Can AI-Driven Mines and Refineries Beat the US Critical Minerals Supply Shortage?

To to set the stage I think it's important to think about the the scale of the problem that's kind of presenting us, you know, much like oil and gas, mineral processing is is a depleted industry. So so what that means is that, you know, even if demand is staying steady, the world still needs to identify and bring new mines and refineries into production to make up for the fact that we're depleting the current resource base. Now you stack on that, you know, massive demand growth across AI infrastructure, electric vehicles, grid expansion, aerospace, defense, robotics, you name it. You kind of have this perfect storm of a supply crunch coming online, you know, and we anticipate needing, you know, over 500 new mines and refineries to be built over over the next 10 years. And and compounding on that problem is actually that, you know, these mines are getting harder, more expensive, and more complex to to develop. So, you know, we've gone and for most part mined most of the let's say good resources that we found today. And so what that means is that, you know, overall, if we take copper as an example, overall grades that we're finding these are decreasing quite significantly. You know, we used to be mining 1% copper and and now we're, you know, new mines are being developed at, you know, half or 0.6% copper. So that means there, you know, to make the same amount of copper you're having to move almost twice as much ore. The mineralogy is getting more complex, which makes the refineries more complex, and then the mining side is also getting more difficult as well. So, new discoveries are getting deeper, which means, you know, more advanced technology to get to the ore, and means longer times to develop, longer times to cash flow. That means temperature becomes more challenging in the underground, it means water infiltration becomes more challenging underground. So all of this is making it, you know, you know, it's getting harder to find these new assets, getting more expensive to develop them, and we need a lot more of them. And so hopefully that sets a little bit of stage for how I like the technical challenges that we're facing. You know, on top of that, you know, the Western mineral companies have kind of continued to I I'd like cede market position to to China in general. And you know, within that, you know, lands, you know, if you can look at this as as a glass half empty and say, you know, a large chunk of the US mining workforce is going to retire in the next 10 years, you know, we're slower and more expensive than developing projects as they are in China. But I think if you look at this as a glass half full, you know, it's it's essentially showing that that there is a roadmap to execute on this and and it is possible, right? Most of these major Chinese mining companies have all been started in the last, you know, 15, 10, 15, 20 years. Which really isn't isn't that long in the grand scheme of kind of the market that we're playing in. Um so it's showing that you know, new entrants to the market can can show up and really develop and scale their businesses and really you know, not surprisingly the way you need to do that is is to, you know, execute quickly and and execute cheaply. Um so there are some structural disadvantages that the US is facing in terms of labor availability, cost and schedule and you know, we here at at Mariana Minerals really think that, you know, the answer to all three of those problems is, you know, being autonomy first and and software first in the way that we develop and execute projects and and ultimately operate them. Um and we think that's like true true vertical integration, you know, there's a lot of companies out there selling spot solutions, but we think that that the value really comes into like building an overall integrated platform across project development, across, you know, mine operations and across refinery operations. So what does that mean in particular for for Mariana Minerals? You know, really we see ourselves as a project developer and owner and operator of the mine and mineral refining assets. There's actually been a lot of venture capital dollars that have funneled into the exploration game, you know, using big data analytics, using machine learning, using reinforcement learning to try and hone in on where we should be looking for or and and metals. And that's really exciting to see and then and very excited to see kind of what gets developed out of that. But really, you know, one out of maybe every thousand, you know, projects, exploration projects will ever actually turn into a mine. And so there's actually more of a bottleneck in the actual project development life cycle. So, you know, after a discovery gets made by a junior mining company, you know, their stock price will tend to climb, you know, maybe there's some excitement and really, you know, they're trying to get acquired because those teams are small, they're really out there trying to, you know, de-risk the geology of an asset, they're not there to build a project. Um and so, you know, eventually if they're not acquired, you know, the market tends to realize that, you know, the the likelihood of this project ever getting developed is quite low. Um and so, their their market cap tends to to drop pretty drastically. And these projects enter like what's called like the orphan period where there's some amount of geologic de-risking that's been done, uh but there's no one actually out there who the the market really believes is going to be able to develop and bring this mine into production. Um and that's where we see, you know, ourselves as Mariana Minerals coming into with a with a new value proposition to really accelerate the development and deployment of these