going to do the mining? Or

1:02

is it only a tool to

1:04

find these sorts of assets or resources?

1:07

That's a central question. So our

1:09

business is focused on exploration and it's

1:11

focused on exploration for a couple

1:13

of reasons. One is because there's way

1:15

more value to be created there.

1:17

And the second is that's where technology

1:19

can be really differentiating. The economics

1:22

of exploration are really

1:24

quite extraordinary. With a few

1:26

million dollars of capital, you can

1:28

create 100 to 1 ,000 times return. Exploration

1:31

is a very old business. Think about

1:33

gold miners back in the middle of

1:35

the 19th century. If you can get

1:37

the right claims, you can strike it

1:39

rich if you can dig in the

1:41

right places. It's about where you look

1:43

and how effectively you can look. And

1:45

so the unit economics of discovery are

1:47

really extraordinary. The problem with exploration as

1:49

a business is that the success rate's

1:51

really low. You have to try many,

1:53

many different places before you can

1:55

find something and the problem keeps getting

1:57

harder. But that's also the reason

1:59

why technology is so differentiating. We're

2:01

looking for things that are harder and

2:04

harder to find. It used to be

2:06

that you could find minerals literally with

2:08

your eyeballs by walking across the ground

2:10

and prospecting. And a lot of the

2:12

copper ore minerals that form at the

2:14

surface are modified by the air and

2:16

the water and the surface environment turn

2:18

blue and green like the patina on

2:20

the Statue of Liberty. Anything

2:22

you can find by traipsing across the

2:24

ground with your eyes has been found

2:26

by now. And we need more intelligent

2:28

ways of looking for minerals in places

2:30

that are concealed. They're literally underground and

2:33

concealed by the rocks. And so technology

2:35

is a way to create differentiation to

2:37

be a much better explorer. And once

2:39

we find things, there's a continuum from

2:41

you had a good idea and you

2:43

collected some rock samples, you found something

2:45

underground, you have many different holes,

2:47

and you've established that you've got something

2:49

continuous. to, oh, it's going

2:51

to be economic to mineness, to we're

2:54

designing the mine, to we're building the

2:56

mine. There's a whole spectrum. And the

2:58

technology that we use to find resources

3:00

and define those resources helps set a

3:02

project up to be a more economical

3:04

mine as well. So we continue to

3:06

contribute technology and stay involved in projects

3:08

as they evolve. What sort of data

3:11

are you using in order to actually

3:13

identify a mine site or a potential

3:15

site? Okay, so there's a huge amount of

3:17

data. Humans have been collecting data about the

3:19

Earth for as long as humans have been

3:21

looking at rocks, right? And there's an enormous

3:23

amount of data, a great deal of which

3:25

is actually in the public domain. And the

3:28

length scales are very different. Start with the

3:30

global length scale. What can you know about

3:32

the entire Earth? Well, you can look at

3:34

satellite imagery. And you can look at satellite

3:36

imagery in different colors. And so you can

3:38

get a sense of the rocks that

3:40

are exposed at the surface. And there are

3:42

data sets that tell you about the structure

3:45

of the continents and the ancient continents that

3:47

collided and where the sort of ancient

3:49

continental proto -continents were and where those crashed

3:51

into each other a long time ago

3:53

and formed mountain ranges. You

3:55

zoom in and you go to another

3:57

length scale and you can fly

3:59

airborne surveys with sensors on them that

4:01

can detect the magnetic properties and the

4:03

density and the electrical conductivity of the

4:06

rocks. Go out and collect rock samples

4:08

and measure what they're made out of,

4:10

all the concentrations of different chemical elements

4:12

and likewise for soil samples. And these

4:14

are standard types of data that are

4:16

used in the industry. And there's a

4:18

huge number of these old data sets.

4:20

that are in the public domain. Most

4:24

private companies have to disclose their

4:26

data to regulators any place you

4:28

look. Typically, a

4:30

number of other companies have looked there

4:32

before and haven't yet found anything. But this

4:34

data is, even when it's in structured

4:36

form, it is spread out over tens of

4:38

thousands of different repositories. There's nowhere you can

4:40

go where this is all aggregated in one

4:42

place. You both have to do a

4:44

lot of really hard technical work to get

4:47

it together. And you have to

4:49

do a lot of scientific work to use

4:51

judgment about what this data actually means

4:53

and whether or not it's fit for purpose.

4:55

There's all kinds of messy problems with

4:57

the data. But a lot

4:59

of this data is unstructured as

5:01

well. And geologists use a lot

5:03

of words. There's a very rich

5:05

lexicon of geological vocabulary for rocks

5:07

and time periods. There's a lot

5:10

of text data and reports that

5:12

are filed by companies, often with

5:14

regulators that become public after a

5:16

period of time. And there's an

5:18

enormous amount of data in maps

5:20

of various kinds. One of my

5:22

favorite data sets that we use

5:24

around the world, a set of

5:26

maps from Zambia from almost 100

5:28

years ago, the originals are hand -painted

5:31

on linen. And we got

5:33

a tip from an elderly geologist

5:35

on which drawer in the state archives

5:37

to look in that had this

5:39

particular collection of maps. And you could

5:41

never collect data like this again.

5:43

It's incredibly labor intensive. And now there's

5:45

lots of farms and people living

5:47

there. You can't go traipsing across their

5:49

ground looking at the rocks. But

5:52

these observations were made by skilled geologists.

