Most important take away

America’s AI ambitions are ultimately a physical infrastructure problem: the US grid runs on pre-WWII mechanical systems and critical mineral supply chains sit 50 years behind China. Winning the next industrial era requires applying the Tesla playbook (techno-optimism, risk tolerance, vertical integration, software-first operations) to mines, refineries, and grid-scale power electronics, paired with durable industrial policy and co-located domestic supply chains.

Summary

Erin Price-Wright of a16z interviews two Tesla alumni now building the physical layer beneath the AI economy: Turner Caldwell, co-founder/CEO of Mariana Minerals, and Drew Baglino, founder/CEO of Heron Power. The conversation reframes the AI race as a contest of atoms, not just algorithms — energy generation, grid transmission, critical mineral extraction, refining, and advanced manufacturing are the binding constraints.

Key themes:

• The physical stack is the bottleneck. Models, chips, and data centers all rest on electricity, transformers, copper, lithium, and silicon carbide. Innovation has flooded the edge of the grid (EVs, batteries, chips) while the grid itself still uses ~100-year-old mechanical, steel-oil-copper transformer designs from overseas suppliers — a fragile and strategically exposed position.
• Mariana Minerals is a vertically integrated, software-first mining and refining company (~25% software/ML engineers) running an operating copper mine in SE Utah and building a lithium refinery in Texas, with a goal of 10 projects in 10 years. Its three internal systems — Capital ProjectOS (agentic project lifecycle management), PlantOS (RL-based refinery control), and MineOS (short-interval autonomous mining control) — exist to compress 5-year build timelines and 3-5 year ramp times that put the US 50 years behind China.
• Heron Power builds solid-state transformers that use silicon carbide power semiconductors and software to replace steel, oil, and copper in grid power conversion at data centers, solar, and battery installations — capitalizing on US-led silicon carbide manufacturing that originated in DOE/Navy/academic partnerships.
• US labor cost is a red herring; in modern automated factories it’s under 5-10% of COGS. The real disadvantage is the absence of co-located supply chains. China clusters everything needed for a 7,000-part car within a 3-hour drive; the US must deliberately build similar manufacturing/energy zones.
• Workforce: re-industrialization requires creative sourcing from analog industries — Baglino hired battery talent from high-speed bottling and syringe plants; Caldwell taps oil and gas for mining/refining talent, plus software engineers from ride-share, ad-tech, and lending where the optimization math is nearly identical.

Business strategy lessons (from the Tesla playbook):

• Vertical integration beats selling SaaS into legacy operators. Mariana explicitly is not a SaaS company — it builds and operates the assets because the gate to software adoption is operator culture (pen, paper, 150 spreadsheets), so software engineers must sit next to operators with shared incentives, not as forward-deployed contractors.
• Bet on techno-optimism applied to “old and archaic” systems — most legacy players try autonomy for a year, fail, and shelve it. Tesla’s edge was barreling through pain when the outcome justifies it.
• Existential focus (“will the paycheck clear?”) forces best work; a clear mission becomes a talent magnet that lets startups pick from the best and retain them through visible impact and growth — something multi-product industrial conglomerates structurally cannot replicate.
• High risk tolerance enables fast decisions without fear-paralysis; clear refusal to abandon projects with worthwhile outcomes is a cultural moat.
• Local jurisdictional alignment is the difference between magic and stagnation — Baglino built the Megapack 3 factory in a JC Penney warehouse in 11 months because every party chose to say “yes” at every code-compliance step.

Career advice:

• Career trajectory is fastest where impact is visible and the company is growing — pick mission-driven, high-growth environments where you can rotate across functions (as both guests did at Tesla).
• For engineers/operators worried about pivoting industries: optimization algorithms in dog-walking, Uber routing, lending, and ad-tech transfer directly to plant and mine optimization. Analog-industry experience is a feature, not a bug.
• Working on “unsexy” infrastructure (grid hardware, mining) is strategically high-leverage right now — the talent magnet effect means top operators are increasingly choosing these companies.

Actionable policy asks:

• Apply the last 50 years of oil-and-gas incentive structures to a national minerals mandate to mobilize private capital with confidence that the policy rug won’t be pulled.
• Durable industrial policy that suppliers and financiers can plan around.
• Federal-state coordination to designate co-located energy + manufacturing build-out zones where local jurisdictions are aligned to say “yes.”
• A federal highway-trust-fund equivalent for the electric grid to fund linear transmission infrastructure connecting industrial zones — a master plan instead of today’s patchwork.

Chapter Summaries

• Cold open and framing: AI dominance and reindustrialization are physical projects — energy, mining, refining, manufacturing, grid. Concerns about AI straining the grid should be a call to action, not a reason to pause progress.
• Introductions: Mariana Minerals (software-first mining and refining, operating copper mine in Utah, lithium refinery being built in Texas, 10 projects in 10 years) and Heron Power (solid-state transformers using silicon carbide to replace steel-oil-copper grid hardware).
• Why American champions matter: silicon carbide was developed via US DOE/Navy/academic R&D — commercialization should happen here. On minerals, the US is 50 years behind China; even with permitting reform, build-out (5 years) and ramp (3-5 years) are the deeper bottleneck Mariana targets.
• Leaving Tesla for the grid: Baglino saw enormous innovation at the grid’s edge but none on the other side of the wire. Local jurisdictional alignment determines whether projects fly (11-month Megapack 3 build) or die.
• Labor and supply chains: factory labor is <10% of COGS; the real US gap is co-located supply chains. China’s 3-hour-drive industrial clusters are the model. Modern factory jobs are skilled, well-paid, technical work.
• Critical minerals sovereignty: Mariana bets on full mine-to-refinery integration plus autonomy — LLM-assisted engineering, agentic procurement, RL-based control of variable refinery feedstocks, short-interval autonomous mining decisions. Software adoption gated by operator culture, so engineers must sit with operators.
• The Tesla model: techno-optimism toward archaic systems, high risk appetite for fast decisions, refusal to abandon worthwhile outcomes, existential focus forcing best work, mission as talent magnet, and visible impact driving retention.
• Industrial workforce in 2026: hire creatively from analog industries — bottling plants, syringe manufacturing, oil and gas, ride-share and ad-tech optimization engineers. Both companies will add ~500 construction/operations jobs per initial facility.
• Closing asks: minerals incentives modeled on oil-and-gas precedent; durable industrial policy; federal-state co-located energy/manufacturing zones; a federal grid trust fund analogous to the highway trust fund.