AI transcript
0:00:03 There is a phrase that’s common in the mining world,
0:00:05 which goes something like,
0:00:08 if you can’t grow it, you must mine it.
0:00:10 That’s so central to understand
0:00:12 that so much of our physical world around us,
0:00:16 our homes, our cars, everything on our table tops,
0:00:19 these things require materials that must be mined
0:00:21 from the earth’s surface and below.
0:00:25 That is Connie Chan, a 16Z general partner.
0:00:28 Connie has led investments into all kinds of companies,
0:00:30 from live shopping to religious super apps,
0:00:32 to AI leasing agents.
0:00:35 But one of Connie’s most important bets
0:00:37 isn’t your standard technology story.
0:00:39 It’s in a mineral exploration company,
0:00:41 one that uses artificial intelligence,
0:00:43 but also human intelligence
0:00:48 to find critical materials across five continents.
0:00:49 Now, for many of you,
0:00:51 this is a topic that is just starting to bubble up,
0:00:54 because if you want to build the future,
0:00:55 so many of the new technologies
0:00:57 that we’re talking about and dreaming about
0:00:59 are going to require more metals.
0:01:03 One very clear example is for electric vehicles.
0:01:05 EVs require massive batteries,
0:01:07 and those batteries need more copper,
0:01:09 more lithium, more nickel.
0:01:12 And there’s very clearly a big supply gap
0:01:14 that’s coming in a couple of decades.
0:01:16 EV cars right now globally account
0:01:18 for already 14% of car sales.
0:01:22 In China, it was the majority of car sales in 2024.
0:01:25 And these EVs require four X amount of copper
0:01:26 as a normal gas vehicle.
0:01:29 So if we want to power the spring revolution,
0:01:32 we definitely need more mining.
0:01:34 And it’s not just electric vehicles.
0:01:36 If you think about data centers,
0:01:38 BHP estimates by 2050,
0:01:40 we’re going to have 6% to 7% of the world’s copper
0:01:43 going directly to data centers.
0:01:45 Now, you might say, well, 2050 sounds so far away,
0:01:48 but the reality is it can take years to find the mine,
0:01:50 years to find the deposit,
0:01:52 years to then estimate the size of the deposit,
0:01:54 figure out how to extract it best,
0:01:56 then years to build out the mine,
0:02:00 and then decades to actually extract all of that metal.
0:02:04 Which means if we need more metal in one, two, three decades,
0:02:06 we have to find that metal today.
0:02:09 So as this topic moves to the forefront
0:02:12 of not only the technology conversation,
0:02:14 but the national conversation,
0:02:17 we brought in three guests from all over the COBOL team.
0:02:20 – Tom Hunt, VP of Technology at COBOL.
0:02:22 My whole career has been at the junction
0:02:25 of technology and climate change.
0:02:27 – My name is MfK Makai.
0:02:30 I am Zambian and I’m a trained mining and civil engineer.
0:02:34 I’ve been in the industry for a little over 16 years.
0:02:38 – Hi, George Gilchrist, VP of Geosciences for COBOLT,
0:02:40 geologist here in Johannesburg, South Africa.
0:02:44 – Tom, MfK and George have worked in a combination of projects
0:02:48 from solar printing to working in existing mines.
0:02:49 And in today’s episode,
0:02:53 we explore what makes these metals truly irreplaceable,
0:02:55 the current process for discovery,
0:02:58 how technology and data are changing the game,
0:03:00 plus how to decide when to drill, when?
0:03:04 – The single drill hole can cost up to a million dollars.
0:03:07 – So without further ado, let’s get started.
0:03:11 As a reminder, the content here
0:03:13 is for informational purposes only.
0:03:15 Should not be taken as legal, business, tax,
0:03:16 or investment advice,
0:03:19 or be used to evaluate any investment or security
0:03:20 and is not directed at any investors
0:03:23 or potential investors in any A16Z fund.
0:03:26 Please note that A16Z and its affiliates
0:03:27 may also maintain investments
0:03:29 in the companies discussed in this podcast.
0:03:32 For more details, including a link to our investments,
0:03:35 please see a16z.com/disclosures.
0:03:42 – To kick off, I’d like to talk about
0:03:44 why are these metals so critical?
0:03:46 – For the energy transition,
0:03:49 we will need to build about two billion electric vehicles,
0:03:51 which means that we actually have to discover
0:03:53 about a thousand new mines
0:03:56 in order to provide the lithium, nickel, copper, and cobalt
0:03:58 that’s going to go into those vehicles.
0:04:00 After we’ve built those vehicles,
0:04:02 we can recycle the batteries,
0:04:04 but we need to put the batteries into those cars
0:04:06 to begin with.
0:04:10 So there is no substitute for copper or lithium.
0:04:14 Copper is the second most conductive metal after silver,
0:04:17 and unless we find an enormous pile of silver,
0:04:20 we’re gonna be using copper long into the future.
0:04:23 Lithium is both the lightest element
0:04:25 and the most electronegative element,
0:04:28 and having worked on next-generation batteries,
0:04:30 there’s really no substitute for the energy density
0:04:32 you can get at lithium.
0:04:35 So these metals are irreplaceable
0:04:37 in the global supply chain for the energy transition,
0:04:41 and they’re also critical for the buildout of solar,
0:04:42 of the utility-scale batteries
0:04:45 of all kinds of new power generation,
0:04:47 as well as the next-generation data centers.
0:04:49 So there’s no shortage of demand
0:04:52 when it comes to these critical metals.
0:04:55 – What is the main problem that we’re trying to solve though?
0:04:56 Is there a shortage of supply?
0:04:58 Are these metals rare?
0:05:00 Are they very difficult to find?
0:05:03 – There’s plenty of metal in the Earth’s crust,
0:05:07 and the question is how do we find places
0:05:10 where the history of the Earth has concentrated
0:05:11 these metal deposits to the point
0:05:15 where we can extract them both cost-effectively
0:05:17 and with minimal environmental impact?
0:05:20 So the more concentrated a metal deposit is,
0:05:22 the less rock you have to process
0:05:25 in order to extract a certain amount of metal.
0:05:29 – Can you talk about how these metals are not that rare?
0:05:31 Does that mean we can find them in our backyard?
0:05:35 Should we be searching for it locally, domestically?
0:05:39 Like, where do these metal deposits usually live?
0:05:41 – Yeah, I think my kids and my neighbors dog-serve me,
0:05:43 try to look them over at the yard for metals,
0:05:46 but they don’t concentrate in very many places.
0:05:49 There’s a very unique combination of factors
0:05:53 that are required to concentrate any of the metals
0:05:56 into a small, mineable target,
0:05:59 and these will vary depending on what you’re looking for.
0:06:01 So one of the tricks in exploration
0:06:04 is to really become super familiar
0:06:06 with the deposit style that you’re targeting,
0:06:09 to understand where these controls are coming together.
0:06:11 So if you’re looking for copper,
0:06:13 it might be in a very different place
0:06:15 to where you might be looking for lithium.
0:06:18 The techniques that you use are different.
0:06:19 Even within copper,
0:06:21 there will be different types of deposits
0:06:22 that will have very different characteristics
0:06:25 and require different tools, different approaches.
0:06:27 You have to be really flexible.
0:06:31 There’s no formulated approach to making a discovery.
0:06:32 And the question now is,
0:06:36 can we do it faster or more effectively than we have been?
0:06:39 Can we use more of the data that we have at our disposal
0:06:41 to guide those decisions
0:06:43 and to really help identify
0:06:45 where the most prospective areas are?
0:06:49 And Cobble goes around the world to find the best rocks.
0:06:51 We go to the high arctic,
0:06:53 if that’s where the geological conditions
0:06:55 for our formation were best,
0:06:57 and the Central African Copper Belt
0:07:00 is one of the best copper locations in the entire world,
0:07:02 which is why we’re there.
0:07:03 – So give me a sense.
0:07:06 What does it mean to be high grade or low grade?
0:07:09 – Copper is quite a stark example.
0:07:12 Many of the big deposits that are mined
0:07:14 for copper or called porphyry copper deposits,
0:07:18 particularly in South America, Indonesia, and other places,
0:07:23 they will be by mass, half a percent, 0.6% copper.
0:07:25 In the copper belts,
0:07:28 the average deposits will be about two to 3% copper,
0:07:30 and the really high grade deposits,
0:07:33 such as the one that we’ve discovered at Mngomba,
0:07:36 will be a 5% to 6% copper.
0:07:38 So an order of magnitude higher in grade,
0:07:42 and that has significant advantages, obviously, economically,
0:07:43 but also environmentally,
0:07:46 you can mine a much smaller footprint
0:07:48 and produce a large volume of copper,
0:07:51 so it’s a very attractive target.
0:07:55 – So you’re saying if someone was extracting a ton of rock
0:07:58 that potentially less than 1% of that ton
0:07:59 would be usable copper?
0:08:01 – Yes, you’re always going to lose a little bit
0:08:04 in the mining and a little bit in the processing,
0:08:07 and so you end up getting less than 1% of that mass
0:08:10 that you’ve mined is your actual product.
0:08:11 – Right, George, a lot of mines today
0:08:15 are being expanded as opposed to new mines being discovered.
0:08:16 Why is that?
0:08:19 – It’s really hard to find new mines.
0:08:22 So 40, 50, 60 years ago,
0:08:25 a lot of the earth hadn’t been tested.
0:08:27 A lot of the deposits were at surface.
0:08:30 They might have had an expression at surface.
0:08:32 – Meaning I can see it with my eyes on the ground
0:08:34 or just digging with the shovel?
0:08:36 – Yes, so copper, if it gets to surface,
0:08:39 it will oxidize and form minerals that look green
0:08:40 or look blue.
0:08:42 And so in the copper belt, for instance,
0:08:45 there would be hills that had green staining on them.
0:08:49 So it wasn’t difficult to discover those deposits,
0:08:51 but those have largely been found.
0:08:54 I haven’t stumbled into a green hill
0:08:56 that’s just waiting to be mined.
0:08:59 So now we’ve got to start looking underneath the surface
0:09:01 and we need a lot more tools, a lot more data.
0:09:05 And so it’s easier to expand your existing mine
0:09:08 than spend the money to try and make a discovery elsewhere.
0:09:12 – So who usually discovers these deposits to date?
0:09:16 – A lot of the discoveries are made by smaller companies
0:09:19 that are really focused on just making discoveries.
0:09:21 They’re willing to take more risks.
0:09:23 They are willing to test new technology.
0:09:26 They’re willing to look deeper or undercover
0:09:28 or in areas that people might have turned away
0:09:31 or had the incorrect geology approach
0:09:32 to the wrong model.
0:09:36 So that attitude to exploration that drives success.
0:09:38 – And maybe before we jump into
0:09:40 how technology is changing mining,
0:09:42 maybe can you share with us
0:09:45 how has mining exploration changed over the decades?
0:09:47 One of the stories I’ve always loved in mining
0:09:50 was there was a story of how someone made
0:09:52 a huge gold deposit discovery
0:09:55 based off the cover of a National Geographic Magazine.
0:09:56 Because just like you said,
0:09:58 it was being expressed on the surface.
0:10:02 So one, two decades ago, what did exploration look like then?
0:10:05 And then we can contrast that to today.
0:10:07 A lot of the work we’re doing today
0:10:09 is still the same groundwork.
0:10:10 We’re still moving onto ground.
0:10:13 We’re still mapping the geology,
0:10:15 taking samples of the soil.
0:10:17 That’s been done for a long time.
0:10:18 It’s very effective.
0:10:21 What has changed is the level of technology
0:10:24 in terms of the geophysical methods that are available.
0:10:28 So the ability to measure the properties of the rock.
0:10:29 Is it magnetic?
0:10:32 Is one rock denser than another rock?
