Summary

America’s AI constraint is increasingly “atoms and not algorithms”: minerals, power conversion, factories, and transmission must scale alongside models and chips. Turner Caldwell says the US is “50 years behind” China on critical-mineral supply, while Drew Baglino sees a century-old grid that remains largely mechanical, overbuilt, fragile, and dependent on overseas suppliers.
Permitting alone cannot close the minerals gap because execution after approval is still painfully slow. A project can take five years to build and another three to five years to reach operating rate; Mariana Minerals therefore targets design, construction, procurement, and ramp-up, aiming to build 10 projects in 10 years.
Both founders are embedding software in physical-asset operations rather than merely selling standalone software. Mariana uses agentic workflows and reinforcement learning across mines and refineries, including an effort to “remove humans from the loop,” while Heron Power uses silicon and software to replace “steel, oil, and copper” in grid-scale power conversion.
US manufacturing’s central disadvantage is supply-chain geography, not factory wages. Baglino estimates labor represents less than 10%—and potentially less than 5%—of cost of goods sold in a modern automated factory; China’s advantage is that the roughly 7,000 parts needed for a car can sit within a three-hour drive.
The Tesla operating model offers industrial startups a combination incumbents struggle to reproduce: techno-optimism, risk tolerance, mission, and persistence. Caldwell’s sharpest formulation was that Tesla keeps “barreling through the challenges as long as the outcome is worth it”; Baglino added the concentrating effect of knowing “whether or not the paycheck will clear” depends on execution.
The requested policy is durability and planning certainty. Caldwell wants the minerals equivalent of the tools built for oil and gas over the past 50 years; Baglino wants manufacturing zones, aligned jurisdictions, and a “federal highway trust fund for the grid” so private capital and suppliers can plan confidently.

Deep dive

1. AI’s industrial stack begins beneath the model layer

Erin Price-Wright’s premise: AI dominance and reindustrialization are “physical projects”—energy, mining, refining, manufacturing, and transmission. Every model, factory, and autonomous system ultimately requires materials and electricity delivered where and when needed.
Caldwell described Mariana Minerals as a software-first operator, not a SaaS vendor: roughly one-quarter of its staff are software or machine-learning engineers. Its Capital Project OS, Plant OS, and Mine OS support projects from development through autonomous operation; Mariana already operates a Southeast Utah copper mine producing high-purity copper materials and is building a Texas lithium refinery, with a goal of 10 projects in 10 years.
Baglino’s Heron Power builds solid-state transformers, applying decades of improvement in power semiconductors to the grid itself. The objective is to use “silicon and software to replace steel, oil, and copper” at data centers, solar and battery installations, and other large electricity loads.

2. America’s minerals gap persists after the license arrives

Caldwell’s blunt assessment was that the US is “50 years behind” specifically China, and a couple of decades behind globally. Faster permitting and more available project-level finance help, but neither fixes slow design, construction, and ramp-up.
His load-bearing timeline: even after receiving a license to operate, a minerals project can take five years to build and another three to five years to reach operating rate. Catching China therefore requires the US not merely to lower barriers, but to execute faster than China.
Baglino argued that US commercialization should follow US invention. Federal agencies, academia, and industry helped develop advanced power semiconductors, and the world’s leading silicon-carbide producer is US-based; failing to manufacture the applications domestically transfers the accumulated benefits abroad.

3. The grid needs modern controls and industrial density

Baglino left Tesla after watching rapid innovation at the grid’s edge—EVs, Supercharging, and Megapack—while “on the other side of the wire, there’s really been no change.” Mechanical systems developed over 100 years ago offer little monitoring or control, producing an overbuilt yet fragile system with too few, often overseas, suppliers.
His counterexample to inevitable US construction delay was Tesla’s Lathrop Megafactory: a JCPenney warehouse produced its first product 11 months later. The decisive variable was alignment—local authorities can use the process for a code-compliant project “to say no at every step” or help it reach yes.
On competitiveness, Baglino pushed back against labor-cost fatalism: modern automated factories put the labor differential below 10% of cost of goods sold, perhaps below 5%. China’s deeper advantage is co-location; everything needed to build a 7,000-part car can be within a three-hour drive.

4. Autonomy requires operating-team integration

Mariana integrates mining through refining because the handoff between them creates market inefficiencies. Its autonomy bet covers LLM-assisted engineering, procurement, construction resource balancing, and short-interval operational control—not a detached software layer sold to incumbents.
In refineries, reinforcement learning could continuously tune temperatures, flows, chemical additions, and residence times as heterogeneous feedstock changes. Caldwell said the goal is to “remove humans from the loop” because the US lacks a deep labor pool carrying the tacit knowledge needed to stabilize variable feedstock quickly.
Yet Caldwell’s qualification is crucial: technology penetration is paced by operating teams whose current stack may be “pen and paper and maybe 150 spreadsheets.” Mariana therefore seats software engineers beside operators under shared incentives, designing tools around frontline problems and the culture that must adopt them.

5. Tesla’s execution culture and a call for durable industrial policy

Caldwell distilled Tesla into three advantages: belief that archaic systems can be reinvented, appetite for risk that enables fast decisions, and refusal to give up on valuable outcomes through the challenges. Baglino added the “do or die” focus created when the company’s future success—and whether paychecks clear—rests on the team executing.
Both founders recruit from analogous industries rather than waiting for ready-made labor pools. For Tesla’s 4680 program, Baglino was responsible with his team for building a 50-gigawatt-hour battery facility in Texas and hired people from high-speed bottling and syringe manufacturing. Caldwell sees mining talent in oil and gas, while its optimization software resembles systems used for Uber rides, dog-walking apps, lending, and advertising.
Erin noted that Mariana’s initial projects should add over 500 construction jobs plus additional full-time roles in the next 18 months, while Heron’s first large factory is expected to create about 500 jobs.
Their actionable ask was durable policy. Caldwell wants to examine the toolkit used for oil and gas over the past 50 years and provide incentives that mobilize private capital behind minerals projects without fear that “the rug” will be pulled; Baglino proposed pre-aligned energy-manufacturing zones, jurisdictions getting to yes, and a “federal highway trust fund for the grid” so suppliers and financiers can plan, while improving resilience and lowering costs.

Is The Soft Landing Officially Dead?Inside Anthropic's $100 Billion Al Compute Commitment | CFO Krishna Rao