Summary

Critical minerals are physical inputs to AI, grids, batteries, data-center infrastructure, vehicles, and defense systems. Turner Caldwell’s mass-flow call is emphatic: “We need a lot of aluminum. We need an insane amount of copper. We need more iron. We need more zinc,” while lithium production capacity must roughly 4× over the next 10 years if planned batteries are to be built.
A mineral project is a site-specific chain from sub-1% ore to high-purity metal, not a fungible factory template. Ore bodies change grade and impurity mix as mining proceeds—“the Earth is heterogeneous”—so every flowsheet is bespoke and needs flexibility. That makes recovery and adaptability economically decisive because every lost atom must be mined again.
The venture thesis is to own the mine-to-refinery operating system because selling point technology into incumbent miners has become a “death spiral.” Multi-billion-dollar plants resist changes that might create multimillion-dollar downtime, pilots can miss commercial builds that arrive perhaps once every five years, and operators distrust outsiders touching their “cash register.” Mariana, emerging with $85 million raised, is betting vertical integration can capture efficiencies that point-solution SaaS vendors cannot capture on their own—but it also imports the partner’s risk into its expanded risk profile.
Mariana’s software bet targets two large sources of wasted time: a roughly three-week construction-information lag and refinery-control problems with about 1,000 interacting variables. Capital Project OS would automate engineering and procurement workflows; Plant OS would use reinforcement learning to optimize refinery operations, including recovery, energy, and reagents, across 24–48-hour feedback loops. Caldwell wants to remove humans from many operating decisions.
China’s moat is skilled execution capacity as much as policy or capital. Caldwell saw 13,000 people mobilized at a Chinese-backed Indonesian nickel refinery during construction and commissioning; a U.S. project might struggle to field one-tenth as many. Indonesia now supplies “something like 70%” of global nickel, illustrating how labor depth and downstream buildout compound geopolitical leverage.
The portfolio call is countercyclical: diversify, navigate frothy markets, and build at commodity troughs. In the rare-earth discussion, Caldwell calls the market “a little bit of a frothy market,” while focusing on lithium and regarding copper demand and declining grades as “pretty hard to ignore.” The operating bet is that software-controlled circuits can process lower-grade copper without meaningful cost inflation.
U.S. supply security needs faster exploration, permitting, and demand support. Exploration over more than five acres on federal land can require BLM approval, while price floors or fixed-price offtakes—like the cited MP Materials arrangement—could unlock infrastructure capital that will not underwrite commodity volatility. Mariana’s mission is concrete but hedged: build 10 increasingly large projects in 10 years, expand overseas and perhaps underwater, and restore confidence that complex minerals infrastructure can be built “cost-effectively, time-effectively and responsibly.”

Deep dive

1. The bottleneck begins long before a battery or magnet exists

Caldwell starts with ubiquity: critical minerals underpin aerospace, defense, renewable energy, storage, AI and ordinary electronics—“your phone,” AirPods, screens and laptops. Mining and refining remain invisible despite determining whether many downstream products can be made.
The physical chain begins with exploration—“you’ve got to find the rocks”—then permitting, mine planning, extraction and separating ore from waste. Typical ore reaches the surface below 1% concentration, and definitely below 5% absent a world-class deposit, before mechanical, thermal or chemical concentration produces an intermediate product.
Refining can take something like a 10% concentrate to a 50% intermediate product and then to high-purity metal; specialty chemistry then makes sulfates or hydroxide salts, followed by engineered cathode or anode materials. Magnets similarly require the right rare-earth blend, casting, sintering and precision machining before reaching motors.
Torenberg’s question—wouldn’t each mine require its own setup?—gets a categorical answer: “It’s very bespoke.” Grade, impurities, target-metal concentration, and flotation or leaching behavior vary by asset and over time. A library of metallurgical unit operations is stitched into a site-specific flowsheet with substantial human input, making flexibility valuable as the ore body changes.