0:10:34 We can measure that through what we call gravity measurements.
0:10:36 We can use seismic surveys
0:10:41 to try and identify the shapes of deposits deep underground.
0:10:43 And those technologies have advanced dramatically.
0:10:45 They’ve become airborne.
0:10:46 So we can fly over deposits.
0:10:49 So we don’t need to build roads and bridges.
0:10:51 We can now access the ground easy.
0:10:53 They’ve become better in their resolution,
0:10:55 the quality of the data.
0:10:59 But with that has come significant volumes of data.
0:11:01 And so the ability to process
0:11:05 and extract the value out of that data is not a challenge.
0:11:09 – Tom, tell us how is AI being used for exploration
0:11:12 in a way that wasn’t true a decade ago?
0:11:15 – Yeah, I think the types of data that George described
0:11:17 are what we need to squeeze
0:11:19 in order to have the insights
0:11:22 that guide our exploration programs.
0:11:24 So we might start at the continent scale,
0:11:27 would say satellite data across all of Australia.
0:11:30 And we want to be able to find particular rock types
0:11:32 that might be indicative at the surface
0:11:34 of deposits that are very deeper.
0:11:37 So we can use image recognition
0:11:40 and classification algorithms developed
0:11:41 for other applications.
0:11:45 We can adapt those to then apply to our specific instance
0:11:47 of adding to insights that might lead
0:11:50 to finding these underground deposits.
0:11:53 But just one data source is not nearly enough.
0:11:56 We would also take magnetic data
0:11:57 across the entire continent.
0:12:00 We will take whatever kinds of geochemical data,
0:12:03 government data sets such as geologists
0:12:04 who’ve walked around, picked up a rock
0:12:07 and sent that rock in for chemical analysis.
0:12:10 They’re databases that are available
0:12:12 but extremely poorly structured
0:12:14 with multiple analysis methods
0:12:17 that we need to be able to clean up
0:12:19 so that we can actually feed it into our algorithms
0:12:22 and serve it up to our geoscientists.
0:12:25 We also take geostructural data.
0:12:27 So what is the topography?
0:12:29 Where are the mountains and what slopes
0:12:31 are the different rock layers coming in?
0:12:33 We can take all that data and use that
0:12:37 to inform the local interpretation of the geology.
0:12:41 So that’s going from the continent scale
0:12:43 to what we call the camp scale
0:12:46 or roughly 10 kilometers by 10 kilometers
0:12:49 as we narrow down the search for one of these deposits.
0:12:52 And then we need a whole different set of algorithms
0:12:54 in order to go from that 10 kilometer scale
0:12:57 to where do we actually drill
0:12:59 to try to answer the question of what’s underground?
0:13:03 You really can’t tell what’s underground until you drill.
0:13:06 And the single drill hole can cost up to a million dollars.
0:13:09 So we want to be able to optimize
0:13:10 where we place that drill hole.
0:13:13 So those are some of the types of algorithms
0:13:17 that we’ve built to give superpowers to our geologists.
0:13:19 – Quality of the data is so important.
0:13:21 And you mentioned a lot of the data
0:13:24 that Cobalt uses previously was very unstructured.
0:13:28 Give us a taste of just how difficult it was
0:13:31 to digitize the data to begin with
0:13:34 and then put it in a usable format.
0:13:38 – So very talented geologists have been walking
0:13:41 the surface of the earth for over a hundred years
0:13:43 and collecting sometimes handwritten records,
0:13:47 sometimes physical samples that are analyzed,
0:13:51 sometimes hand drawn maps, sometimes digital maps.
0:13:55 So they’re all this very diverse set of unstructured data.
0:13:58 And these are then sometimes in government databases,
0:14:01 either paper or digital archives,
0:14:04 or they are with potential joint venture partners
0:14:06 who might have huge pile of papers
0:14:09 that go to date back 50 or 100 years.
0:14:12 And there’s an incredible wealth of information there,
0:14:14 but information that hasn’t been exploited
0:14:16 ’cause it’s been locked up in these paper records.
0:14:18 So yeah, we’ve had teams out scanning
0:14:20 some of these paper records.
0:14:23 The job isn’t done once you have the records scanned.
0:14:26 We really want to get as much information
0:14:28 out of this raw data as we can.
0:14:31 And so for example, in Finland,
0:14:33 there’s a wealth of historic data, but it’s in Finnish.
0:14:37 And so we need to be able to both translate that.
0:14:40 And many of the words that people use to describe rocks
0:14:41 are highly specialized.
0:14:43 And so we need special translation modules
0:14:45 to make sure that those come through correctly
0:14:49 and then extract the structured data
0:14:50 out of these unstructured reports.
0:14:53 So for example, somebody may describe a certain rock type
0:14:55 and a certain grain size on that rock
0:14:57 and a certain chemistry of that rock.
0:14:59 And we would like to take that into a table
0:15:01 that our algorithms can then use.
0:15:03 – So it’s fascinating to me
0:15:05 that we basically have all these geoscientists
0:15:09 that then pair with these data scientists.
0:15:11 How do these teams work together
0:15:14 given that they’re coming from such different backgrounds?
0:15:15 – We’re actually fortunate
0:15:17 that we live in the world of technology
0:15:19 and somebody sitting in Silicon Valley
0:15:23 is working with someone sitting here in Central Africa
0:15:24 or in Australia.
0:15:27 And we’ve been very deliberate on making sure
0:15:29 that our data science software engineering
0:15:32 and our geoscience teams are highly collaborative
0:15:36 and are paired in being able to extract the rock
0:15:39 that we have drilled, upload it into the cloud
0:15:41 and anyone around the world within Cobalt
0:15:45 can look with the same precision at the same information
0:15:48 as if they were standing in front on the ground
0:15:50 here in South Central Africa.
0:15:53 We also have the abilities to actually look at the images
0:15:55 through all different models that we’re building
0:15:57 to analyze our core.
0:15:59 And that’s something that we build within the company
0:16:02 through the tech stack that we have.
0:16:04 – Even just the languages, the words,
0:16:07 how do you teach each other about your craft?
0:16:12 If you are a scientist, a biochemist, a geologist,
0:16:13 you’re bringing together people
0:16:15 of all different backgrounds.
0:16:17 How do you get everyone on the same page?
0:16:18 – One of the things we kind of do
0:16:20 with your first drawing, Cobalt,
0:16:23 we have some sort of nomenclature session
0:16:27 of what the basics mean in terms of geology and rocks,
0:16:30 in terms of also the tech and marrying those together,
0:16:33 like creating our own sort of internal glossary
0:16:35 that is easy for people from backgrounds
0:16:36 who have never been on a mine
0:16:38 or never been around exploration
0:16:41 and vice versa, people who have never understood
0:16:44 the different AI models that are used in the US.
0:16:46 And that starts quite early
0:16:48 when you’re inducting a new team member.
0:16:51 If you talk about some of the local staff we have
0:16:53 down to community members,
0:16:55 we’ve had indigenous words for copper,
0:16:56 like the word mukuba,
0:16:58 and we help teach our North America
0:16:59 and our South African colleagues,
0:17:02 like this word mukuba is copper in our local language.
0:17:04 It’s literally one language in a way.
0:17:07 We also have a Zambian data scientist on the team
0:17:09 and they come and spend many, many weeks on site.
0:17:11 They get to go to a drill rig.
0:17:13 They get to suggest many new ideas
0:17:15 with drilling companies, drilling contractors
0:17:17 who’ve kind of worked in a certain way
0:17:18 over industry for many, many years.
0:17:19 So we look at the rig and say,
0:17:23 how can we be more efficient in extracting the score
0:17:25 and collecting information
0:17:29 in a manner that reduces the time to process the information?
0:17:31 So we’ve been fortunate to work with contractors
0:17:34 that allow us to trial our hardware
0:17:37 that Tom and his team build, ship it out to Zambia,
0:17:39 put it next to a driller,
0:17:42 do some basic training on how do we capture
0:17:44 imaging of the core, like 360 imaging of the core
0:17:46 as it’s coming out of the ground,
0:17:48 which is really revolutionary.
0:17:51 Whereas the standard was you have to wait a couple of days,
0:17:54 take an image, take it right, stitch it together.
0:17:56 And a data scientist working with a geoscientist
0:17:59 is also getting training from a geoscientist
0:18:02 on just the type of lithology that they see in the rock,
0:18:03 what we’re looking for
0:18:06 and how do we make better interpretations
0:18:09 and better predictions in a meaningful way
0:18:11 through either a mixture of AI
0:18:12 and HI through all the experience
0:18:14 of the brilliant geoscientists
0:18:16 who’ve worked in industry for many, many years.
0:18:18 – That’s a great point how Cobalt is innovating
0:18:21 on both the hardware and the software front.
0:18:24 – That’s right, to feed the algorithms
0:18:27 and the geoscientists who are going to make
0:18:31 the next generation of discoveries of these deposits,
0:18:34 we need to have as many data types as we can
0:18:38 and put all of that data in front of them.
0:18:40 So we found some opportunities
0:18:42 where we could push the state of the art
0:18:45 to be able to collect more flavors of data more easily.
0:18:49 And so one of those is with hyperspectral imaging.
0:18:52 And hyperspectral just means many, many different colors
0:18:55 all the way from the visible through the infrared.
0:18:58 And what’s really exciting about hyperspectral imagery
0:18:59 is that in the infrared,
0:19:03 you can actually measure the absorption
0:19:05 of different molecules.
0:19:09 And so you can actually read out chemistry using light.
0:19:13 And we implemented that method.
0:19:15 So hyperspectral imagery on an airborne system
0:19:19 so that we can fly over the Cobalt claims,
0:19:22 take terabytes of data with a light aircraft
0:19:24 and then come back and process that data
0:19:27 with very high spatial and spectral resolution,
0:19:31 add together the ground truth that geologists
0:19:33 who’ve walked the ground and picked up rock samples
0:19:36 and then scan those with infrared spectroscopy
0:19:38 combined with this airborne imaging
0:19:41 to really build a system that hasn’t been built
0:19:43 in mineral exploration before
0:19:45 to be able to survey thousands of square kilometers
0:19:48 and automatically interpret exactly
0:19:49 what rocks are on the ground.
0:19:52 So compared to a hand drawn geologic map
0:19:54 that has a rich history,
0:19:57 we can actually make a data driven geologic map.
0:19:58 – That’s great.
0:19:59 And George, given that you’ve worked
0:20:03 at more traditional exploration of mining companies before,
0:20:06 maybe share some examples of how this technology
0:20:08 or the data has surprised you.
0:20:10 – Yeah, there’ve been a number of examples.
0:20:13 Some of the countries have big data sets
0:20:15 and these are data sets that have been accumulated
0:20:17 over the years from explorers
0:20:18 and different parts of the country.
0:20:21 Each of those explorers was looking for something specific.
0:20:23 And so they didn’t assay necessarily for everything.
0:20:26 They didn’t measure every element in every sample.
0:20:29 They were just targeting a few.
0:20:32 And now we might be looking for a very specific element
0:20:34 and it’s only available in a small portion
0:20:35 of that data set.
0:20:37 Normally you think, oh, it’s such a pity
0:20:39 that we have this huge data set
0:20:41 but we can only use a small part of it.
0:20:44 Whereas the data scientists will say, well, that’s okay.
0:20:46 That element that we’re interested in
0:20:49 has relationships to other elements.
0:20:53 And we can have a really good idea
0:20:55 of what the grade of that element would be
0:20:56 given the grade of all the other elements
0:20:59 that we know at each of the sample points.
0:21:01 So we can test it by taking out the examples
0:21:03 where we do have that grade.
0:21:05 We can then estimate what that grade would be
0:21:06 and we compare it.
0:21:08 And that’s remarkably close
0:21:09 because we’re not just comparing it
0:21:11 to one element or two elements.
