Winning the AI Race (All-In and Hill & Valley Forum, 2025)

An All-In event in Washington co-produced with the Hill & Valley Forum, held the day the administration published its AI Action Plan. The transcript Kyle supplied (auto-generated, and garbled in places — names especially) runs from the opening through the Kelly Loeffler conversation; the raw text is kept privately with the wiki’s attachments. Filed from the Apple Note “The Anduril Thesis” (created 2025-02-28), where it sits among the reindustrialisation sources.

Why it belongs with The Anduril Thesis

It is the policy-and-industry version of the same argument: manufacturing capacity, energy and a trained workforce are what decide the contest with China, and AI is useful mainly as a multiplier on all three. Almost every speaker lands on the same claim — that AI is creating industrial jobs rather than destroying them — which is also the claim most in need of independent evidence.

The frame (hosts)

David Sacks, six months into his role as the White House AI and crypto adviser, and Jacob Helberg of the Hill & Valley Forum set it up as a race the US must win. David Friedberg argues the press narrative of AI destroying jobs is wrong on the ground, and that the story is new industries plus the industries that enable AI (mining, energy, chips, fabs, data centres). Chamath Palihapitiya casts it as a choice between treating technology optimistically or with mistrust — and says the mistrust let rivals emerge.

Michael Kratsios — the AI Action Plan

The Director of the Office of Science and Technology Policy describes a plan ordered by an executive order in the President’s first week: roughly 90 concrete actions the executive branch can take, organised by Sacks into three pillars — innovation, infrastructure, and exporting the American AI ecosystem.

• A public request for input drew over 10,000 responses, including from Hollywood actors.
• It is deliberately an action plan: things achievable in six to nine months.
• Deregulation happens agency by agency, because AI will be embedded in drones, cars and medical diagnostics rather than regulated as one thing.
• Government data — above all the Department of Energy’s national-lab datasets — should be cleaned up and opened for scientific discovery.
• Infrastructure includes categorical permitting exclusions for data centres and power on federal land, and retraining electricians and HVAC workers, whom data-centre builders say they cannot find.
• On federal preemption of state AI laws the plan is silent; Sacks separately calls a patchwork of state regimes — over a thousand bills — a national-security problem.

Chris Power (Hadrian) — reindustrialising with AI factories

Chris Power’s presentation is a history lesson with a business plan attached: great powers rise on industrial capacity, offshore it once comfortable, and are exposed when conflict comes. His claims about the gap with China are stark — US munitions run out within days in war games and take years to replenish, and Chinese shipbuilding capacity dwarfs American. The skilled workforce under the US defence base, he says, is ageing out faster than it can be replaced.

Hadrian’s answer is AI-run precision factories. He claims roughly 4x the productivity of a typical US machine shop (which he puts at ~20% uptime), 10x per worker, training new staff in about 30 days, and a workforce drawn entirely from people who had never worked in a factory. A new Arizona factory, chosen for permitting, energy and regulation, is promised within six months with 350+ jobs, and 2026 factories are aimed at submarines, ships and munitions. In Q&A he names energy as the decisive subsidy China provides, and says tariffs matter for reshoring commercial volume. Chamath discloses that he led Hadrian’s Series C and sits on its board. These are company claims made on stage by a CEO and his lead investor.

Jake Loosararian (Gecko Robotics) — AI that creates energy

Gecko’s robots crawl, climb, swim and fly through power plants, refineries, ships and aircraft to capture condition data that inspectors once gathered by hand. His argument inverts the usual one: AI can unlock energy as well as consume it. His worked example is a gas plant running below its rated output, where combined robot and operational data traced the loss to a steam problem — about a 1% efficiency gain. Extrapolated across the US thermal fleet, he puts that at nearly 12 GW without building anything new, and he cites a Department of Energy study warning of steeply rising blackout risk by 2030. He also mentions a new shipbuilding agreement signed during a presidential visit to Pittsburgh.

Worker vignettes

Short films from a Tampa General Hospital nurse, a Navy submarine-parts supplier whose quoting process shrank from days to minutes, a hospital in Michigan’s Upper Peninsula, and Panasonic Energy’s Nevada battery plant, which recruits technicians from casinos and auto shops. Every one frames AI as returning time to the core job.

Shyam Sankar (Palantir) — superpowers, not efficiency

In conversation with Christian Garrett of Hill & Valley and 137 Ventures, Shyam Sankar argues the goal should be making workers many times more productive, not marginally more efficient. Adoption goes best where institutions let frontline workers build their own applications, and worst where tools are imposed from the top. His sharpest examples: electronic health records roughly halved how many patients a clinician can see per hour, and battery-plant training that took three years now takes three months with an AI assistant. He describes Palantir’s American Tech Fellows programme — a four-week boot camp placing overlooked, often blue-collar builders with customers — and goes as far as declaring the traditional college degree dead.

Paul Buchheit — who gets to build

The Gmail creator (now starting a new firm, Standard Capital) argues that plain-English programming widens the pool of founders dramatically, that robotics is surging among Y Combinator startups, and that intelligence plus energy are the inputs to wealth. He recalls that OpenAI was incubated at YC out of worry that AI would be locked up inside Google, prefers a world of many foundation models plus open source to keep closed vendors honest, and says AI assistants now compete directly with social media for young people’s time.

Kelly Loeffler (SBA) with Keith Rabois — small manufacturers

The SBA administrator’s headline example is a 60-person bicycle factory in Indiana making bikes in the US again. She says the prior administration barred SBA-backed loans from funding AI purchases and that she reversed it; the SBA guarantees rather than lends directly, handles on the order of 2,000 Main Street loans a week, and — by her account — has returned to pre-pandemic staffing while running at no cost to taxpayers. Keith Rabois argues AI gives small firms access to information, products and back-office capacity once reserved for large companies, and predicts a barbell in which small firms eat into the mid-market.

What to keep in mind

The event is advocacy — an administration, its allies and portfolio companies on one stage. The most useful material for The Anduril Thesis is Power’s history of industrial decline and his energy-subsidy point, and Loosararian’s “AI creates energy” inversion; the job-creation claims are asserted rather than evidenced.

Connections

• The Anduril Thesis · Hill & Valley
• Hadrian · Chris Power · Palantir · Shyam Sankar
• David Sacks · Chamath Palihapitiya · Keith Rabois