Mariana Minerals

I. Right now

About forty miles southeast of Moab, a robot dog is on patrol.

It walks the catwalks of a copper plant that, until recently, had been more or less left for dead. Around it, autonomous drills bite into a hillside, haul trucks decide their own routes, and a software stack called MarianaOS reads it all back as telemetry - block model on one screen, plant chemistry on another, capital project burndown on a third. The mine is called Copper One. The company running it is Mariana Minerals. The point is not the dog. The point is that someone finally treated a mine like a system worth instrumenting.

The company is not yet two years old. It has roughly ninety people on payroll, a dual-coast footprint (San Francisco for software, Houston for engineering), and a Series A worth $85 million led by Andreessen Horowitz, with Breakthrough Energy Ventures and Khosla Ventures rolling forward from the seed. Lithium, copper, and nickel projects are in the pipeline, with three more in the customary stealth. The stated cadence: ten projects in ten years. Mining, an industry that measures projects in geological time, is being asked to keep up.

II. The problem they saw

The bottleneck of the AI age turned out to be a hole in the ground.

Every fashionable trend of the last five years - electrification, AI infrastructure, defense reshoring, the energy transition - terminates in the same shopping list. Copper for the wires. Lithium and nickel for the batteries. Cobalt, graphite, rare earths for everything else. The hyperscalers built data centers in 24 months. The mines that feed them, on average, take fifteen to twenty years to permit, finance, and commission. One of these timelines is going to have to give.

Critical minerals were also, awkwardly, an industry the United States had largely outsourced. China dominates refining for most of the strategic list. The Department of Energy started using the word "critical" in earnest. Capital noticed. Talent did not, particularly - mining had spent a decade telling young engineers to go elsewhere, and they obliged. The result, by 2024, was an industry with enormous demand, generous policy, and a labor base that had quietly forgotten how to build new mines fast.

This is the gap Mariana Minerals walked through. The premise is almost embarrassingly simple: the limiting factor is not geology, not metallurgy, not even capital. It is decision latency. Every drill program waits on a consultant. Every plant change waits on a binder of paperwork. Every project waits on the next. Replace the binders with software and the consultants with engineers who actually live on site, and the clock changes.

III. The founders' bet

A Tesla metals lead, an Affirm ML lead, and an energy CFO walked into a copper mine.

This sounds like the setup for a joke. It is, in fact, the cap table. Turner Caldwell ran Tesla's metals and minerals team, where he reportedly moved billions of dollars across lithium and battery facilities - which is the polite way of saying he learned, painfully and in public, what mining can do to a car company's schedule. Juan Lozano, the CTO, scaled ML systems at Affirm and Kite. Baker Tilney, the CFO, had built and sold an integrated energy company to Vitol. They had, between them, the three skills mining traditionally separates with continental drift: operations, software, and capital.

The thesis, as told to investors, was almost contrarian by 2024 standards: not exploration, not a junior miner staking new claims, not a chemistry breakthrough. Instead, take real assets - brownfield projects, mines that already exist - and run them differently. Buy the dirt, install the software, and let MarianaOS do what mining had always asked humans and clipboards to do. It is, in some ways, the most boring possible startup pitch. In others, it is the most aggressive: it bets that the moat in mining is operational excellence, and that operational excellence is, finally, a software problem.

IV. The product

MarianaOS is three operating systems pretending, politely, to be one.

The stack has three named layers, and the names are unembarrassed about their software heritage. MineOS handles the geology and the moving parts above ground: block modeling, drill scheduling, autonomous fleet dispatch. PlantOS handles the chemistry and the steel: process optimization in the refinery, sensor fusion across pumps and reactors, anomaly detection on equipment that would, in another life, have been minded by a man with a clipboard. CapitalProjectOS handles the thing mining traditionally fears most: capital project execution. Schedules. Procurement. The cost overruns nobody likes to discuss in the quarterly call.

Fig. 1 - The four-headed dog of the modern mine. (Yes, one of them is a literal dog.)

The robot dog is the part that makes for good copy, and Mariana knows it. At Copper One, Spot - Boston Dynamics' canine wedge into industry - has been integrated directly into the MarianaOS stack. It walks, it scans, it reports, and the data lands in the same backend that handles the drill telemetry. The mine site runs on a private 5G network, courtesy of Celona, because if you are going to insist that everything be telemetry, you had better not let the WiFi drop.

V. The proof

Numbers, with the usual asterisks.

The case for Mariana, ultimately, has to be made in dollars per ton. The early numbers, which the company tells anyone who will listen, are these: refining costs roughly 30% below conventional baselines, mining costs roughly 50% below. These are not audited financials. They are operator-reported figures from a single asset still ramping. They are also, if even directionally true, the kind of numbers that bend an industry's cost curve in ways that decade-long permitting reform never quite managed.

Fig. 2 - A chart that, if even half right, is the kind that ends careers and starts companies.

The customer base is the part Mariana does not need to invent. Refined copper, lithium and nickel sell into an audience that is, at the moment, panicking quietly: battery makers, EV manufacturers, grid storage developers, the hyperscalers building AI campuses that draw nuclear-plant levels of power, and a defense industrial base finally permitted to care about supply chains. The unsexy truth of critical minerals is that demand is not the question. Reliable, domestic, on-time supply is.

VI. The mission

"Building a better minerals company." That's it. That's the line.

Mariana's stated mission, to its credit, is short. Building a better minerals company. It is the sort of thing that sounds either humble or smug depending on the morning. The reason it works as a mission is that "better" has, at Mariana, a series of testable definitions: faster to commission, cheaper to operate, safer because there are fewer people in front of the rock, domestic enough to matter to policy, and visible enough to its operators that decisions can happen in hours instead of board meetings.

The longer version is implied by the cadence: ten projects in ten years. Mining has not done that in living memory. The industry's last great operating innovation arguably arrived with in-pit haul truck dispatch in the 1990s, and even that took a decade to spread. Mariana wants to spread MarianaOS across its own portfolio first, then - one suspects, though the company is careful not to say so on the record - license the stack to others.

VII. Why it matters tomorrow

The grid we are about to need does not, currently, exist.

Picture the next decade honestly. Data centers double in footprint. EV penetration crosses thresholds in markets that will not retreat. Defense procurement quietly normalizes the phrase "secure supply chain." Every one of those vectors lands on the same metals. The U.S. can either source them on Beijing's calendar or build the mines itself, and the second option requires somebody to run those mines without taking the standard fifteen years.

Mariana, of course, may not pull it off. The risk register is honest: brownfield mines come with the geology and the lawsuits of whoever ran them last. Permitting is improved but not solved. Commodity prices swing. Autonomy fails in dust. The company is not invincible. But it is, for now, the most interesting operator in an industry that has not had a genuinely new operator in a generation - and the cap table suggests serious people agree.

VIII. Back at Copper One

The dog is still walking.

It steps over a hose, scans a flange, and uploads a frame to MarianaOS. Somewhere a thousand miles away in San Francisco, an engineer notices a pressure trend, drags a slider, and the plant adjusts. The hillside is still a hillside; the rock is still rock. Mining, fundamentally, has not been replaced. It has, more quietly, been instrumented. That is the thing forty miles southeast of Moab. That is, in fact, what is new.