0:21:14 We’ll be looking at relationships from numerous elements.
0:21:16 And so suddenly there’s huge data sets
0:21:18 that look like at a trunk.
0:21:20 We’re actually able to use the full value
0:21:22 and the spread of that data.
0:21:24 And that allows us to move into areas
0:21:25 where other people wouldn’t be.
0:21:26 That’s one way.
0:21:30 Another way is Tom spoke about data that you digitize, maps.
0:21:33 If I scan a map, I can look at it on a computer screen,
0:21:36 but I have to look at it really carefully.
0:21:38 Now I can just look for a search term
0:21:42 and the 12 maps that have that term will pop up
0:21:44 and it’ll show me where that term is on the map,
0:21:45 straight away.
0:21:46 And then I’m able to interrogate,
0:21:49 maybe it’s the name of the drill hole
0:21:52 or it’s a certain element that I’m looking for
0:21:55 in a report on a series of maps.
0:21:57 And suddenly all of this information
0:21:59 is just so much quicker to interrogate.
0:22:02 I can spend my time applying my geological training
0:22:05 and my experience rather than spend my time
0:22:06 just opening and closing things.
0:22:09 So that’s a huge advantage to be able to advance
0:22:11 the search for the new discoveries.
0:22:13 – With all of this data,
0:22:16 how do you guys know what to prioritize?
0:22:17 How do you decide what kind of information
0:22:18 is more important?
0:22:22 And then how does that guide the actual work on the field?
0:22:26 – In a world that is rich in data that becomes the challenge
0:22:29 is how do you know what to actually focus on?
0:22:32 Some of that will come from our experience
0:22:34 with working with data from other projects.
0:22:37 So the geoscientists will be interacting
0:22:39 with the data scientists saying in this environment,
0:22:42 we know that these factors are really critical.
0:22:44 And this is where the collaboration
0:22:46 between the geoscientists and the data scientists
0:22:50 becomes so important that it’s not two separate entities.
0:22:53 It’s very collaborative and specific.
0:22:56 And so there’s no off-the-shelf option.
0:22:58 We’re not developing a tool that we consult
0:23:01 when exploration company that will help them discover
0:23:02 in any environment.
0:23:04 Everything we’re doing is tailored
0:23:06 to the areas that we’re working.
0:23:09 – And one key aspect is also modeling the uncertainty
0:23:10 of what’s underground.
0:23:12 And there’s incredible uncertainty
0:23:14 of what’s just under the surface.
0:23:17 And so by being able to map out the regions
0:23:20 of high uncertainty or low uncertainty,
0:23:22 that can also allow us to optimize
0:23:24 where we collect the next data point.
0:23:27 And so we can start with these publicly available
0:23:30 or joint venture style data sets.
0:23:32 But ultimately we have to go to the field
0:23:34 and collect our own data.
0:23:36 – And I guess that uncertainty also brings us
0:23:38 to the question of with all of this technology,
0:23:42 how has it improved our accuracy when we do drill?
0:23:45 Is there a reduction that you see already
0:23:47 in the number of drills it’s taking
0:23:49 for us to get more information?
0:23:52 – So we want to basically quantify uncertainty
0:23:55 and by being able to drill in an area
0:23:59 where we’ll maximize the amount of information we get
0:24:01 that will inform both the geoscience
0:24:04 and data science team of what is happening
0:24:07 within the underlying rocks is very different
0:24:11 than I’ll say the traditional way of from top’s point,
0:24:14 you have a camp of a 10 by 10 square kilometer
0:24:17 and you place a grid drill and hope for the best.
0:24:21 We are very targeted on the areas
0:24:24 that both the geoscience and data science team
0:24:26 want to drill a hole to the angle the hole
0:24:29 should be drilled at to what we’ll call the pierce point
0:24:33 of where we think the resource may lie.
0:24:36 And within a 24 month period,
0:24:39 the level of precision has encouraged a lot of confidence
0:24:42 particularly on our project in Zambia,
0:24:43 we’ve moved up to 10 rigs.
0:24:46 And obviously with each rig,
0:24:48 we know there’s a level of uncertainty attached to it
0:24:51 but we want to quantify that as much as possible.
0:24:54 – So as you talk about drilling for information then,
0:24:56 does that mean sometimes we’re drilling
0:24:58 not just to look for the resource
0:25:00 but we’re drilling whatever hole will give us
0:25:03 the most information that will either confirm
0:25:06 or deny a bunch of hypotheses?
0:25:10 – Yes, and we want to basically falsify the hypotheses quickly
0:25:12 because you could be drilling into perpetuity
0:25:15 but to somebody somewhere that’s a cost.
0:25:19 So even if you drill and you falsify a particular hypothesis,
0:25:21 you do not find what you’re looking for
0:25:24 or you affirm what you think was happening in the system,
0:25:26 that is a lot of information
0:25:29 that normally in traditional mining exploration,
0:25:32 someone’s like, oh, we didn’t find what we’re looking for
0:25:35 but we learned something from that particular hole.
0:25:38 And that learning takes us back to saying,
0:25:41 all right, scratch off hypothesis number seven,
0:25:43 let’s come up with a new hypothesis.
0:25:46 It’s basically data-driven decision-making.
0:25:50 – Yeah, a lot of the decisions around how to explore
0:25:53 would be driven by what is the sort of precedent
0:25:54 in this environment?
0:25:56 What grid size would we be drilling?
0:25:58 What did the neighbors draw?
0:26:01 What’s been the traditional drill spacing?
0:26:04 Some of it might be driven by perceived regulatory requirements
0:26:06 to reach a certain density of data.
0:26:09 The interesting thing is that as geologists,
0:26:13 we know that we have a limited understanding
0:26:14 of what’s deep underground.
0:26:16 And when you talk to one of the famous things
0:26:19 about geologists is that they get a room of geologists
0:26:23 together, you’ll get n plus one number of opinions.
0:26:24 You know, that’s where the joke is,
0:26:26 there’s four geologists, there’s five opinions.
0:26:28 And inherently we know that,
0:26:31 but that isn’t built into the decision-making process.
0:26:34 We aren’t building models that account
0:26:36 for all the different possibilities
0:26:37 that we know could exist.
0:26:41 And so typically we will anchor on one model
0:26:44 and our drilling and our sampling and our exploration
0:26:47 will focus on that model and we’ll keep testing that model
0:26:49 until either eventually we realize it’s not working
0:26:52 or we hope it’s going to hold.
0:26:55 Whereas at Cobalt, we are happy to hold multiple models
0:26:57 at the same time.
0:27:00 And we will test each of those models simultaneously
0:27:02 ’cause you’re testing the same space.
0:27:05 But we design the drill holes to maximize
0:27:08 how many of these models can we effectively test
0:27:10 with each of these holes that we’re drilling?
0:27:13 Where are the specific areas that are gonna solve
0:27:14 some of the problems that we need to know
0:27:18 about which hypothesis is valid and which one isn’t?
0:27:19 And so actually building that in
0:27:22 and being able to show the uncertainty
0:27:24 and how we are reducing that uncertainty
0:27:27 as we continue to drill is a really key focus
0:27:30 and it’s something that’s very unusual in the exploration.
0:27:34 – One example of applying our AI tools
0:27:37 to lithium exploration is actually in Canada
0:27:41 where we knew that the right geologic processes
0:27:43 had happened within this belt of granite.
0:27:46 We knew that this is the right flavor of granite
0:27:48 that might host a lithium deposit,
0:27:52 but we didn’t know where in hundreds of square kilometers
0:27:57 with no road access where you had to helicopter in a crew
0:28:01 to go look at the ground where there might be signs
0:28:02 of lithium.
0:28:07 And in order to investigate this very large area,
0:28:10 the traditional way would be to drop off a field team
0:28:13 and maybe pick them up at the end of the summer
0:28:16 and they will have walked across as much terrain
0:28:19 as they could walk across and found whatever lithium
0:28:21 they could find at the surface.
0:28:24 That’s not fast enough or cheap enough.
0:28:26 And so we knew we could do better than that
0:28:28 standard practice.
0:28:31 And so what we did was start with satellite data
0:28:36 and start with reports across a significant fraction of Quebec
0:28:40 about where lithium had been found previously.
0:28:43 And then we matched the satellite imagery
0:28:48 with those records of lithium-bearing rocks
0:28:52 and we’re then able to predict where in our claims
0:28:54 there might be lithium-bearing rocks.
0:28:56 So then we sent our people out in the field
0:29:00 to either confirm or deny the existence of lithium
0:29:02 at the places where our machine learning models were
0:29:05 predicting there might be lithium.
0:29:09 And they reported back immediately via satellite link
0:29:13 and said, guys, you just sent us to a vast field of lichen.
0:29:14 There aren’t even rocks here.
0:29:16 Everything’s covered in lichen.
0:29:18 And that’s not what we were looking for.
0:29:22 But that led us update our models overnight
0:29:26 so we could rerun models across 1,000 square kilometers
0:29:29 overnight with this new ground truth of what our people had
0:29:30 actually found on the ground.
0:29:32 And so with the addition of both the false positives
0:29:35 and the true positives, we were able to improve the accuracy
0:29:38 of our model prediction by over an order of magnitude.
0:29:40 And then in the following several weeks
0:29:43 that we had this helicopter supported field program,
0:29:45 we were actually able to find spodgamine, which
0:29:48 is the lithium-bearing mineral that we were looking for.
0:29:50 So that’s an example of compressing
0:29:54 what might have taken years of expensive remote field work
0:29:57 into the period of a week or so by being able to iterate
0:30:00 and update these machine learning models.
0:30:03 If I compare what is happening in the oil and gas space
0:30:06 and how they marry traditional methods and technology,
0:30:10 how does that compare with the mining space?
0:30:13 Is mining ahead, behind of oil and gas,
0:30:15 or are there shared learnings?
0:30:17 The technology used in the mining industry
0:30:20 is dramatically behind what we’ve
0:30:22 used to search for oil and gas.
0:30:26 And part of our job is to take these amazing technologies that
0:30:29 have been developed to find fossil fuels
0:30:32 and then adjust them so that perhaps they’re
0:30:37 smaller and cheaper so they’re applicable to discovering
0:30:39 metals rather than oil and gas.
0:30:42 And so examples of that are directional drilling,
0:30:46 where in oil and gas now you can precisely control
0:30:49 the trajectory of exactly where you want to drill.
0:30:52 This is new for the mining industry.
0:30:55 We need to make this technology smaller
0:30:59 so that it’s appropriate for the kinds of brakes that we use.
0:31:01 Another example is geophysical techniques
0:31:05 that were initially invented for finding oil and gas.
0:31:09 We can adjust those to whether that’s seismic imaging,
0:31:12 whether that’s electromagnetic imaging.
0:31:13 We can adjust those so that they’re
0:31:16 suitable for finding the kinds of more bodies we’re looking for.
0:31:19 So really, it’s taking tools that have been developed
0:31:21 by other industries, such as oil and gas,
0:31:23 or in the case of these artificial intelligence
0:31:26 algorithms, tools that have been developed for image
0:31:28 processing, we can take those and then apply them
0:31:30 to the kinds of complex problems that we face.
0:31:34 Oftentimes, we’re talking about copper, nickel, lithium.
0:31:38 Does the approach work for other metals underground?
0:31:42 And how do you guys think about the expanse of what
0:31:45 this technology can be applied towards?
0:31:48 Yeah, it definitely has applicability
0:31:49 across any of the search spaces.
0:31:51 The processes that are being developed
0:31:54 are solving data problems.
0:31:58 They are looking for controls on mineralization.
0:32:00 And you can adjust those as you’re
0:32:01 looking at a different commodity.
0:32:04 So what’s controlling a nickel deposit?
0:32:06 It’s very different to what’s controlling a lithium deposit.