5:54

and the rocks haven't moved. So

5:56

there's no expiration date on the data.

5:58

And so you can take data

6:00

sets like this that provide ground truth

6:02

and use it for training machine

6:04

learning models based on modern airborne geophysical

6:06

surveys and modern satellite imagery. And

6:08

it's the combination of all these many

6:10

different data sets of different types

6:12

of data and the systematic use of

6:15

structured and unstructured data that's really

6:17

powerful. In a pre -cobald world, or

6:19

just with, like, I mean, maybe you

6:21

can just tell us, like, who

6:23

the largest couple other explorers are out

6:25

there. Like, how do you go

6:27

look for lithium? Oh, yeah.

6:29

Okay. So, again, you've got this

6:31

different set of length scales, right?

6:33

You start with the Earth. And

6:35

you say, okay, I'm interested in

6:38

lithium. What's the recipe

6:40

for making a lithium deposit? What

6:42

is an ore deposit in the first

6:44

place? So there's an enormous amount of

6:46

lithium in the Earth's crust. The central

6:48

problem is that the lithium that's in

6:50

your driveway is in very low concentration.

6:53

The lithium that's in the granites that

6:56

you can see out your window

6:58

is not economical to extract. It's too

7:00

dilute. A lot of the minerals

7:02

that we're looking for, the metals we're

7:04

looking for, their concentration in the

7:06

crust is a few tens of parts

7:08

per million. The crust is really

7:10

big, so there's a lot of metals.

7:12

So what we're looking for are

7:15

those places in the Earth's crust where

7:17

natural processes, geologic processes in Earth's

7:19

history, have gathered up a bunch of

7:21

metals from a really large volume

7:23

of rock. And then they've moved them

7:25

and they have concentrated them and

7:27

then redeposited them in a much more

7:29

concentrated form. More like 1 %

7:31

copper or 1 % lithium or even more

7:34

than that. And then you can take it

7:36

the rest of the way to 100 % with

7:38

industry. So that's what an ore

7:40

deposit is. And not only

7:42

is there lots of lithium

7:44

and lots of copper in

7:46

the crust, but there are

7:48

actually many, many places where

7:50

those geological processes have happened,

7:52

even though they're rare in

7:54

the Earth as a whole.

7:56

And so the problem is,

7:58

where are those special places

8:00

where these natural processes happened?

8:02

And how can we find

8:04

those? And we talk about...

8:07

is an information problem because the

8:09

scarce resource is not lithium

8:11

or copper metal in the ground.

8:13

It's actually information. The

8:15

scarce resources are not the ore deposits. The

8:18

scarce resource is the information about where

8:20

the ore deposits are located. So you have

8:22

to first understand, well, how is an

8:24

order deposit formed? You have to know the

8:26

recipe and you have to have some

8:28

ideas about where those processes might have been

8:30

occurring on the earth and how they're

8:32

going to be expressed in the data sets.

8:34

Then you can marshal the data and

8:36

you can start asking questions of the data.