2. Following Tesla’s cost stack led Caldwell upstream

Caldwell’s path ran from Tesla factory design and construction into battery-cell manufacturing with Panasonic in Japan, then cathode manufacturing and refining. The pull was always “big things” and infrastructure: “If you want to have a big impact on the world, you have to build things at scale.”
Following cost revealed the chain in reverse. Factory equipment looked expensive until cells exposed the cost of their components; engineered materials then exposed the metals beneath them. That progression turned his question from how to manufacture batteries into why their metals cost so much.
His central incentive mismatch: manufacturing expects higher volume to lower unit cost through scale, whereas constrained mining supply treats greater demand as a reason for price to rise. “If you want more of it, it’s going to cost more”—the opposite of what a fast-scaling customer wants from suppliers.
The intellectual draw spans extremes: recover a target metal from 1% ore by solving chemistry at atom or micron scale, then deploy kilometer-scale infrastructure. Japanese battery manufacturing added another lesson: high-throughput precision improves through Kaizen and relentless iteration, not constant radical changes to individual operations.

3. Vertical integration is an assumption of risk, not a slogan

Tesla initially integrated because needed components did not exist: it was “do or die.” Integration continued when suppliers lacked incentives to invest, scale or innovate at Tesla’s desired pace, forcing development in-house to reach the required component specifications.
Caldwell’s caution is as important as the upside: “It takes a lot of guts.” Bringing an activity inside transfers the partner’s risk into the company’s expanded risk profile, so management must believe it is better positioned to control that risk.
Mining’s automation deficit partly reflects its capital cycle. Exploration and development consume capital, and a new mine can spend years removing waste or sinking a shaft before earning revenue; by then “the capital starts to get tired,” and automation gets cut if people can still drive the trucks and drills.
That fallback is disappearing as both trade and engineering labor pools contract, especially around remote assets. Mine sites can be calmer than cinematic images suggest; developing-country operations in Indonesia or Africa generally retain more visible human activity and less automation.

4. Incumbents’ operating model turns point technology into a trap

Caldwell says Freeport-McMoRan, Rio Tinto and BHP have digital-innovation arms but outsource much of this work; McKinsey and Palantir effectively act as consultants, and less than half of recommendations are often adopted. A valuable model should sometimes recommend counterintuitive moves beyond the local optimum humans have already found.
Trust is difficult when billions of dollars are operating reliably. Managers know that experimenting outside the current envelope might stop the plant, and the desirable culture—accepting a model’s counterintuitive recommendation—is particularly hard to build inside a large incumbent.
New $5 billion–$10 billion projects are commonly handed to an EPC contractor, even though the bespoke mine, processor and refinery should be treated as the product. Caldwell argues that EPCs have shifted somewhat from turnkey delivery toward selling hours, studies and reports, leaving the operator with less control over what it inherits.
Price-Wright’s description of “calcified” customers lands because fixing the status quo or achieving a step-change requires doing perhaps a thousand things, while each individual change carries plant-downside risk. Pilots are easy to approve, but commercial mines may appear only every five years; mistime the build cycle and a vendor waits another five. Caldwell calls the result a “death spiral.”

5. China’s execution depth and orphan assets create Mariana’s opening

Caldwell credits China’s top-down recognition of minerals, but says the neglected advantage is that “the talent pool is insane”—large, skilled and experienced. At a Chinese nickel operation in Indonesia, 13,000 people worked through construction and commissioning; mobilizing even one-tenth of that in the U.S. would be difficult.
The industry has meanwhile split exploration from development. Junior miners “don’t mine, they explore”: they define a resource, increase its value and hope to sell it to a major, sometimes while competing in Canada for capital with the cannabis industry.
Majors prefer multibillion-dollar deposits large enough to “underwrite their own inefficiency.” Smaller discoveries can enter an “orphan period” despite containing usable metal, because they cannot command an acquisition premium or independently justify development.
Mariana’s proposed wedge is to build and operate those supposedly subscale assets more efficiently, then scale toward the size of major-miner operations. McEntush defines the company as a “vertically integrated, software-first minerals project developer and operator,” owning detailed engineering, permitting, construction, commissioning and operations rather than selling tools into them.