0:32:11 This is the beauty of not approaching a commercial software
0:32:13 vendor to try and solve your problems.
0:32:16 You still have the data science team and the geoscience team
0:32:18 working so closely together to say,
0:32:21 these are the problems in this specific environment.
0:32:24 Even if you’re thinking of nickel exploration,
0:32:26 nickel deposits, all of them are different.
0:32:28 Yeah, the technology really is a toolbox
0:32:31 that we can apply to different exploration programs
0:32:34 around the world in multiple mineral commodities.
0:32:36 And so the same exact tool that you’d
0:32:39 use in Australia to find lithium is probably
0:32:43 different from the one that you’d use to find copper in the Arctic.
0:32:45 But they have many parallels.
0:32:46 And when I started at Cobold, I was
0:32:48 new to the mineral exploration industry.
0:32:51 And I was thinking that there’d be an incredibly sophisticated
0:32:54 way that society finds the materials
0:32:56 that we use on a day-to-day basis.
0:32:59 But actually, the tools in mineral exploration
0:33:01 are shockingly primitive.
0:33:03 And we can rapidly move beyond the state of the art
0:33:05 in mineral exploration.
0:33:09 I think one anecdote about the state of the industry
0:33:14 is just how much it relies on looking at rocks,
0:33:17 where a good geologist can look at a rock
0:33:20 and tell you from the context of what’s around that rock,
0:33:22 from the grain size, from the color,
0:33:25 from the mineralogy of that rock,
0:33:28 can tell you just so much about what
0:33:31 has happened over the last billion years at that location.
0:33:34 And yet that’s all done with the human eye.
0:33:37 And so it’s said that the best geologist is the one that’s
0:33:38 seen the most rocks.
0:33:41 Well, what if we make a system that
0:33:46 can look at more rocks than any human has ever looked at before
0:33:50 by just analyzing the imagery of the core that we’ve
0:33:55 drilled at Mingonba, we now have nearly 100 kilometers
0:33:58 of rock core that we’ve pulled out of the ground?
0:34:00 That would take years for a person
0:34:03 to look at every little part of that core.
0:34:07 And yet this is perfectly suited to machine learning
0:34:10 algorithms that can look at millions of images
0:34:11 in the blink of an eye.
0:34:15 And so being able to take the real knowledge that
0:34:19 is embedded in hundreds of years of geoscience
0:34:25 and then apply new emerging tools to help analyze and extract
0:34:27 real information from those techniques,
0:34:29 I think it’s just incredibly promising.
0:34:30 That’s amazing.
0:34:32 Lots of times when I’m reading in the headlines now,
0:34:35 I see this phrase, critical minerals.
0:34:38 Do you guys have a sense of which critical minerals
0:34:41 are important to national interest, for example?
0:34:44 Or how would you define what is a critical mineral?
0:34:46 Yeah, if you look at all the lists of critical minerals,
0:34:49 almost every element on the periodic table
0:34:50 is in one of those lists.
0:34:53 And so to me, what a critical mineral means
0:34:57 is something that’s important for our economy
0:35:00 and our defense industrial base.
0:35:04 And if you look at where our economy is trending,
0:35:07 our economy is trending towards electrification,
0:35:09 towards electrification of everything.
0:35:11 And so the most important critical minerals
0:35:16 are those that allow us to electrify the economy.
0:35:18 And the most important materials for electrifying
0:35:22 the economy really come down to copper and lithium
0:35:23 being the main ones.
0:35:25 And then there’s a long tail of other critical minerals
0:35:27 beyond that.
0:35:29 Yeah, I think there’s a lot of minerals
0:35:32 that can perform a similar role, some just more efficiently
0:35:33 than others.
0:35:38 And as supply concerns or prices dictates,
0:35:40 some can be substituted out.
0:35:43 But some elements are just so fantastic at what they do
0:35:47 that they are the critical minerals for electrification.
0:35:51 Copper is just so good at transmitting electricity.
0:35:54 Lithium is so good in batteries.
0:35:58 And it’s really hard to see how you can, on a large scale,
0:36:00 remove such elements.
0:36:01 Other considerations are elements
0:36:05 where the supply is concentrated in very small areas.
0:36:08 Things that on local or shorter timescales
0:36:10 might become really critical.
0:36:11 Yeah, tell me more about those.
0:36:13 Are they concentrated in specific continents?
0:36:16 Or is it even at a country level?
0:36:18 Rare earth elements haven’t always
0:36:22 been top of the exploration ladder in terms of what people
0:36:24 are really interested in looking for.
0:36:26 And so deposits will go through phases
0:36:29 where you will discover deposits, define them,
0:36:30 and then prices will drop.
0:36:32 And those deposits are known about.
0:36:33 But no one’s interested in them.
0:36:35 And then there’ll be a price shock.
0:36:36 And they will not jump back into them.
0:36:39 And so rare earth deposits are widely spread.
0:36:43 They are well represented in the geological record.
0:36:45 But they haven’t been widely explored for.
0:36:47 So lithium is actually a really good example
0:36:49 that for many, many, many years, people
0:36:51 were not that interested in lithium.
0:36:53 It was used in very niche applications.
0:36:55 And it’s only since batteries have rarely
0:36:58 become so critical that suddenly people like, wait a minute,
0:37:01 where’s the historic database on lithium deposits for?
0:37:04 It’s small because people haven’t been focused on it.
0:37:06 And I think some of the rare earth elements
0:37:09 fall into those categories where for a long time they’ve
0:37:15 been quite niche and now are becoming a lot more important
0:37:17 to technologies going forward.
0:37:18 So traditionally, as a geologist,
0:37:21 when you’re thinking about what to go explore for before it
0:37:25 might be in terms of market size or financially,
0:37:27 does it make sense to go explore for this,
0:37:29 given where the prices are for those metals?
0:37:31 And now it seems like there’s another driver, which
0:37:33 could be national interest.
0:37:36 Are there certain metals that even if the numbers don’t
0:37:40 pan out today, there might be other incentives
0:37:43 to build up more deposits for these other metals?
0:37:44 Yes.
0:37:46 And I think in the mining industry,
0:37:52 there hasn’t been a large transition of sustained demand
0:37:53 for certain elements.
0:37:56 And that’s happening and has been in progress now
0:37:58 for a few years and will continue.
0:38:01 And that’s quite a change from what’s happened historically.
0:38:03 We’re always going to need iron ore.
0:38:04 We’re always going to need copper.
0:38:08 But in the last few years, other elements
0:38:10 have suddenly become a lot more critical than they’ve
0:38:12 been in the past.
0:38:14 And it only looks like the demand is
0:38:16 going to grow for those elements.
0:38:19 And there’s a fundamental shift and a lot of companies
0:38:22 are refocusing efforts into these elements that
0:38:24 haven’t had a lot of love from the exploration world
0:38:26 for a long time.
0:38:29 It’s more complicated than just a supply and demand
0:38:30 on an open market.
0:38:33 There is a country competition where some of those elements
0:38:36 might not be available.
0:38:40 And so that also becomes a consideration, definitely.
0:38:40 Yeah.
0:38:44 And that country competition and the impact on countries,
0:38:48 what is Zambia’s response to this new cobalt approach?
0:38:51 And how has it impacted, also, does Africa as a whole?
0:38:54 It’s been absolutely positive.
0:38:57 Definitely a breath of fresh air for all of us
0:38:59 who’ve been in the sector here.
0:39:02 And generally, as a continent, there
0:39:05 is this big drive because we have this huge youthful population
0:39:07 to drive industrialization.
0:39:09 This is what many African governments want.
0:39:12 And an investment like cobalt into Zambia
0:39:16 is stimulating opportunities, knowing that beyond what we do
0:39:18 in exploration and eventually mining,
0:39:21 we’re actually driving the development of the youngest
0:39:24 population on the continent and a lot of other trade
0:39:27 and business opportunities, even for US enterprises,
0:39:30 looking to do business in Zambia and in different parts
0:39:31 of Africa.
0:39:36 So it’s a seed that basically is getting in the door,
0:39:38 meeting the right officials, and stating the intention
0:39:41 very transparently, very clearly,
0:39:43 and operating within the bounds of the law
0:39:45 that we know globally.
0:39:47 And that’s quite exciting for the country,
0:39:50 but it’s also getting the neighbors waking up and saying,
0:39:52 wait a minute, if we integrate more,
0:39:54 there’s this big project in Zambia.
0:39:55 How do we participate in it?
0:39:57 How do we use the Lobito Corridor,
0:40:01 which we know will be a conduit for some of these minerals
0:40:04 that will help us drive electrifying the world?
0:40:07 You’re talking about a continent with a billion people,
0:40:09 or these billion people in coming decades
0:40:12 who need all sorts of consumer items and electronics,
0:40:14 and the US gets the firsthand look into that
0:40:18 by having access into Africa through the Corridor.
0:40:21 Just along the corridor, the population of the three countries
0:40:24 combined is close to the US population as a whole.
0:40:27 And growing, even as we build these mines,
0:40:30 how we mine, when we mine,
0:40:31 and how much material we impact,
0:40:34 and even how we’re able to look at the environment
0:40:38 through the mining process is going to rapidly transform
0:40:40 basically on everything that we’re doing
0:40:41 with the information we have
0:40:43 and how we can analyze it statistically.
0:40:45 So mines of the future are also going
0:40:46 to be fundamentally different.
0:40:48 I know we talked about exploration,
0:40:50 but as earlier said by Connie,
0:40:53 the image of mining has been poor in past decades,
0:40:56 and something we want to do is change the image of mining,
0:40:59 and that change will come to how we do a lot more things
0:41:01 with more prediction, more precision,
0:41:04 even as we build out many, many projects,
0:41:06 and that basically becomes institutionalized
0:41:08 in the industry.
0:41:09 – I think from my perspective,
0:41:13 the industry is perceived as quite an established,
0:41:18 mature industry, and from a geology perspective,
0:41:19 a lot of the easy deposits of perception
0:41:22 as they’ve been found, and that’s only going to get harder,
0:41:25 and it doesn’t sound like a good marketing touch.
0:41:29 But the reality is actually this time now
0:41:33 is an amazing time to be in exploration.
0:41:36 The ability to take new technologies
0:41:39 and apply them to solve the problems,
0:41:42 and knowing that you’re absolutely critical
0:41:45 to ensuring that we can solve the problems
0:41:47 that we need to solve globally.
0:41:49 We don’t do our job properly.
0:41:52 We’re going to slow down the ability to solve problems.
0:41:55 – Yeah, that sort of focus has been missing,
0:41:57 I think, from the industry for a long time,
0:41:59 and it’s awesome to be part of that now.
0:42:04 – All right, that is all for today.
0:42:06 If you did make it this far, first of all, thank you.
0:42:08 We put a lot of thought into each of these episodes,
0:42:11 whether it’s guests, the calendar tetris,
0:42:12 the cycles with our amazing editor, Tommy,
0:42:14 until the music is just right.
0:42:16 So if you like what we’ve put together,
0:42:21 consider dropping us a line at ratethespodcast.com/a16c.
0:42:23 And let us know what your favorite episode is.
0:42:26 It’ll make my day, and I’m sure Tommy’s too.
0:42:28 We’ll catch you on the flip side.
0:42:31 (upbeat music)
0:42:34 (upbeat music)
0:42:37 (upbeat music)
0:00:05 which goes something like,
0:00:08 if you can’t grow it, you must mine it.
0:00:10 That’s so central to understand
0:00:12 that so much of our physical world around us,
0:00:16 our homes, our cars, everything on our table tops,
0:00:19 these things require materials that must be mined
0:00:21 from the earth’s surface and below.
0:00:25 That is Connie Chan, a 16Z general partner.
0:00:28 Connie has led investments into all kinds of companies,
0:00:30 from live shopping to religious super apps,
0:00:32 to AI leasing agents.