8:38

You can make hypotheses and then you

8:41

can narrow down on some specific portion of

8:43

the earth. And then you actually, what

8:45

you want to do is you want to

8:47

go acquire the land. I guess another

8:49

overlay may be sort of... geography relative to

8:51

governance of the country, regulatory ability to

8:53

actually mine things like my census, for example,

8:55

the U .S. has a pretty diverse range

8:57

of deposits. We just kind of don't

8:59

want to mine certain locations anymore or certain

9:02

types of we don't want to do

9:04

certain types of mining. And so it's a

9:06

bit more of a regulatory issue in

9:08

some cases versus can we find stuff? Is

9:10

that a correct understanding or is it

9:12

these things are rare enough and scarce enough

9:14

that you really have to scour the

9:16

ends of the earth to find them? Regulatory

9:19

constraints are really important, but

9:21

at the same time, you

9:23

can't be too narrow in

9:25

your initial filter because they

9:27

are rare enough. You

9:29

want to put yourself in the

9:31

place where you have the highest

9:33

probability of success. You want to

9:35

start with the best prior that

9:37

you can. That

9:40

way, your likelihood of success is going

9:42

to be much higher. It isn't just

9:44

a function of regulations. We

9:46

consider security of property rights. If we find

9:48

something, we have to be able to develop

9:50

it into a mine that is going to

9:52

produce for decades, or we have to be

9:54

able to sell it to someone who

9:56

would do that. And so you have to

9:59

be able to rely the fact

10:01

that you can continue to own the property for

10:03

that period and that you will... The tax

10:05

rates and the royalty rates will be consistent over

10:07

that period of time. Development

10:09

is challenging because you don't just

10:11

have regulators. You have lots of different

10:13

community interests. And these things are

10:15

extremely local. The U .S. is not

10:17

monolithic. You have state regulators. And within

10:19

a state, you have many different

10:21

communities, many different indigenous groups. And this

10:24

is true the world over. It's

10:26

true in Zambia. There are 50 different

10:28

chiefdoms. And so you have traditional

10:30

leaders everywhere that you work. technical success

10:32

is not very helpful. Success is

10:34

you find something that is really economic

10:36

to develop that either we can

10:38

develop or we can sell to somebody

10:40

who can develop it. And so

10:42

if we don't actually have... the so

10:44

-called social license to operate. If we

10:46

haven't invested in the relationships with

10:48

the community to be able to build

10:50

and we haven't started in a

10:53

place where that's possible, then we're not

10:55

going to be successful. But these

10:57

are hyper -local problems for sure. Josh,

10:59

can you give us a sense of

11:01

just like the scale of the

11:03

operation for Kobold today and like, you

11:05

know, where you are looking, where

11:07

you own land, where you're drilling, what

11:09

you've discovered? Absolutely. So we operate

11:11

exploration projects. Basically, the company

11:13

does two things. We

11:15

find places that are prospective for

11:17

making discoveries, and then we

11:19

go test our hypotheses by going

11:22

and collecting data, collecting rock

11:24

samples, flying airborne surveys, drilling holes

11:26

to get samples of rock

11:28

from below the ground. And we

11:30

develop technology that we use

11:32

for guiding our decision making. So

11:34

our exploration portfolio is more

11:36

than 60 projects, and they're

11:38

on four continents. They're in

11:40

North America, Europe, Australia, and

11:42

critically in Africa, targeting copper

11:44

and lithium and nickel and

11:46

cobalt and likely other commodities

11:48

to come. And again,

11:50

in all of these cases, we

11:52

own the exploration rights, either

11:54

ourselves or in combination with a

11:56

joint venture partner. And we

11:59

are operating the exploration programs. Almost

12:01

all of these are pre -discovery

12:03

opportunities. They're seeds we've planted.

12:05

Any of them could become great

12:07

ore deposits. And what we

12:09

have in Zambia is really an

12:11

extraordinary deposit. It is the

12:13

highest grade large copper deposit that

12:15

is not yet a mine.

12:17

The average of operating copper mines

12:19

today is that the concentration

12:21

of copper in the ore is

12:23

about 0 .6%. So if you

12:25

mine 1 ,000 kilograms of ore,

12:27

not including the non -ore rocks

12:29

all around it, there's six

12:31

kilograms of copper in it that

12:33

you can potentially extract. And

12:35

the Mingomba deposit in Zambia, the

12:37

core of it is over

12:40

5 % copper, and it's very

12:42

large. And that's extraordinary. That means

12:44

the economics are much better

12:46

because if you compare a high

12:48

-grade and a low -grade deposit,

12:50

a 5 % and a 0 .5

12:52

% deposit, if they're producing the

12:54

same amount of copper, they

12:56

have the same revenue. But

12:58

the high -grade deposit, if you have 10

13:00

times the grade, it means you are producing

13:02

10 times less rock, at least. You

13:04

have 10 times less stuff to haul out

13:07

of the ground, 10 times less waste, 10

13:09

times smaller plant. So that means the

13:11

economics are far better, the capital intensity

13:13

is lower, the operating costs are lower,

13:15

and it means the environmental footprint is

13:17

smaller. So those are the things that

13:20

we are looking for. In a commodity

13:22

business, everybody sells copper for the same

13:24

price. It's a global commodity market. And

13:26

our ability to make money depends on

13:28

what our margin is. That means we

13:30

need to be a low cost producer

13:32

and we want low capital intensity assets.

13:34

And so that is the definition of

13:36

the exploration problem is find the highest

13:38

quality assets. And in Zambia so far,

13:40

we have a quite extraordinary and really

13:42

world class copper deposit. Can you tell

13:44

us a little bit more about the

13:46

technology that you're using? Obviously, you mentioned

13:48

you're mixing sort of older school data,

13:51

modern. image

13:53

-based data, et cetera. And then you

13:55

have to kind of data mine

13:57

it or extrapolate where these potential deposits

13:59

are. What sort of models are

14:01

you using? What approaches are you using?

14:03

How do you think about overall

14:06

what you're building from a sort of

14:08

AI and data perspective? For sure.

14:10

So Cobalt's technology is a full stack

14:12

system for guiding exploration decision -making. So

14:14

there are dozens of different products

14:17

that work together and they fit on

14:19

three themes. The first one is

14:21

sensors. hardware that we have

14:23

developed that collects new kinds of

14:25

data about the Earth. The second is

14:27

the data system for taking all

14:29

of the data that we're collecting, all

14:31

the historic data. in structured data

14:33

from many different kinds and a huge

14:35

corpus of unstructured data and getting

14:37

this all in one system so that

14:39

we can interact with it systematically. And

14:42

rather than hunting and pecking through this, we can

14:44

interact with the whole corpus of data at the same

14:47

time. LOMs and other

14:49

technologies are very powerful for

14:51

being able to interact with

14:53

all of these different types

14:55

of information. And the third

14:57

theme are models, dozens of

14:59

different models for making better

15:01

predictions about where and how

15:03

to look. And so these

15:05

models, again, they operate at

15:07

many different length scales. So

15:09

there's models trained on satellite

15:11

imagery or our proprietary hyperspectral

15:13

airborne imagery. And you've got

15:15

some rock samples on the ground. And

15:17

so we can predict from imagery what

15:20

types of rocks we're going to find

15:22

at the surface and what the properties

15:24

of those rocks are going to be.

15:26

And then what's really exciting is that

15:28

it's not just that we have a

15:30

model. or a model for lithium pegmatites,

15:32

or a model for mafic to ultra -mafic

15:34

rocks that might host nickel deposits. It's

15:36

that we make a prediction and develop

15:38

an initial set of hypotheses on that.