6. Capital Project OS is designed to collapse construction latency

The organizational target is stark: begin with an exceptional team, then use LLM-driven engineering, procurement and construction workflows to let roughly 200 parent-company employees accomplish what might require 10,000 today. Much of the initial opportunity is eliminating lists, manual database transfers and avoidable project churn.
Large projects can carry a roughly three-week delay between field reality and the consolidated schedule used to make decisions. In the meantime, crews “stand in circles every morning,” ask what everyone will do that day, execute, and send brief reports that take far too long to become measured progress and revised priorities.
Capital Project OS aims to democratize live field data and run construction more like manufacturing. Caldwell’s connective insight is that “a mining project is a big civil construction project. It just never ends”; the same feedback-oriented software stack can address mining operations where underground equipment has historically been misplaced inside maze-like workings.

7. Plant OS treats a refinery as a robot with delayed feedback

Caldwell describes large refineries as “effectively big robots”: sensors and telemetry observe the plant, actuators control it, but humans still make many higher-level decisions atop basic temperature and pH set-point loops. The hard variable is feedstock, whose grade and impurity profile changes continuously with the ore body.
Operators currently blend feed to suppress that variability. Mariana wants to invert the logic with a “hyperdynamic” flexible circuit that responds to changing ore, optimizing the mine and refinery together while first reducing reagent and energy consumption.
Caldwell’s benchmark is DeepMind’s data-center thermal work after Google acquired it in 2016 or 2017. With weather, building load and roughly nine control variables, the system reportedly cut energy consumption 30%–40%; a refinery may present about 1,000 control variables.
Recovery is the biggest cost lever because “every atom that you lose” requires another atom to be mined. The circuit is an interconnected, high-latency web: downstream operations recycle reject streams upstream, and one change may take 24–48 hours to cascade. Chinese plants can commission in roughly six months; Western projects may take two to four years, with some still uncommissioned three or four years after construction.

8. The portfolio favors proven equipment and commodity troughs

Mariana will initially combine commercially demonstrated unit operations and seek uplift through superior integration and operation. That sequencing also fits project finance, which resists first-of-a-kind facilities; over time, Caldwell wants Mariana to become the customer that helps novel processing companies cross from pilots into commercial deployment.
Even basic industrial supply is a constraint: industrial tanks take unexpectedly long outside China, and a new pump in Australia might require 30 weeks versus a week—or three days—from a manufacturer in China. Asked whether Mariana would vertically integrate into equipment manufacturing, Torenberg jokes, “I don’t think so. I hope not”; Caldwell answers that the question depends on partner and supplier incentives.
McEntush frames mining as a massive market largely untouched by technology and argues that an end-to-end owner is needed to capture efficiency; Price-Wright adds that geopolitical urgency, improved technology and a hard-tech talent pool willing to work “in the middle of the desert” make the timing compelling. The shared alternative to point solutions is to control the full project and operating lifecycle.
Caldwell emphasizes mass flow over fashionable labels: aluminum, copper, iron and zinc require the largest absolute growth. Copper is the electrification and grid workhorse; aluminum is underestimated in transmission and is the most-consumed metal in defense applications, while zinc is needed to galvanize expanding steel infrastructure.

9. Lithium, copper and policy support define the near-term build

Lithium production capacity needs to grow roughly 4× over 10 years if projected batteries are built, though Caldwell hedges that demand forecasts materialize only when supply exists. Nickel serves high-temperature and corrosion-resistant alloys plus high-energy batteries; after Chinese-backed expansion, “something like 70%” of global supply now comes from Indonesia.
Rare earths remain essential but small by volume, with chemically intensive solvent-extraction circuits, relatively low recoveries and know-how concentrated in China. In that discussion, Caldwell calls the market “a little bit frothy”; diversification lets Mariana pick spots across cycles, with lithium a focus and copper’s declining grades a longer-duration opportunity.
On federal land, exploring more than roughly five acres can require a BLM plan of operations before expanding exploration over a larger area. Caldwell argues that U.S. rare-earth reserve estimates may reflect limited exploration rather than a fundamental lack of geological presence, while slow project approvals also repel talent that wants to see hard work become physical infrastructure.
His largest policy lever is demand support: fixed-price offtakes or floors such as the cited MP Materials arrangement could mobilize infrastructure investors seeking annuity-like returns. Government capital helps too, but can add federal permitting requirements to a state-land project. Mariana will start domestically, expand overseas and possibly underwater, pursuing “10 projects in 10 years” as proof the capability to build has been unlocked.

Random Ramblings July 2025 CME Group: The House Always Wins - [Business Breakdowns, EP.224]