0:00:35 But one of Connie’s most important bets
0:00:37 isn’t your standard technology story.
0:00:39 It’s in a mineral exploration company,
0:00:41 one that uses artificial intelligence,
0:00:43 but also human intelligence
0:00:48 to find critical materials across five continents.
0:00:49 Now, for many of you,
0:00:51 this is a topic that is just starting to bubble up,
0:00:54 because if you want to build the future,
0:00:55 so many of the new technologies
0:00:57 that we’re talking about and dreaming about
0:00:59 are going to require more metals.
0:01:03 One very clear example is for electric vehicles.
0:01:05 EVs require massive batteries,
0:01:07 and those batteries need more copper,
0:01:09 more lithium, more nickel.
0:01:12 And there’s very clearly a big supply gap
0:01:14 that’s coming in a couple of decades.
0:01:16 EV cars right now globally account
0:01:18 for already 14% of car sales.
0:01:22 In China, it was the majority of car sales in 2024.
0:01:25 And these EVs require four X amount of copper
0:01:26 as a normal gas vehicle.
0:01:29 So if we want to power the spring revolution,
0:01:32 we definitely need more mining.
0:01:34 And it’s not just electric vehicles.
0:01:36 If you think about data centers,
0:01:38 BHP estimates by 2050,
0:01:40 we’re going to have 6% to 7% of the world’s copper
0:01:43 going directly to data centers.
0:01:45 Now, you might say, well, 2050 sounds so far away,
0:01:48 but the reality is it can take years to find the mine,
0:01:50 years to find the deposit,
0:01:52 years to then estimate the size of the deposit,
0:01:54 figure out how to extract it best,
0:01:56 then years to build out the mine,
0:02:00 and then decades to actually extract all of that metal.
0:02:04 Which means if we need more metal in one, two, three decades,
0:02:06 we have to find that metal today.
0:02:09 So as this topic moves to the forefront
0:02:12 of not only the technology conversation,
0:02:14 but the national conversation,
0:02:17 we brought in three guests from all over the COBOL team.
0:02:20 – Tom Hunt, VP of Technology at COBOL.
0:02:22 My whole career has been at the junction
0:02:25 of technology and climate change.
0:02:27 – My name is MfK Makai.
0:02:30 I am Zambian and I’m a trained mining and civil engineer.
0:02:34 I’ve been in the industry for a little over 16 years.
0:02:38 – Hi, George Gilchrist, VP of Geosciences for COBOLT,
0:02:40 geologist here in Johannesburg, South Africa.
0:02:44 – Tom, MfK and George have worked in a combination of projects
0:02:48 from solar printing to working in existing mines.
0:02:49 And in today’s episode,
0:02:53 we explore what makes these metals truly irreplaceable,
0:02:55 the current process for discovery,
0:02:58 how technology and data are changing the game,
0:03:00 plus how to decide when to drill, when?
0:03:04 – The single drill hole can cost up to a million dollars.
0:03:07 – So without further ado, let’s get started.
0:03:11 As a reminder, the content here
0:03:13 is for informational purposes only.
0:03:15 Should not be taken as legal, business, tax,
0:03:16 or investment advice,
0:03:19 or be used to evaluate any investment or security
0:03:20 and is not directed at any investors
0:03:23 or potential investors in any A16Z fund.
0:03:26 Please note that A16Z and its affiliates
0:03:27 may also maintain investments
0:03:29 in the companies discussed in this podcast.
0:03:32 For more details, including a link to our investments,
0:03:35 please see a16z.com/disclosures.
0:03:42 – To kick off, I’d like to talk about
0:03:44 why are these metals so critical?
0:03:46 – For the energy transition,
0:03:49 we will need to build about two billion electric vehicles,
0:03:51 which means that we actually have to discover
0:03:53 about a thousand new mines
0:03:56 in order to provide the lithium, nickel, copper, and cobalt
0:03:58 that’s going to go into those vehicles.
0:04:00 After we’ve built those vehicles,
0:04:02 we can recycle the batteries,
0:04:04 but we need to put the batteries into those cars
0:04:06 to begin with.
0:04:10 So there is no substitute for copper or lithium.
0:04:14 Copper is the second most conductive metal after silver,
0:04:17 and unless we find an enormous pile of silver,
0:04:20 we’re gonna be using copper long into the future.
0:04:23 Lithium is both the lightest element
0:04:25 and the most electronegative element,
0:04:28 and having worked on next-generation batteries,
0:04:30 there’s really no substitute for the energy density
0:04:32 you can get at lithium.
0:04:35 So these metals are irreplaceable
0:04:37 in the global supply chain for the energy transition,
0:04:41 and they’re also critical for the buildout of solar,
0:04:42 of the utility-scale batteries
0:04:45 of all kinds of new power generation,
0:04:47 as well as the next-generation data centers.
0:04:49 So there’s no shortage of demand
0:04:52 when it comes to these critical metals.
0:04:55 – What is the main problem that we’re trying to solve though?
0:04:56 Is there a shortage of supply?
0:04:58 Are these metals rare?
0:05:00 Are they very difficult to find?
0:05:03 – There’s plenty of metal in the Earth’s crust,
0:05:07 and the question is how do we find places
0:05:10 where the history of the Earth has concentrated
0:05:11 these metal deposits to the point
0:05:15 where we can extract them both cost-effectively
0:05:17 and with minimal environmental impact?
0:05:20 So the more concentrated a metal deposit is,
0:05:22 the less rock you have to process
0:05:25 in order to extract a certain amount of metal.
0:05:29 – Can you talk about how these metals are not that rare?
0:05:31 Does that mean we can find them in our backyard?
0:05:35 Should we be searching for it locally, domestically?
0:05:39 Like, where do these metal deposits usually live?
0:05:41 – Yeah, I think my kids and my neighbors dog-serve me,
0:05:43 try to look them over at the yard for metals,
0:05:46 but they don’t concentrate in very many places.
0:05:49 There’s a very unique combination of factors
0:05:53 that are required to concentrate any of the metals
0:05:56 into a small, mineable target,
0:05:59 and these will vary depending on what you’re looking for.
0:06:01 So one of the tricks in exploration
0:06:04 is to really become super familiar
0:06:06 with the deposit style that you’re targeting,
0:06:09 to understand where these controls are coming together.
0:06:11 So if you’re looking for copper,
0:06:13 it might be in a very different place
0:06:15 to where you might be looking for lithium.
0:06:18 The techniques that you use are different.
0:06:19 Even within copper,
0:06:21 there will be different types of deposits
0:06:22 that will have very different characteristics
0:06:25 and require different tools, different approaches.
0:06:27 You have to be really flexible.
0:06:31 There’s no formulated approach to making a discovery.
0:06:32 And the question now is,
0:06:36 can we do it faster or more effectively than we have been?
0:06:39 Can we use more of the data that we have at our disposal
0:06:41 to guide those decisions
0:06:43 and to really help identify
0:06:45 where the most prospective areas are?
0:06:49 And Cobble goes around the world to find the best rocks.
0:06:51 We go to the high arctic,
0:06:53 if that’s where the geological conditions
0:06:55 for our formation were best,
0:06:57 and the Central African Copper Belt
0:07:00 is one of the best copper locations in the entire world,
0:07:02 which is why we’re there.
0:07:03 – So give me a sense.
0:07:06 What does it mean to be high grade or low grade?
0:07:09 – Copper is quite a stark example.
0:07:12 Many of the big deposits that are mined
0:07:14 for copper or called porphyry copper deposits,
0:07:18 particularly in South America, Indonesia, and other places,
0:07:23 they will be by mass, half a percent, 0.6% copper.
0:07:25 In the copper belts,
0:07:28 the average deposits will be about two to 3% copper,
0:07:30 and the really high grade deposits,
0:07:33 such as the one that we’ve discovered at Mngomba,
0:07:36 will be a 5% to 6% copper.
0:07:38 So an order of magnitude higher in grade,
0:07:42 and that has significant advantages, obviously, economically,
0:07:43 but also environmentally,
0:07:46 you can mine a much smaller footprint
0:07:48 and produce a large volume of copper,
0:07:51 so it’s a very attractive target.
0:07:55 – So you’re saying if someone was extracting a ton of rock
0:07:58 that potentially less than 1% of that ton
0:07:59 would be usable copper?
0:08:01 – Yes, you’re always going to lose a little bit
0:08:04 in the mining and a little bit in the processing,
0:08:07 and so you end up getting less than 1% of that mass
0:08:10 that you’ve mined is your actual product.
0:08:11 – Right, George, a lot of mines today
0:08:15 are being expanded as opposed to new mines being discovered.
0:08:16 Why is that?
0:08:19 – It’s really hard to find new mines.
0:08:22 So 40, 50, 60 years ago,
0:08:25 a lot of the earth hadn’t been tested.
0:08:27 A lot of the deposits were at surface.
0:08:30 They might have had an expression at surface.
0:08:32 – Meaning I can see it with my eyes on the ground
0:08:34 or just digging with the shovel?
0:08:36 – Yes, so copper, if it gets to surface,
0:08:39 it will oxidize and form minerals that look green
0:08:40 or look blue.
0:08:42 And so in the copper belt, for instance,
0:08:45 there would be hills that had green staining on them.
0:08:49 So it wasn’t difficult to discover those deposits,
0:08:51 but those have largely been found.
0:08:54 I haven’t stumbled into a green hill
0:08:56 that’s just waiting to be mined.
0:08:59 So now we’ve got to start looking underneath the surface
0:09:01 and we need a lot more tools, a lot more data.
0:09:05 And so it’s easier to expand your existing mine
0:09:08 than spend the money to try and make a discovery elsewhere.
0:09:12 – So who usually discovers these deposits to date?
0:09:16 – A lot of the discoveries are made by smaller companies
0:09:19 that are really focused on just making discoveries.
0:09:21 They’re willing to take more risks.
0:09:23 They are willing to test new technology.
0:09:26 They’re willing to look deeper or undercover
0:09:28 or in areas that people might have turned away
0:09:31 or had the incorrect geology approach
0:09:32 to the wrong model.
0:09:36 So that attitude to exploration that drives success.
0:09:38 – And maybe before we jump into
0:09:40 how technology is changing mining,
0:09:42 maybe can you share with us
0:09:45 how has mining exploration changed over the decades?
0:09:47 One of the stories I’ve always loved in mining
0:09:50 was there was a story of how someone made
0:09:52 a huge gold deposit discovery
0:09:55 based off the cover of a National Geographic Magazine.
0:09:56 Because just like you said,
0:09:58 it was being expressed on the surface.
0:10:02 So one, two decades ago, what did exploration look like then?
0:10:05 And then we can contrast that to today.
0:10:07 A lot of the work we’re doing today
0:10:09 is still the same groundwork.
0:10:10 We’re still moving onto ground.
0:10:13 We’re still mapping the geology,
0:10:15 taking samples of the soil.
0:10:17 That’s been done for a long time.
0:10:18 It’s very effective.
0:10:21 What has changed is the level of technology
0:10:24 in terms of the geophysical methods that are available.
0:10:28 So the ability to measure the properties of the rock.
0:10:29 Is it magnetic?
0:10:32 Is one rock denser than another rock?
0:10:34 We can measure that through what we call gravity measurements.
0:10:36 We can use seismic surveys
0:10:41 to try and identify the shapes of deposits deep underground.
0:10:43 And those technologies have advanced dramatically.
0:10:45 They’ve become airborne.
0:10:46 So we can fly over deposits.
0:10:49 So we don’t need to build roads and bridges.
0:10:51 We can now access the ground easy.
0:10:53 They’ve become better in their resolution,
0:10:55 the quality of the data.
0:10:59 But with that has come significant volumes of data.