15:40

And then when our team gets on

15:43

the ground, every day that they're in

15:45

the field, they are collecting new training

15:47

data. And they're not just going to

15:49

places where we have high confidence in

15:51

what the rocks are, because we're not

15:53

going to learn anything. We're going to

15:55

places where the models are highly uncertain.

15:58

And the new training data, a

16:00

small amount of additional ground truth,

16:02

can dramatically improve the predictive power

16:04

of our models. And so what

16:06

happens is you have geoscientists in

16:08

the field making observations. And using

16:10

those observations, we are retraining those

16:12

models every day and serving new

16:14

predictions out to the team. You've

16:16

got this duet of data scientists,

16:18

of technologists and geologists working together

16:21

on the same problem. Of

16:23

like a hypothesis and then like validation

16:25

or invalidation? Is it like I'm imagining

16:27

like, OK, at this set of spots

16:29

in Zambia, I am going to go

16:31

20 feet below the surface or whatever

16:33

it is. And going to find this

16:35

concentration of something. Absolutely. Yeah. So a

16:37

whole bunch. Let me give you a

16:39

whole bunch of examples. Right. So one

16:41

example is I'm going to go. this

16:43

location and I'm going to pegmatites are

16:45

the container rocks for lithium deposits. And

16:47

we predict that there's, it's just a

16:49

name for a rock, like a granite

16:52

or something like that. Okay. We're going

16:54

to predict that we have these special

16:56

rocks, pegmatites that might contain lithium. We're

16:58

going to predict that there's one in

17:00

this location and we're going to go

17:02

then land on it and we're to

17:04

sample those rocks and we're going to

17:06

look at it and see. Okay. That's

17:08

a prediction we're making at the surface.

17:10

And we're making predictions in 3D and

17:12

we're saying, okay, here now. I think

17:14

there is a layer of conductive rocks

17:16

here, and I think those conductive rocks

17:19

are prospective for hosting nickel and copper

17:21

and cobalt. And I think this rock

17:23

layer is, we're going to intersect this

17:25

rock layer between 200 and 300 meters

17:27

below surface, and it's going to be

17:29

highly conductive. It's going to have some

17:31

distribution for how much sulfur and how

17:33

much nickel and copper in it. And

17:36

more than that, I'm to say the

17:38

best place to test this set of

17:40

hypotheses is by putting a hole at

17:42

this location and by drilling it in

17:44

this direction. And then other times

17:47

there's a known layer of rock and

17:49

we're saying, okay, we think this layer continues

17:51

out in this direction. And here is

17:53

a surface where we're predicting this layer is

17:55

going to be at this depth and

17:57

it's going to be this thick and it's

17:59

going to have this much copper in

18:01

it. And you're going to get a probability

18:03

distribution for all of these at any

18:05

given point. Those are the kinds of predictions

18:07

that we're making. And then we go

18:10

collect a piece of information and then condition

18:12

the model on the new data and

18:14

serve out a new prediction. And

18:16

on the third theme

18:18

of sensors, we use

18:20

everything that's available today

18:22

that we can get

18:24

from a service provider.

18:26

But most mining companies

18:28

are not as... not

18:30

as keen to use new data types

18:32

and as keen to invest in new

18:34

kinds of technologies. And sometimes

18:37

we need to go build our

18:39

own. And so an example of this

18:41

is our hyperspectral imaging technology. There

18:43

were new imaging chips available that were

18:45

not yet deployed in the service

18:47

market. The mining industry was adopting them

18:49

too slowly. We built our own

18:51

hyperspectral imaging system in less than a

18:53

year. We had it flying on

18:56

a light aircraft and we're surveying areas

18:58

that we're interested in and using

19:00

it for getting data in 600 colors

19:02

at dramatically lower cost of acquisition

19:04

and much, much faster time to deliver

19:06

processed images. So that we're using,

19:08

we're integrating that information with other types

19:10

of data and using that to

19:13

make decisions where to go in the

19:15

first place and then how to

19:17

change our exploration plans while we're in

19:19

the field. Was there any

19:21

tool or data set that was most

19:23

crucial for that murky discovery you

19:25

made in Zambia? Was there a piece

19:27

of data that others had overlooked?

19:29

Was it just looking at that geography?

19:31

Was it a specific tool? There

19:33

is no one piece of data that

19:35

enabled that. And that's really a

19:37

critical theme. Often new technologies are invented

19:40

in this industry where people think,

19:42

ah, this is going to be the

19:44

silver bullet. It's going to help

19:46

us find. It's going to help us

19:48

find all the ore deposits or this data set

19:50

alone is going to let us do that. Actually,

19:52

the data is very high dimensional. And when you

19:54

can add dimensionality to the data, then you can

19:56

have improved predictive power. And so

19:58

that's the story there as it is

20:00

everywhere else. It's a combination of new

20:02

analytical methods. the ability to quantify uncertainty

20:04

and understand the range of possibilities, and

20:06

critical scientific insights about the way that

20:09

these OR systems are formed. And all

20:11

of those things in combination are what

20:13

make it possible. There is no way

20:15

to isolate the AI from the HI.

20:17

There's no way to isolate one piece

20:19

of data that's uniquely powerful. And that's

20:21

one of the reasons that I think

20:23

is limited innovation as well, is that

20:25

we think, oh... This new airborne gravity

20:27

gradiometry invented in the 1990s was going

20:29

to find all the ore deposits. It

20:31

doesn't, but it's really powerful. We're really

20:33

happy when we can get that data.