projects to get those metals into production. You know, and and how do we do that? You know, really we see three three main verticals that we're working on, you know, building both our team and our software platform to address. And I'll spend a little bit of time talking about each of these in a bit more detail, but you know, Capital Project OS is is really about trying to, you know, more closely match the Chinese EPC execution model of doing, you know, process development, process engineering, detail engineering, and construction in as accelerated and compressed as a time frame as possible. You know, kind of eschewing a bit of the traditional stage-gate EPC project execution. And to do that obviously introduces some risk. And so, how can you use, you know, agentic engineering and advancements in LLMs to really create an integrated data system uh to to make that possible. Uh you know, MineOS is really about deploying autonomy and getting to as as, you know, near real-time mine planning as possible. Um and PlantOS is is really about trying to leverage advancements in reinforcement learning and machine learning to build optimized models of of the refinery and then eventually, you know, get real-time control systems to optimize for what we call like reward functions, which could be, you know, profit, yield, throughput, etc. Today, I'll spend a little bit of time talking about each of those. You know, you know, Capital Project OS, you know, to execute a large, you know, billion-dollar, multi-billion-dollar capital project, you know, requires, you know, quite literally an army of people, right? You have you have hundreds of engineers and designers working on different parts of the plant. You have thousands of construction technical professionals. You have to do thousands of engineering tasks. You have to release thousands of purchase orders. You have to manage, you know, inflow and receipt of, you know, steel, pipe, cable, equipment, you name it. You know, and today this is all, you know, really a highly manual process. You know, big EPC companies have built their business models around trying to manage these these data flows. But we see a ton of opportunity in building an integrated software system that's going to allow us to automate as much as possible so that, you know, instead of needing a dozen project controls professionals and a dozen project engineers to manage data information flows, you know, really you have a software platform and the ecosystem that allows you to feed the live status of project execution at any given point you know, when you look and evaluate potential changes, really see and and automate the impact of the cost and schedule impacts of making those changes all kind of in a unified software platform. On on the mine side, you know, there's you know, hundreds of activities going on every day in the mine from drilling and blasting, from mucking, from hauling, from, you know, moving material to the refinery. You know, and really here, you know, there are spot solutions in terms of automation. And where we at Mariana see a lot of the value is really on sitting on top of those and kind of what we call the orchestration layer where how can you actually create an integrated planning platform that's, you know, constantly updating your your geologic model and your mine plan where every time you drill and you blast, you're learning something new about the ore that you didn't know before. How can you capture all of that data and get to real-time mine planning so that you're constantly making the best decision with the information you have available in front of you? And then the last one we have is what we call like plant OS, you know, where, you know, today when you're building a new mine or refining project, you know, typically you'll build a steady state heat material balance. You'll do your best to optimize that heat material balance. You know, but really mining's a highly dynamic and highly variable process. So, you know, your ore grade is constantly changing, your impurity profile is constantly changing. And that's from project to project, but even within a single mining pit, you're actually changing every time you go through the mining process. And so really you need your refinery if you want to truly optimize your yield and reduce your opex, you really have to be responding dynamically to what it is that you're actually mining at any given time. So, we spend a lot of time building first principles based and then also empirical based machine learning reinforcement learning models to build simulations of what we expect the plant to do based on the orders coming in and the different operating conditions you run the refinery at. And once you build a, you know, highly accurate model on how the refinery is going to operate, you know, you can start to put dollars and cents signs to all the inputs and levers to essentially create models that will, you know, predict your your OPEX and then you can then go in and optimize in real time to, you know, maximize yield throughput and and refining. And so that's what we call plant plant OS. You know, really we see those three as like critical to unlocking cost, schedule and and, you know, performance of our mines to be able to go compete compete with China and really go scale the the rest of the the western mineral production. And you know, we're we're going really rapidly right now. So I think our team is kind of we're we're 200 people primarily based in the US and then really have an active mine in in Southeast Utah where we're deploying a lot of the autonomous technology that I just talked about as well as what we're doing with the refinery out in Meeker, so that's where that's really kind of the the first demonstration of a plant OS. So kind of really exciting work going on across the board. [music]