0:11:01 And so the ability to process
0:11:05 and extract the value out of that data is not a challenge.
0:11:09 – Tom, tell us how is AI being used for exploration
0:11:12 in a way that wasn’t true a decade ago?
0:11:15 – Yeah, I think the types of data that George described
0:11:17 are what we need to squeeze
0:11:19 in order to have the insights
0:11:22 that guide our exploration programs.
0:11:24 So we might start at the continent scale,
0:11:27 would say satellite data across all of Australia.
0:11:30 And we want to be able to find particular rock types
0:11:32 that might be indicative at the surface
0:11:34 of deposits that are very deeper.
0:11:37 So we can use image recognition
0:11:40 and classification algorithms developed
0:11:41 for other applications.
0:11:45 We can adapt those to then apply to our specific instance
0:11:47 of adding to insights that might lead
0:11:50 to finding these underground deposits.
0:11:53 But just one data source is not nearly enough.
0:11:56 We would also take magnetic data
0:11:57 across the entire continent.
0:12:00 We will take whatever kinds of geochemical data,
0:12:03 government data sets such as geologists
0:12:04 who’ve walked around, picked up a rock
0:12:07 and sent that rock in for chemical analysis.
0:12:10 They’re databases that are available
0:12:12 but extremely poorly structured
0:12:14 with multiple analysis methods
0:12:17 that we need to be able to clean up
0:12:19 so that we can actually feed it into our algorithms
0:12:22 and serve it up to our geoscientists.
0:12:25 We also take geostructural data.
0:12:27 So what is the topography?
0:12:29 Where are the mountains and what slopes
0:12:31 are the different rock layers coming in?
0:12:33 We can take all that data and use that
0:12:37 to inform the local interpretation of the geology.
0:12:41 So that’s going from the continent scale
0:12:43 to what we call the camp scale
0:12:46 or roughly 10 kilometers by 10 kilometers
0:12:49 as we narrow down the search for one of these deposits.
0:12:52 And then we need a whole different set of algorithms
0:12:54 in order to go from that 10 kilometer scale
0:12:57 to where do we actually drill
0:12:59 to try to answer the question of what’s underground?
0:13:03 You really can’t tell what’s underground until you drill.
0:13:06 And the single drill hole can cost up to a million dollars.
0:13:09 So we want to be able to optimize
0:13:10 where we place that drill hole.
0:13:13 So those are some of the types of algorithms
0:13:17 that we’ve built to give superpowers to our geologists.
0:13:19 – Quality of the data is so important.
0:13:21 And you mentioned a lot of the data
0:13:24 that Cobalt uses previously was very unstructured.
0:13:28 Give us a taste of just how difficult it was
0:13:31 to digitize the data to begin with
0:13:34 and then put it in a usable format.
0:13:38 – So very talented geologists have been walking
0:13:41 the surface of the earth for over a hundred years
0:13:43 and collecting sometimes handwritten records,
0:13:47 sometimes physical samples that are analyzed,
0:13:51 sometimes hand drawn maps, sometimes digital maps.
0:13:55 So they’re all this very diverse set of unstructured data.
0:13:58 And these are then sometimes in government databases,
0:14:01 either paper or digital archives,
0:14:04 or they are with potential joint venture partners
0:14:06 who might have huge pile of papers
0:14:09 that go to date back 50 or 100 years.
0:14:12 And there’s an incredible wealth of information there,
0:14:14 but information that hasn’t been exploited
0:14:16 ’cause it’s been locked up in these paper records.
0:14:18 So yeah, we’ve had teams out scanning
0:14:20 some of these paper records.
0:14:23 The job isn’t done once you have the records scanned.
0:14:26 We really want to get as much information
0:14:28 out of this raw data as we can.
0:14:31 And so for example, in Finland,
0:14:33 there’s a wealth of historic data, but it’s in Finnish.
0:14:37 And so we need to be able to both translate that.
0:14:40 And many of the words that people use to describe rocks
0:14:41 are highly specialized.
0:14:43 And so we need special translation modules
0:14:45 to make sure that those come through correctly
0:14:49 and then extract the structured data
0:14:50 out of these unstructured reports.
0:14:53 So for example, somebody may describe a certain rock type
0:14:55 and a certain grain size on that rock
0:14:57 and a certain chemistry of that rock.
0:14:59 And we would like to take that into a table
0:15:01 that our algorithms can then use.
0:15:03 – So it’s fascinating to me
0:15:05 that we basically have all these geoscientists
0:15:09 that then pair with these data scientists.
0:15:11 How do these teams work together
0:15:14 given that they’re coming from such different backgrounds?
0:15:15 – We’re actually fortunate
0:15:17 that we live in the world of technology
0:15:19 and somebody sitting in Silicon Valley
0:15:23 is working with someone sitting here in Central Africa
0:15:24 or in Australia.
0:15:27 And we’ve been very deliberate on making sure
0:15:29 that our data science software engineering
0:15:32 and our geoscience teams are highly collaborative
0:15:36 and are paired in being able to extract the rock
0:15:39 that we have drilled, upload it into the cloud
0:15:41 and anyone around the world within Cobalt
0:15:45 can look with the same precision at the same information
0:15:48 as if they were standing in front on the ground
0:15:50 here in South Central Africa.
0:15:53 We also have the abilities to actually look at the images
0:15:55 through all different models that we’re building
0:15:57 to analyze our core.
0:15:59 And that’s something that we build within the company
0:16:02 through the tech stack that we have.
0:16:04 – Even just the languages, the words,
0:16:07 how do you teach each other about your craft?
0:16:12 If you are a scientist, a biochemist, a geologist,
0:16:13 you’re bringing together people
0:16:15 of all different backgrounds.
0:16:17 How do you get everyone on the same page?
0:16:18 – One of the things we kind of do
0:16:20 with your first drawing, Cobalt,
0:16:23 we have some sort of nomenclature session
0:16:27 of what the basics mean in terms of geology and rocks,
0:16:30 in terms of also the tech and marrying those together,
0:16:33 like creating our own sort of internal glossary
0:16:35 that is easy for people from backgrounds
0:16:36 who have never been on a mine
0:16:38 or never been around exploration
0:16:41 and vice versa, people who have never understood
0:16:44 the different AI models that are used in the US.
0:16:46 And that starts quite early
0:16:48 when you’re inducting a new team member.
0:16:51 If you talk about some of the local staff we have
0:16:53 down to community members,
0:16:55 we’ve had indigenous words for copper,
0:16:56 like the word mukuba,
0:16:58 and we help teach our North America
0:16:59 and our South African colleagues,
0:17:02 like this word mukuba is copper in our local language.
0:17:04 It’s literally one language in a way.
0:17:07 We also have a Zambian data scientist on the team
0:17:09 and they come and spend many, many weeks on site.
0:17:11 They get to go to a drill rig.
0:17:13 They get to suggest many new ideas
0:17:15 with drilling companies, drilling contractors
0:17:17 who’ve kind of worked in a certain way
0:17:18 over industry for many, many years.
0:17:19 So we look at the rig and say,
0:17:23 how can we be more efficient in extracting the score
0:17:25 and collecting information
0:17:29 in a manner that reduces the time to process the information?
0:17:31 So we’ve been fortunate to work with contractors
0:17:34 that allow us to trial our hardware
0:17:37 that Tom and his team build, ship it out to Zambia,
0:17:39 put it next to a driller,
0:17:42 do some basic training on how do we capture
0:17:44 imaging of the core, like 360 imaging of the core
0:17:46 as it’s coming out of the ground,
0:17:48 which is really revolutionary.
0:17:51 Whereas the standard was you have to wait a couple of days,
0:17:54 take an image, take it right, stitch it together.
0:17:56 And a data scientist working with a geoscientist
0:17:59 is also getting training from a geoscientist
0:18:02 on just the type of lithology that they see in the rock,
0:18:03 what we’re looking for
0:18:06 and how do we make better interpretations
0:18:09 and better predictions in a meaningful way
0:18:11 through either a mixture of AI
0:18:12 and HI through all the experience
0:18:14 of the brilliant geoscientists
0:18:16 who’ve worked in industry for many, many years.
0:18:18 – That’s a great point how Cobalt is innovating
0:18:21 on both the hardware and the software front.
0:18:24 – That’s right, to feed the algorithms
0:18:27 and the geoscientists who are going to make
0:18:31 the next generation of discoveries of these deposits,
0:18:34 we need to have as many data types as we can
0:18:38 and put all of that data in front of them.
0:18:40 So we found some opportunities
0:18:42 where we could push the state of the art
0:18:45 to be able to collect more flavors of data more easily.
0:18:49 And so one of those is with hyperspectral imaging.
0:18:52 And hyperspectral just means many, many different colors
0:18:55 all the way from the visible through the infrared.
0:18:58 And what’s really exciting about hyperspectral imagery
0:18:59 is that in the infrared,
0:19:03 you can actually measure the absorption
0:19:05 of different molecules.
0:19:09 And so you can actually read out chemistry using light.
0:19:13 And we implemented that method.
0:19:15 So hyperspectral imagery on an airborne system
0:19:19 so that we can fly over the Cobalt claims,
0:19:22 take terabytes of data with a light aircraft
0:19:24 and then come back and process that data
0:19:27 with very high spatial and spectral resolution,
0:19:31 add together the ground truth that geologists
0:19:33 who’ve walked the ground and picked up rock samples
0:19:36 and then scan those with infrared spectroscopy
0:19:38 combined with this airborne imaging
0:19:41 to really build a system that hasn’t been built
0:19:43 in mineral exploration before
0:19:45 to be able to survey thousands of square kilometers
0:19:48 and automatically interpret exactly
0:19:49 what rocks are on the ground.
0:19:52 So compared to a hand drawn geologic map
0:19:54 that has a rich history,
0:19:57 we can actually make a data driven geologic map.
0:19:58 – That’s great.
0:19:59 And George, given that you’ve worked
0:20:03 at more traditional exploration of mining companies before,
0:20:06 maybe share some examples of how this technology
0:20:08 or the data has surprised you.
0:20:10 – Yeah, there’ve been a number of examples.
0:20:13 Some of the countries have big data sets
0:20:15 and these are data sets that have been accumulated
0:20:17 over the years from explorers
0:20:18 and different parts of the country.
0:20:21 Each of those explorers was looking for something specific.
0:20:23 And so they didn’t assay necessarily for everything.
0:20:26 They didn’t measure every element in every sample.
0:20:29 They were just targeting a few.
0:20:32 And now we might be looking for a very specific element
0:20:34 and it’s only available in a small portion
0:20:35 of that data set.
0:20:37 Normally you think, oh, it’s such a pity
0:20:39 that we have this huge data set
0:20:41 but we can only use a small part of it.
0:20:44 Whereas the data scientists will say, well, that’s okay.
0:20:46 That element that we’re interested in
0:20:49 has relationships to other elements.
0:20:53 And we can have a really good idea
0:20:55 of what the grade of that element would be
0:20:56 given the grade of all the other elements
0:20:59 that we know at each of the sample points.
0:21:01 So we can test it by taking out the examples
0:21:03 where we do have that grade.
0:21:05 We can then estimate what that grade would be
0:21:06 and we compare it.
0:21:08 And that’s remarkably close
0:21:09 because we’re not just comparing it
0:21:11 to one element or two elements.
0:21:14 We’ll be looking at relationships from numerous elements.
0:21:16 And so suddenly there’s huge data sets
0:21:18 that look like at a trunk.
0:21:20 We’re actually able to use the full value
0:21:22 and the spread of that data.
0:21:24 And that allows us to move into areas
0:21:25 where other people wouldn’t be.
0:21:26 That’s one way.
0:21:30 Another way is Tom spoke about data that you digitize, maps.
0:21:33 If I scan a map, I can look at it on a computer screen,
0:21:36 but I have to look at it really carefully.