20:35

We go collect it ourselves. But these

20:37

are incremental improvements to predictive power, but

20:39

it's only possible if you can work

20:41

with all of these different data sets

20:44

together in a unified way. How does

20:46

a project like this get valued? Like

20:48

if you sell it to somebody else

20:50

or you develop it, it sounds like,

20:52

well, copper is whatever price it is.

20:54

And, you know, take some risk on

20:56

that over time. And then there's, you

20:58

know, cost of operation based on basically

21:00

how concentrated the deposit is and then

21:02

like how large it is. And then

21:04

those those kind of give you some

21:06

sort of cash flow model for the

21:08

business. That's exactly right. Yeah, this is

21:10

it's actually really easy to value a

21:12

natural resource asset like this. They all

21:14

trade on their present value of future

21:17

production. which is very knowable. It is

21:19

much easier to know what a mine

21:21

is going to produce 20 years from

21:23

now than it is to know what

21:25

a SaaS company's sales volume is going

21:27

to be 20 years from now and

21:29

how it's going to be priced, right?

21:31

I feel attacked. They're

21:34

very different kinds of businesses. You think you

21:36

build a mine that can move, say, 10 million

21:38

tons of ore per year, and then what

21:40

you're going to do is you're going to dig

21:42

10 million tons of ore per year, and

21:44

you're going to dig the highest grade part first

21:46

and the next highest grade, and on average,

21:48

it's going to produce. whatever percent copper it's going

21:50

to produce. And so it's very

21:52

simple, right? The cash flow is, the revenue

21:54

is what the commodity price is. The volume

21:56

is based on the size of the mine

21:58

you and how you cost it. The cost

22:00

is very knowable because you need to know,

22:02

well, how many trucks do you need to

22:05

move? And how much water do you need

22:07

to pump? And what does it cost? pump

22:09

the water. And that's all straightforward stuff. And

22:11

then you need the capital costs, which is

22:13

like, okay, I'm going to build a plant.

22:15

And these things are, you know, they're big

22:17

vessels. Like you got a tank and you

22:19

got a crusher and the crusher has got

22:21

some steel balls or the mill has some

22:23

steel balls in it. And these are knowable

22:25

things. They're typically built. So you can figure

22:27

out what the margin is going to be.

22:30

You can see what the capital profile is

22:32

going to be. You have to assess what

22:34

fraction of the copper can you recover. then

22:36

those are your sensitivities. Like, ah, I think

22:38

we can get 90 % of the copper.

22:40

If we can get 92, the economics are

22:42

juicier. If we only get 88, it's a

22:45

little dilutive. But those are the uncertainties. And

22:47

then you discount that according to, well, what

22:49

is the risk profile of the asset? What

22:51

stage is it? How close are you to

22:53

production? And

22:55

you might demand a higher rate

22:57

of return if you are in

22:59

a less stable jurisdiction. And so

23:01

it's quite straightforward. And actually, we

23:03

know with high confidence what the

23:06

sales volume will be from Mingomba

23:08

20 years from today. And

23:10

that's amazing. There's potential upside if

23:12

we find more and more resource. And

23:14

that's one of the things about

23:16

these deposits is once you get underground

23:18

and you start mining... then you

23:21

learn more and more about the geology

23:23

and you keep finding extensions. And

23:25

so mines are often designed, you underwrite

23:27

an investment based on the first

23:29

20 years, and then actually many of

23:31

these mines operate for decades, many

23:33

decades longer, 50, 60, 70 years, because

23:36

the resource keeps going and you

23:38

can keep adding to it as you

23:40

go. So it's actually pretty straightforward

23:42

to understand how these are valued. And

23:44

these are hard assets. There's a

23:46

property interest. The market values

23:48

these accordingly. They all trade on their

23:50

present value of future production. How

23:53

successful are exploration companies in general

23:55

today? Like if I look at, if

23:57

I start, I don't know how

23:59

to ask this question, but if I

24:01

start sampling 100 sites, I have

24:03

100 theses. Like, do I find one?

24:06

Do I find zero? Do I

24:08

find 10? And like, how much better

24:10

do you think cobalt can be?

24:12

Yeah. This is the key question, right?

24:14

Which is like, what is the

24:16

success rate in the industry and how

24:18

much better we hope we can

24:20

do? So in the industry, it's gotten

24:23

10x worse in the last 30

24:25

years because the problem has gotten harder

24:27

and the industry is slow to

24:29

innovate. The way to think about it is

24:31

not the number of successes. You can go look. studies

24:35

that will say, you know, a half a

24:37

percent success rate or something like that. But what

24:39

actually counts as an attempt is ambiguous. And

24:41

the way that we think about it

24:44

is that, you know, the key resource

24:46

input is you have to invest some

24:48

capital to run an exploration program. You

24:50

have to put a geologist on a

24:52

helicopter and go out and take samples.