0:21:38 Now I can just look for a search term
0:21:42 and the 12 maps that have that term will pop up
0:21:44 and it’ll show me where that term is on the map,
0:21:45 straight away.
0:21:46 And then I’m able to interrogate,
0:21:49 maybe it’s the name of the drill hole
0:21:52 or it’s a certain element that I’m looking for
0:21:55 in a report on a series of maps.
0:21:57 And suddenly all of this information
0:21:59 is just so much quicker to interrogate.
0:22:02 I can spend my time applying my geological training
0:22:05 and my experience rather than spend my time
0:22:06 just opening and closing things.
0:22:09 So that’s a huge advantage to be able to advance
0:22:11 the search for the new discoveries.
0:22:13 – With all of this data,
0:22:16 how do you guys know what to prioritize?
0:22:17 How do you decide what kind of information
0:22:18 is more important?
0:22:22 And then how does that guide the actual work on the field?
0:22:26 – In a world that is rich in data that becomes the challenge
0:22:29 is how do you know what to actually focus on?
0:22:32 Some of that will come from our experience
0:22:34 with working with data from other projects.
0:22:37 So the geoscientists will be interacting
0:22:39 with the data scientists saying in this environment,
0:22:42 we know that these factors are really critical.
0:22:44 And this is where the collaboration
0:22:46 between the geoscientists and the data scientists
0:22:50 becomes so important that it’s not two separate entities.
0:22:53 It’s very collaborative and specific.
0:22:56 And so there’s no off-the-shelf option.
0:22:58 We’re not developing a tool that we consult
0:23:01 when exploration company that will help them discover
0:23:02 in any environment.
0:23:04 Everything we’re doing is tailored
0:23:06 to the areas that we’re working.
0:23:09 – And one key aspect is also modeling the uncertainty
0:23:10 of what’s underground.
0:23:12 And there’s incredible uncertainty
0:23:14 of what’s just under the surface.
0:23:17 And so by being able to map out the regions
0:23:20 of high uncertainty or low uncertainty,
0:23:22 that can also allow us to optimize
0:23:24 where we collect the next data point.
0:23:27 And so we can start with these publicly available
0:23:30 or joint venture style data sets.
0:23:32 But ultimately we have to go to the field
0:23:34 and collect our own data.
0:23:36 – And I guess that uncertainty also brings us
0:23:38 to the question of with all of this technology,
0:23:42 how has it improved our accuracy when we do drill?
0:23:45 Is there a reduction that you see already
0:23:47 in the number of drills it’s taking
0:23:49 for us to get more information?
0:23:52 – So we want to basically quantify uncertainty
0:23:55 and by being able to drill in an area
0:23:59 where we’ll maximize the amount of information we get
0:24:01 that will inform both the geoscience
0:24:04 and data science team of what is happening
0:24:07 within the underlying rocks is very different
0:24:11 than I’ll say the traditional way of from top’s point,
0:24:14 you have a camp of a 10 by 10 square kilometer
0:24:17 and you place a grid drill and hope for the best.
0:24:21 We are very targeted on the areas
0:24:24 that both the geoscience and data science team
0:24:26 want to drill a hole to the angle the hole
0:24:29 should be drilled at to what we’ll call the pierce point
0:24:33 of where we think the resource may lie.
0:24:36 And within a 24 month period,
0:24:39 the level of precision has encouraged a lot of confidence
0:24:42 particularly on our project in Zambia,
0:24:43 we’ve moved up to 10 rigs.
0:24:46 And obviously with each rig,
0:24:48 we know there’s a level of uncertainty attached to it
0:24:51 but we want to quantify that as much as possible.
0:24:54 – So as you talk about drilling for information then,
0:24:56 does that mean sometimes we’re drilling
0:24:58 not just to look for the resource
0:25:00 but we’re drilling whatever hole will give us
0:25:03 the most information that will either confirm
0:25:06 or deny a bunch of hypotheses?
0:25:10 – Yes, and we want to basically falsify the hypotheses quickly
0:25:12 because you could be drilling into perpetuity
0:25:15 but to somebody somewhere that’s a cost.
0:25:19 So even if you drill and you falsify a particular hypothesis,
0:25:21 you do not find what you’re looking for
0:25:24 or you affirm what you think was happening in the system,
0:25:26 that is a lot of information
0:25:29 that normally in traditional mining exploration,
0:25:32 someone’s like, oh, we didn’t find what we’re looking for
0:25:35 but we learned something from that particular hole.
0:25:38 And that learning takes us back to saying,
0:25:41 all right, scratch off hypothesis number seven,
0:25:43 let’s come up with a new hypothesis.
0:25:46 It’s basically data-driven decision-making.
0:25:50 – Yeah, a lot of the decisions around how to explore
0:25:53 would be driven by what is the sort of precedent
0:25:54 in this environment?
0:25:56 What grid size would we be drilling?
0:25:58 What did the neighbors draw?
0:26:01 What’s been the traditional drill spacing?
0:26:04 Some of it might be driven by perceived regulatory requirements
0:26:06 to reach a certain density of data.
0:26:09 The interesting thing is that as geologists,
0:26:13 we know that we have a limited understanding
0:26:14 of what’s deep underground.
0:26:16 And when you talk to one of the famous things
0:26:19 about geologists is that they get a room of geologists
0:26:23 together, you’ll get n plus one number of opinions.
0:26:24 You know, that’s where the joke is,
0:26:26 there’s four geologists, there’s five opinions.
0:26:28 And inherently we know that,
0:26:31 but that isn’t built into the decision-making process.
0:26:34 We aren’t building models that account
0:26:36 for all the different possibilities
0:26:37 that we know could exist.
0:26:41 And so typically we will anchor on one model
0:26:44 and our drilling and our sampling and our exploration
0:26:47 will focus on that model and we’ll keep testing that model
0:26:49 until either eventually we realize it’s not working
0:26:52 or we hope it’s going to hold.
0:26:55 Whereas at Cobalt, we are happy to hold multiple models
0:26:57 at the same time.
0:27:00 And we will test each of those models simultaneously
0:27:02 ’cause you’re testing the same space.
0:27:05 But we design the drill holes to maximize
0:27:08 how many of these models can we effectively test
0:27:10 with each of these holes that we’re drilling?
0:27:13 Where are the specific areas that are gonna solve
0:27:14 some of the problems that we need to know
0:27:18 about which hypothesis is valid and which one isn’t?
0:27:19 And so actually building that in
0:27:22 and being able to show the uncertainty
0:27:24 and how we are reducing that uncertainty
0:27:27 as we continue to drill is a really key focus
0:27:30 and it’s something that’s very unusual in the exploration.
0:27:34 – One example of applying our AI tools
0:27:37 to lithium exploration is actually in Canada
0:27:41 where we knew that the right geologic processes
0:27:43 had happened within this belt of granite.
0:27:46 We knew that this is the right flavor of granite
0:27:48 that might host a lithium deposit,
0:27:52 but we didn’t know where in hundreds of square kilometers
0:27:57 with no road access where you had to helicopter in a crew
0:28:01 to go look at the ground where there might be signs
0:28:02 of lithium.
0:28:07 And in order to investigate this very large area,
0:28:10 the traditional way would be to drop off a field team
0:28:13 and maybe pick them up at the end of the summer
0:28:16 and they will have walked across as much terrain
0:28:19 as they could walk across and found whatever lithium
0:28:21 they could find at the surface.
0:28:24 That’s not fast enough or cheap enough.
0:28:26 And so we knew we could do better than that
0:28:28 standard practice.
0:28:31 And so what we did was start with satellite data
0:28:36 and start with reports across a significant fraction of Quebec
0:28:40 about where lithium had been found previously.
0:28:43 And then we matched the satellite imagery
0:28:48 with those records of lithium-bearing rocks
0:28:52 and we’re then able to predict where in our claims
0:28:54 there might be lithium-bearing rocks.
0:28:56 So then we sent our people out in the field
0:29:00 to either confirm or deny the existence of lithium
0:29:02 at the places where our machine learning models were
0:29:05 predicting there might be lithium.
0:29:09 And they reported back immediately via satellite link
0:29:13 and said, guys, you just sent us to a vast field of lichen.
0:29:14 There aren’t even rocks here.
0:29:16 Everything’s covered in lichen.
0:29:18 And that’s not what we were looking for.
0:29:22 But that led us update our models overnight
0:29:26 so we could rerun models across 1,000 square kilometers
0:29:29 overnight with this new ground truth of what our people had
0:29:30 actually found on the ground.
0:29:32 And so with the addition of both the false positives
0:29:35 and the true positives, we were able to improve the accuracy
0:29:38 of our model prediction by over an order of magnitude.
0:29:40 And then in the following several weeks
0:29:43 that we had this helicopter supported field program,
0:29:45 we were actually able to find spodgamine, which
0:29:48 is the lithium-bearing mineral that we were looking for.
0:29:50 So that’s an example of compressing
0:29:54 what might have taken years of expensive remote field work
0:29:57 into the period of a week or so by being able to iterate
0:30:00 and update these machine learning models.
0:30:03 If I compare what is happening in the oil and gas space
0:30:06 and how they marry traditional methods and technology,
0:30:10 how does that compare with the mining space?
0:30:13 Is mining ahead, behind of oil and gas,
0:30:15 or are there shared learnings?
0:30:17 The technology used in the mining industry
0:30:20 is dramatically behind what we’ve
0:30:22 used to search for oil and gas.
0:30:26 And part of our job is to take these amazing technologies that
0:30:29 have been developed to find fossil fuels
0:30:32 and then adjust them so that perhaps they’re
0:30:37 smaller and cheaper so they’re applicable to discovering
0:30:39 metals rather than oil and gas.
0:30:42 And so examples of that are directional drilling,
0:30:46 where in oil and gas now you can precisely control
0:30:49 the trajectory of exactly where you want to drill.
0:30:52 This is new for the mining industry.
0:30:55 We need to make this technology smaller
0:30:59 so that it’s appropriate for the kinds of brakes that we use.
0:31:01 Another example is geophysical techniques
0:31:05 that were initially invented for finding oil and gas.
0:31:09 We can adjust those to whether that’s seismic imaging,
0:31:12 whether that’s electromagnetic imaging.
0:31:13 We can adjust those so that they’re
0:31:16 suitable for finding the kinds of more bodies we’re looking for.
0:31:19 So really, it’s taking tools that have been developed
0:31:21 by other industries, such as oil and gas,
0:31:23 or in the case of these artificial intelligence
0:31:26 algorithms, tools that have been developed for image
0:31:28 processing, we can take those and then apply them
0:31:30 to the kinds of complex problems that we face.
0:31:34 Oftentimes, we’re talking about copper, nickel, lithium.
0:31:38 Does the approach work for other metals underground?
0:31:42 And how do you guys think about the expanse of what
0:31:45 this technology can be applied towards?
0:31:48 Yeah, it definitely has applicability
0:31:49 across any of the search spaces.
0:31:51 The processes that are being developed
0:31:54 are solving data problems.
0:31:58 They are looking for controls on mineralization.
0:32:00 And you can adjust those as you’re
0:32:01 looking at a different commodity.
0:32:04 So what’s controlling a nickel deposit?
0:32:06 It’s very different to what’s controlling a lithium deposit.
0:32:11 This is the beauty of not approaching a commercial software
0:32:13 vendor to try and solve your problems.
0:32:16 You still have the data science team and the geoscience team
0:32:18 working so closely together to say,
0:32:21 these are the problems in this specific environment.
0:32:24 Even if you’re thinking of nickel exploration,
0:32:26 nickel deposits, all of them are different.
0:32:28 Yeah, the technology really is a toolbox
0:32:31 that we can apply to different exploration programs
0:32:34 around the world in multiple mineral commodities.
0:32:36 And so the same exact tool that you’d
0:32:39 use in Australia to find lithium is probably
0:32:43 different from the one that you’d use to find copper in the Arctic.
0:32:45 But they have many parallels.
0:32:46 And when I started at Cobold, I was
0:32:48 new to the mineral exploration industry.
0:32:51 And I was thinking that there’d be an incredibly sophisticated
0:32:54 way that society finds the materials
0:32:56 that we use on a day-to-day basis.
0:32:59 But actually, the tools in mineral exploration
0:33:01 are shockingly primitive.
0:33:03 And we can rapidly move beyond the state of the art
0:33:05 in mineral exploration.
0:33:09 I think one anecdote about the state of the industry
0:33:14 is just how much it relies on looking at rocks,
0:33:17 where a good geologist can look at a rock
0:33:20 and tell you from the context of what’s around that rock,
0:33:22 from the grain size, from the color,
0:33:25 from the mineralogy of that rock,
0:33:28 can tell you just so much about what
0:33:31 has happened over the last billion years at that location.
0:33:34 And yet that’s all done with the human eye.
0:33:37 And so it’s said that the best geologist is the one that’s
0:33:38 seen the most rocks.
0:33:41 Well, what if we make a system that
0:33:46 can look at more rocks than any human has ever looked at before
0:33:50 by just analyzing the imagery of the core that we’ve
0:33:55 drilled at Mingonba, we now have nearly 100 kilometers
0:33:58 of rock core that we’ve pulled out of the ground?
0:34:00 That would take years for a person
0:34:03 to look at every little part of that core.
0:34:07 And yet this is perfectly suited to machine learning
0:34:10 algorithms that can look at millions of images
0:34:11 in the blink of an eye.
0:34:15 And so being able to take the real knowledge that
0:34:19 is embedded in hundreds of years of geoscience
0:34:25 and then apply new emerging tools to help analyze and extract
0:34:27 real information from those techniques,
0:34:29 I think it’s just incredibly promising.
0:34:30 That’s amazing.
0:34:32 Lots of times when I’m reading in the headlines now,
0:34:35 I see this phrase, critical minerals.
0:34:38 Do you guys have a sense of which critical minerals
0:34:41 are important to national interest, for example?
0:34:44 Or how would you define what is a critical mineral?
0:34:46 Yeah, if you look at all the lists of critical minerals,
0:34:49 almost every element on the periodic table
0:34:50 is in one of those lists.
0:34:53 And so to me, what a critical mineral means
0:34:57 is something that’s important for our economy
0:35:00 and our defense industrial base.
0:35:04 And if you look at where our economy is trending,
0:35:07 our economy is trending towards electrification,
0:35:09 towards electrification of everything.
0:35:11 And so the most important critical minerals
0:35:16 are those that allow us to electrify the economy.
0:35:18 And the most important materials for electrifying
0:35:22 the economy really come down to copper and lithium
0:35:23 being the main ones.
0:35:25 And then there’s a long tail of other critical minerals
0:35:27 beyond that.
0:35:29 Yeah, I think there’s a lot of minerals
0:35:32 that can perform a similar role, some just more efficiently
0:35:33 than others.
0:35:38 And as supply concerns or prices dictates,
0:35:40 some can be substituted out.
0:35:43 But some elements are just so fantastic at what they do
0:35:47 that they are the critical minerals for electrification.
0:35:51 Copper is just so good at transmitting electricity.
0:35:54 Lithium is so good in batteries.
0:35:58 And it’s really hard to see how you can, on a large scale,
0:36:00 remove such elements.
0:36:01 Other considerations are elements
0:36:05 where the supply is concentrated in very small areas.
0:36:08 Things that on local or shorter timescales
0:36:10 might become really critical.
0:36:11 Yeah, tell me more about those.
0:36:13 Are they concentrated in specific continents?
0:36:16 Or is it even at a country level?
0:36:18 Rare earth elements haven’t always
0:36:22 been top of the exploration ladder in terms of what people
0:36:24 are really interested in looking for.
0:36:26 And so deposits will go through phases
0:36:29 where you will discover deposits, define them,
0:36:30 and then prices will drop.
0:36:32 And those deposits are known about.
0:36:33 But no one’s interested in them.
0:36:35 And then there’ll be a price shock.
0:36:36 And they will not jump back into them.
0:36:39 And so rare earth deposits are widely spread.
0:36:43 They are well represented in the geological record.
0:36:45 But they haven’t been widely explored for.
0:36:47 So lithium is actually a really good example
0:36:49 that for many, many, many years, people
0:36:51 were not that interested in lithium.
0:36:53 It was used in very niche applications.
0:36:55 And it’s only since batteries have rarely
0:36:58 become so critical that suddenly people like, wait a minute,
0:37:01 where’s the historic database on lithium deposits for?
0:37:04 It’s small because people haven’t been focused on it.
0:37:06 And I think some of the rare earth elements
0:37:09 fall into those categories where for a long time they’ve
0:37:15 been quite niche and now are becoming a lot more important
0:37:17 to technologies going forward.
0:37:18 So traditionally, as a geologist,
0:37:21 when you’re thinking about what to go explore for before it
0:37:25 might be in terms of market size or financially,
0:37:27 does it make sense to go explore for this,
0:37:29 given where the prices are for those metals?
0:37:31 And now it seems like there’s another driver, which
0:37:33 could be national interest.
0:37:36 Are there certain metals that even if the numbers don’t
0:37:40 pan out today, there might be other incentives
0:37:43 to build up more deposits for these other metals?
0:37:44 Yes.
0:37:46 And I think in the mining industry,
0:37:52 there hasn’t been a large transition of sustained demand
0:37:53 for certain elements.
0:37:56 And that’s happening and has been in progress now
0:37:58 for a few years and will continue.
0:38:01 And that’s quite a change from what’s happened historically.
0:38:03 We’re always going to need iron ore.
0:38:04 We’re always going to need copper.
0:38:08 But in the last few years, other elements
0:38:10 have suddenly become a lot more critical than they’ve
0:38:12 been in the past.
0:38:14 And it only looks like the demand is
0:38:16 going to grow for those elements.
0:38:19 And there’s a fundamental shift and a lot of companies
0:38:22 are refocusing efforts into these elements that
0:38:24 haven’t had a lot of love from the exploration world
0:38:26 for a long time.
0:38:29 It’s more complicated than just a supply and demand
0:38:30 on an open market.
0:38:33 There is a country competition where some of those elements
0:38:36 might not be available.
0:38:40 And so that also becomes a consideration, definitely.
0:38:40 Yeah.
0:38:44 And that country competition and the impact on countries,
0:38:48 what is Zambia’s response to this new cobalt approach?
0:38:51 And how has it impacted, also, does Africa as a whole?
0:38:54 It’s been absolutely positive.
0:38:57 Definitely a breath of fresh air for all of us
0:38:59 who’ve been in the sector here.
0:39:02 And generally, as a continent, there
0:39:05 is this big drive because we have this huge youthful population
0:39:07 to drive industrialization.
0:39:09 This is what many African governments want.
0:39:12 And an investment like cobalt into Zambia
0:39:16 is stimulating opportunities, knowing that beyond what we do
0:39:18 in exploration and eventually mining,
0:39:21 we’re actually driving the development of the youngest
0:39:24 population on the continent and a lot of other trade
0:39:27 and business opportunities, even for US enterprises,
0:39:30 looking to do business in Zambia and in different parts
0:39:31 of Africa.
0:39:36 So it’s a seed that basically is getting in the door,
0:39:38 meeting the right officials, and stating the intention
0:39:41 very transparently, very clearly,
0:39:43 and operating within the bounds of the law
0:39:45 that we know globally.
0:39:47 And that’s quite exciting for the country,
0:39:50 but it’s also getting the neighbors waking up and saying,
0:39:52 wait a minute, if we integrate more,
0:39:54 there’s this big project in Zambia.
0:39:55 How do we participate in it?
0:39:57 How do we use the Lobito Corridor,
0:40:01 which we know will be a conduit for some of these minerals
0:40:04 that will help us drive electrifying the world?
0:40:07 You’re talking about a continent with a billion people,
0:40:09 or these billion people in coming decades
0:40:12 who need all sorts of consumer items and electronics,
0:40:14 and the US gets the firsthand look into that
0:40:18 by having access into Africa through the Corridor.
0:40:21 Just along the corridor, the population of the three countries
0:40:24 combined is close to the US population as a whole.
0:40:27 And growing, even as we build these mines,
0:40:30 how we mine, when we mine,
0:40:31 and how much material we impact,
0:40:34 and even how we’re able to look at the environment
0:40:38 through the mining process is going to rapidly transform
0:40:40 basically on everything that we’re doing
0:40:41 with the information we have
0:40:43 and how we can analyze it statistically.
0:40:45 So mines of the future are also going
0:40:46 to be fundamentally different.
0:40:48 I know we talked about exploration,
0:40:50 but as earlier said by Connie,
0:40:53 the image of mining has been poor in past decades,
0:40:56 and something we want to do is change the image of mining,
0:40:59 and that change will come to how we do a lot more things
0:41:01 with more prediction, more precision,
0:41:04 even as we build out many, many projects,
0:41:06 and that basically becomes institutionalized
0:41:08 in the industry.
0:41:09 – I think from my perspective,
0:41:13 the industry is perceived as quite an established,
0:41:18 mature industry, and from a geology perspective,
0:41:19 a lot of the easy deposits of perception
0:41:22 as they’ve been found, and that’s only going to get harder,
0:41:25 and it doesn’t sound like a good marketing touch.
0:41:29 But the reality is actually this time now
0:41:33 is an amazing time to be in exploration.
0:41:36 The ability to take new technologies
0:41:39 and apply them to solve the problems,
0:41:42 and knowing that you’re absolutely critical
0:41:45 to ensuring that we can solve the problems
0:41:47 that we need to solve globally.
0:41:49 We don’t do our job properly.
0:41:52 We’re going to slow down the ability to solve problems.
0:41:55 – Yeah, that sort of focus has been missing,
0:41:57 I think, from the industry for a long time,
0:41:59 and it’s awesome to be part of that now.
0:42:04 – All right, that is all for today.
0:42:06 If you did make it this far, first of all, thank you.
0:42:08 We put a lot of thought into each of these episodes,
0:42:11 whether it’s guests, the calendar tetris,
0:42:12 the cycles with our amazing editor, Tommy,
0:42:14 until the music is just right.
0:42:16 So if you like what we’ve put together,
0:42:21 consider dropping us a line at ratethespodcast.com/a16c.
0:42:23 And let us know what your favorite episode is.
0:42:26 It’ll make my day, and I’m sure Tommy’s too.
0:42:28 We’ll catch you on the flip side.
0:42:31 (upbeat music)
0:42:34 (upbeat music)
0:42:37 (upbeat music)
Critical materials like copper, lithium, and gallium have been mined for decades, but their role in core technologies, geopolitics, and the energy transition have come to a height in recent years.
In this episode, a16z partner Connie Chan discusses how technology is changing the game of identification and exploration, together with leading company KoBold and their VP of Geoscience, VP of Technology, and CEO of Africa.
Resources:
Learn more about KoBold Metals: https://www.koboldmetals.com/
Stay Updated:
Let us know what you think: https://ratethispodcast.com/a16z
Find a16z on Twitter: https://twitter.com/a16z
Find a16z on LinkedIn: https://www.linkedin.com/company/a16z
Subscribe on your favorite podcast app: https://a16z.simplecast.com/
Follow our host: https://twitter.com/stephsmithio
Please note that the content here is for informational purposes only; should NOT be taken as legal, business, tax, or investment advice or be used to evaluate any investment or security; and is not directed at any investors or potential investors in any a16z fund. a16z and its affiliates may maintain investments in the companies discussed. For more details please see a16z.com/disclosures.