24:54

You have to drill holes. If you

24:56

take a portfolio of exploration projects that

24:58

cost some money, say a billion dollars

25:01

industry -wide, how many successes will you

25:03

have? then

25:06

industry -wide, a billion successes will

25:08

have hundreds of failures, but

25:10

eight discoveries as of 30 years

25:12

ago. And today, less than

25:14

one. Less than one high -quality

25:16

economic deposit. And so that's why

25:18

exploration in the aggregate is

25:20

not a great business. So

25:22

Kobold, we target $50 to

25:24

$100 million per discovery. That's the

25:26

goal. That's how well we

25:28

want to do. And so

25:30

far, we now have an extraordinary copper

25:32

deposit and we have succeeded. Now we

25:34

need to do it again and again

25:36

and again. One thing I've heard on

25:38

the capital side, which may or may

25:41

not be true, so be great to

25:43

get your sense of this, is that

25:45

a lot of the people who used

25:47

to buy out and run some of

25:49

these assets in terms of mining assets

25:51

or things like that, at least in

25:53

the Western world, have... run

25:55

into more and more capital constraints because

25:57

the funders have sort of dried

25:59

up in part due to ESG or

26:01

other programs. Has that at all

26:03

been a case or something that's impacted

26:05

your perception of the sorts of

26:07

players that are in this business these

26:09

days? Or do you think that

26:11

really doesn't matter and there's plenty of

26:13

capital availability and it's just hard

26:15

to find these deposits? Yeah, I think

26:18

that the real scarcity is good

26:20

quality or deposits. Great projects

26:22

don't have problems getting funded. Whoever

26:25

owns them. Great projects have lots

26:27

of suitors and people who want

26:29

to buy them. The problem

26:31

is there just aren't very many great

26:33

projects. And so that's what we need to

26:35

do. We need to go find more

26:37

really high quality deposits. Are

26:39

there parts of the world that you

26:41

feel are dramatically underexplored relative to that? It

26:43

varies a lot by commodity. And

26:46

copper... has been

26:48

an exploration target for a long

26:50

time. And so people have been

26:52

looking for copper in South America

26:54

and Central Africa, yet there are

26:56

still parts of these places that

26:59

are quite underexplored. We're very active

27:01

in Zambia, where of course Mingomba

27:03

is, along with a number of

27:05

other exploration projects. The

27:08

parts of Zambia where Mingomba

27:10

lies is... deeper underground where

27:12

you don't have surface expression,

27:14

the deeper parts of the

27:16

basins in Zambia that host

27:18

copper deposits are quite underexplored.

27:22

You have a jurisdiction like Congo

27:24

that has had a number of

27:26

challenges. The exploration potential remains great

27:29

across many commodities. There has been

27:31

a lot of activity, but there

27:33

could be dramatically more activity. And

27:35

lithium, much of the world is

27:38

underexplored for lithium. Lithium hasn't

27:40

been a... exploration target

27:42

until very recently, until the

27:44

growth of lithium -ion batteries

27:46

for big devices like EVs

27:48

and drones and whatnot, not

27:51

just personal devices. The big

27:53

lithium deposits in production today,

27:55

at least the hard rock

27:57

lithium deposits, were found by

27:59

people looking for tantalum for

28:01

capacitors, for the electronics industry

28:04

in the 1980s. And so

28:06

the science of how lithium

28:08

ore deposits form is incipient.

28:10

That's really exciting because a

28:12

little bit of increased scientific

28:14

understanding can be a really

28:16

potent differentiator. So potential for

28:19

big breakthroughs. Are there any

28:21

commodities that you think are

28:23

overstated in terms of their

28:25

scarcity? So an example that I've

28:27

heard is like rare earth minerals may not be

28:29

as rare as people say. And there's deposits, you

28:31

know, more broadly than just in China where it's

28:33

often spoken about or like what are what are

28:35

the things that you think are actually not that

28:37

scarce that people talk about as scarce? That's the

28:39

top of the list. Rare earth. A lot of

28:41

the noise about rare earth is because it has

28:44

the word rare in its name. Yeah. Good

28:46

branding. Not that rare. Also, lithium

28:49

and copper and nickel and cobalt

28:51

are not rare earth elements. The

28:53

rare earth elements are it's a

28:55

well -defined term that not. just

28:57

things that are rare, but neodymium

28:59

and dysprosium, which are important for

29:01

permanent magnets, which are important for

29:04

electric motors and so on. They

29:06

are important. The reason that rare

29:08

earths get noise, besides the name

29:10

rare, is that a concentration of

29:12

downstream processing capacity in China. Spurred

29:18

by Chinese incentives, there's been a lot of

29:20

processing. You have to extract the minerals from the

29:22

ground and then you have to refine them

29:24

into a metal that you can put into a

29:26

product. And there's been a

29:29

huge build out of that, not

29:31

just for rare earth processing, but

29:33

also for lithium and now copper

29:35

smelters as well. And that does

29:37

a couple of things. One. It

29:40

means it's really hard for somebody

29:42

else to go build a processing facility

29:44

because you are competing for feedstock.

29:46

You want to take copper concentrate from

29:48

somewhere and you want to smelt

29:50

it into copper metal. We have to

29:53

go buy your copper concentrate. If

29:55

a Chinese party is willing to buy

29:57

it for more than you because

29:59

they will accept less margin, that makes

30:01

it much harder. It's much harder

30:03

to underwrite a project like that. It's

30:08

had a deterring effect on just

30:10

private commercial actors willing to put

30:12

the capital to work to invest

30:14

in processing capacity. It makes it

30:16

hard for another private actor to

30:18

do the same without guarantees or

30:20

subsidies or something, which we don't

30:22

have any of that as a

30:24

business. That's a strength of Kobold

30:26

is that we just have great

30:28

assets rather than a subsidy. And

30:30

the second is that... Because

30:33

there's so much downstream processing capacity

30:35

in China, then you have the

30:37

raw materials going to China, and

30:39

then you have a concentration of

30:41

the downstream supply chain from there.

30:43

Then you make products from that,

30:45

and so it's a big strength.

30:47

for Chinese manufacturing capacity is you

30:49

have all of these materials landed

30:51

there already. And, you

30:54

know, if you think about that on

30:56

an integrated economic basis, it can

30:58

be very, very powerful. And so that's

31:00

one of the reasons that rare

31:02

earths are in the news a lot

31:04

too. Is there anything that's the

31:06

other way around where you actually worry

31:08

about some commodity or material not

31:11

being able to meet demand for something

31:13

that's industrially important for us? The

31:15

ones that I listed for us are

31:17

the ones where we think both

31:19

there's a lot out there to find.

31:21

The demand tailwinds are really strong. There's

31:24

going to be some depth to those

31:26

commodity markets, so you don't have to have

31:28

a really well -dialed view on commodity prices,

31:30

which we don't. Again, our goal is

31:32

we want to be the low -cost producers. And

31:35

so surprises in that market, not

31:37

great. We are looking

31:39

at other commodities that could be those

31:41

unusual ones, but there isn't one

31:43

today that stands out that we're tackling.

31:46

So it's not that important that we buy

31:48

Greenland. Not

31:52

going to go there. My

31:54

joke version of this is to take over

31:56

Baja because it's already called California. It's

31:59

nice and beachy and sandy. That

32:01

seems like a really great place to annex if you

32:03

were to annex somewhere. I'm

32:07

happy to go to these places regardless of

32:09

which flag. Fair enough. Yeah, me too. It

32:11

actually sounds nice. Only if the algorithms tell

32:13

you, the algorithms and the initial rock samples

32:15

tell you, that's going to be efficient to

32:17

get the lithium out, I suppose. Yeah,

32:19

we need more lithium out of Baja. So let

32:21

me know. A

32:23

lot will go overseas on the

32:26

beach. Josh, when we last saw

32:28

each other, we had like a

32:30

really interesting discussion about how important

32:32

you felt like philosophy was to

32:34

the business and the investments you'd

32:36

made about just how the company

32:38

operates. Can you talk about this

32:40

a little bit? Yeah, Kobold is

32:43

kind of an epistemic project, really.

32:45

Our business is about making better

32:47

predictions. That's what we're doing,

32:49

right? This is the thing we lack

32:51

is information about where the ore deposits

32:53

are. And the thing we do, like

32:55

the actual business activity, we make a

32:57

prediction, make a hypothesis. We go out

33:00

and we deploy capital and we spend

33:02

time testing our hypotheses. And so we

33:04

are successful as a business depending upon

33:06

how good our predictions are. It's what

33:08

the models are meant to do. We're

33:10

making predictions about what the rocks are

33:12

at the surface and what the rocks

33:15

are below the surface and what their

33:17

properties are, like their density, and how

33:19

much copper and nickel and other things

33:21

they contain. So how good are we

33:23

at doing that? Well, we have to

33:25

think hard about on what basis are

33:27

we making those predictions? What things do

33:30

we know about the world? And one

33:32

of the critical elements of this is

33:34

dealing with uncertainty. When you

33:36

have sparse data, then

33:38

you make a prediction about

33:40

everything in between your data

33:42

points. There are many possible

33:44

geologies that are consistent with the data.

33:46

When you make a prediction based on

33:48

data that you have from the surface

33:51

or from an aircraft and you're making

33:53

a prediction about what the properties of

33:55

the rocks are underground, there are many

33:57

possible geologies that are consistent with the

33:59

data. Standard practice in

34:01

the industry is to choose just one

34:03

and make your one best model.

34:05

Because, well, what else you going to

34:07

do? It's hard. You can't work

34:09

with 10 ,000 different models. It's very

34:11

difficult to keep multiple inconsistent hypotheses in

34:13

your mind at the same time.

34:15

But it's what we have to do.

34:17

That is how we become better,

34:19

is by embracing that uncertainty and recognizing

34:21

that our job is to judiciously

34:23

reduce that uncertainty. That's what we do

34:25

when we go out and collect

34:27

data. And the data is useful in

34:29

as much as it reduces uncertainty.

34:31

The way that we think about this

34:33

informs our practice for how we

34:35

actually explore. What is it that our

34:37

teams are doing every day? And

34:39

the scientific culture is one of the

34:41

critical aspects of the business. So

34:43

we have some unusual things. We have

34:46

a document in the company called

34:48

Kobold's Epistemology of Exploration. And

34:50

it really has a small number

34:52

of core ideas in it. Actually,

34:54

epistemology is important for reasons that

34:56

I talked about. that we

34:58

have to make really definite predictions, and

35:00

that means they have to be falsifiable. You

35:03

have to go on record before

35:05

you collect the data about what

35:07

could you observe that would cause

35:09

you to abandon this hypothesis. This

35:11

is how we avoid confirmation bias,

35:13

which is very susceptible to it

35:15

in this business. You come up

35:17

with an idea, and then you

35:19

collect some data, and you figure

35:21

out how to modify your hypothesis

35:23

to accommodate it. And then you

35:25

justify going out and spending more

35:27

time and more money. And really,

35:29

the third idea is that you

35:31

have to work with multiple alternative

35:33

hypotheses. Not just one hypothesis, but

35:35

what are the other possibilities? And

35:37

the point of data collection is

35:39

to distinguish between them. And at

35:41

least one of those hypotheses has

35:43

to be economically relevant. We are

35:45

a business, not a science project,

35:47

right? But the careful thinking about

35:49

what you're doing is really important.

35:51

So the epistemology of exploration. There's

35:53

a lot of vocabulary about this.

35:56

It feels like philosophical vocabulary, but

35:58

it's really important. Oddly,

36:00

we have a chief philosopher who

36:03

is an epistemologist. This is Michael

36:05

Trevins. He wrote a wonderful book

36:07

called The Knowledge Machine about what

36:09

is science and how it is

36:11

different from other ways of knowing.

36:14

And so this really guides

36:16

exploration practice and technology development.

36:18

So lots of the technologies

36:20

are designed to quantify uncertainty

36:22

and then given a set

36:25

of possibilities, determine what information

36:27

can we collect that will most

36:29

effectively reduce that uncertainty. For those

36:31

of us who don't have an

36:33

in -house philosopher, epistemology is the

36:35

study of, you know, what we

36:37

know, like what is knowledge and

36:39

how we know it is knowledge

36:41

and justifiable understanding, right? Maybe

36:44

one last thing on

36:46

this, like how do you,

36:48

you're a math and

36:50

physics guy originally, right? Yes.

36:54

And you went and did consulting and

36:56

you worked in oil and gas.

36:58

You worked in private equity around it.

37:00

And so that feels more relevant. But

37:03

this is like such a

37:05

cool discovery of an interesting problem

37:07

that you might go apply,

37:09

you know, decision making science and

37:11

data to. How

37:13

did you decide that you wanted

37:15

to go work on mining and

37:17

like better exploration? So I've

37:19

always been interested in the

37:22

intersection of. of energy and technology.

37:24

I studied physics because just

37:26

like grappling with hard questions. So

37:28

I did a PhD in

37:30

quantum computing. I've just been interested

37:32

in physics because I like

37:34

working on hard problems. I like

37:37

learning things. But I wanted

37:39

to apply that to the most relevant

37:41

things in our society today, which relate

37:43

to our energy systems. And

37:46

I went and worked with

37:48

energy companies as a management consultant,

37:50

with power companies and oil

37:52

and gas companies and industrial companies

37:54

who make power equipment, oil

37:56

field equipment. And my co -founder,

37:58

Kurt House, and I were doing...

38:00

equity investment in oil and

38:02

gas together in the private equity

38:04

firm whose leaders had sponsored

38:07

his previous startup company. And we

38:09

had become friends as graduate

38:11

students at Harvard together. He also

38:13

studied physics and philosophy as

38:15

an undergraduate and then applied math

38:17

and earth sciences as a

38:19

graduate student. And we would read

38:21

papers on energy topics and

38:23

go to visit power plants and

38:25

coal mines and things like

38:27

that. So we're already

38:29

working the energy system and quite

38:32

interested in how the raw materials relate

38:34

to the global economy. And we

38:36

decided we didn't want to work on

38:38

fossil fuels anymore. This is 2018.

38:40

And we thought from first principles about

38:42

what raw materials the future economy

38:44

will need and where those are going

38:46

to come from. And think about

38:49

all the raw materials that look around

38:51

you at your desk in your

38:53

house. And every one of these products

38:55

ultimately originated... From agriculture, we grew

38:57

it. Or from rocks, we mined it.

39:01

And so what materials are we going

39:03

to need? Well, think about the big

39:05

trends in the global economy. Batteries, AI. Batteries,

39:08

whether it's cars and trucks

39:10

or drones and aircrafts and robots.

39:12

To make a vehicle that has a

39:15

long range and is durable, the

39:17

battery needs lithium, which is different than

39:19

fuel -burning vehicles, which have no lithium

39:21

in them at all. AI.

39:24

Trend I don't have to explain to anyone who's listening

39:26

to this. Huge build out

39:29

of data centers and then the electricity

39:31

to power those data centers require an

39:33

enormous amount of copper. And the scale

39:35

we're talking about here is gigantic. To

39:37

build a future that is powered

39:40

by batteries and AI, by mid

39:42

-century, we will need to mine,

39:44

over the next 25 years, more

39:46

copper than has been mined so

39:48

far in all of human history. To

39:53

get to a high penetration of

39:55

battery -powered devices, we need a

39:57

tenfold increase in lithium production relative

39:59

to today. So where are these

40:02

materials going to from? We have

40:04

to go find more of these

40:06

and recognizing that the problem is

40:08

getting much much harder because innovation

40:10

has been slowing down. It's this

40:12

perfect application where we can use

40:14

technology to create a differentiated business

40:17

and it does something really important

40:19

for our society and that's really

40:21

personally motivating to me and other

40:23

people who joined COBOL. Very exciting.

40:25

It's cool what doing so. No,

40:27

congrats. I I hope you hope

40:30

you find others. Okay, we'll keep

40:32

you posted. Thanks, Josh. Find

40:34

us on Twitter at NoPriorsPod. Subscribe

40:36

to our YouTube channel If you want

40:38

to see our faces, follow the

40:40

show on Apple Podcasts, Spotify, Spotify wherever

40:43

you listen. That way you get a

40:45

new episode every week And up for emails

40:47

or find transcripts for every episode at no .com.