Defense-Tech Silicon Valley Divide

The cultural gap

Silicon Valley and the defense and military world barely talk to each other. The Valley skews Democrat and elite-educated and pacific in its self-image; the warfighter community skews working-class, rural, and middle-American. Tech founders who benefit most from AI, and who will build the AI that displaces white-collar workers, often have no family in the military and no personal experience with warfighters. Alex Karp's framing is blunt: if you are going to meet a general or a warfighter and have never talked to anyone with battlefield experience, "that's probably a huge mistake and probably going to backfire."¹ The failure, in this reading, is relational rather than ideological: worlds that never talk cannot align on anything. Karp adds a historical argument for why the military commands the reverence it does across otherwise divided demographic lines: the United States military, in his account, is the country's most genuinely meritocratic institution, and it integrated racially in Korea before American civilian society did.¹

The nationalization risk

Karp's central warning is structural rather than partisan. If Silicon Valley simultaneously destroys white-collar jobs, the core constituency of one political coalition, and is perceived as hostile to or indifferent about the military, the core priority of the other, then nationalizing the industry becomes the one thing both sides agree on: "If Silicon Valley believes we are going to take away everyone's white collar job and you're going to screw the military, if you don't think that's going to lead to nationalization of our technology, you're crazy."¹ Karp is not arguing nationalization would be right, only that it becomes inevitable absent a course correction.

He also argues the two worlds share a category error on this point: Silicon Valley insists AI is positive-sum and benefits everyone, while privately treating the fight for large language model dominance as viciously zero-sum among labs. What it misses, in Karp's telling, is that the deeper zero-sum game is geopolitical rather than merely competitive: in a world of great-power competition, the decisive question is not who builds the best model but whether America or its adversaries hold the decisive military and technological advantage.

His prescription is industry self-regulation on the model of Hollywood's ratings system, addressing Fourth Amendment questions around AI-enabled surveillance, the fate of the displaced white-collar workforce, and a warfighter maxim that every military application should be optimized to bring soldiers home safely, before Washington imposes a worse regime by itself.

The Anthropic case: punitive escalation

Dario Amodei and Anthropic supply a second data point, and the first instance of a government using a punitive regulatory mechanism against a domestic AI company for declining to comply with its demands. By Amodei's account, Anthropic is the most defense-integrated AI company in Silicon Valley: first on the classified cloud, first with custom national security models, deployed across the intelligence community and Department of Defense. The dispute was not about whether to work with the military at all, but about two specific use cases, domestic mass surveillance and fully autonomous weapons, where Anthropic held red lines. The Department of Defense gave the company a three-day ultimatum to drop the restrictions or be designated a supply chain risk. Secretary Hegseth's subsequent designation of Anthropic as a supply chain risk, a mechanism historically used against Kaspersky Labs and Chinese chip suppliers, marked the first time it had been applied to an American company.²

The irony is that Anthropic had already crossed the bridge Karp advocates crossing: it was the most defense-forward of the major labs, and the response was not simply to find another contractor, the normal market outcome, but to use an unprecedented administrative mechanism to reach into a private contractual relationship. Karp warns about nationalization as a future risk; the Anthropic case reads as a small-scale preview of the same dynamic.

Scale AI and the US-China data picture

Alexandr Wang and Scale AI add a third data point and the sharpest empirical framing in this record of the US-China AI competition. Scale's flagship Department of Defense program, Thunder Forge, based with Indo-Pacific Command in Hawaii, converts military planning processes into AI agent workflows and has compressed decision cycles from 72 hours to 10 minutes.

Wang's competitive analysis names four factors. Espionage is, in his account, the primary explanation for Chinese model progress: "A lot of the secrets about how to train these models leave the frontier labs and make their way back to Chinese labs," so that even talented US labs make progress more slowly than they otherwise would. An energy gap has opened because the US electrical grid has grown roughly flat while China's has doubled in a decade, which Wang calls "just a policy failure." China holds a data advantage through government-run labeling centers across seven cities, voucher subsidies for AI companies, college data-labeling programs, and robotics data factories that even US robotics firms rely on. And a hardware cost gap means an embodied robot costs 20,000 to 30,000 dollars in the US against 2,000 to 4,000 dollars in China, a gap Wang illustrates plainly: "walk down a street in Shenzhen and they've got it."³ His net assessment is a 60-40 to 70-30 US advantage, with the caveat that "there are a lot of worlds where China just catches up or potentially even overtakes."

The divide crossed into banking

Palmer Luckey is a fourth data point, and the first to carry the divide out of AI and hardware entirely and into finance. Where Palantir, Anduril, Anthropic, and Scale AI all sit inside a tech-meets-defense frame, Luckey's bank, Erebor, applies the same logic to the banking system: a US-chartered bank explicitly aligned with the US government, the Department of War, and the intelligence community.⁴

This sharpens the divide thesis in two ways. It shows the divide is sector-general rather than AI-specific: the gap is not simply Silicon Valley versus the military, it is globally-beholden, neutrality-by-default institutions versus explicitly US-aligned ones, wherever they sit in the economy. And it is the constructive counterpart to the Anthropic case: Anthropic shows the divide producing punitive escalation when a company drew red lines, while Erebor shows a founder voluntarily building maximal alignment as a product feature rather than a concession. Luckey, notably, expresses no Anthropic-style red lines in his own account; alignment with the national security state is the pitch, not something to be bounded.

Tensions

Karp speaks from a position of considerable privilege and institutional access, and his warnings about nationalization read differently coming from a company that is itself a government customer than they would from a consumer AI company with no such relationship. The self-regulation analogy to Hollywood's ratings system cuts both ways, since that system also let incumbents capture the regulatory apparatus in ways that raised barriers for new entrants. And how this divide evolves if China's military AI capability closes the gap Wang describes, whether it pushes the US tech industry to coalesce faster around defense or whether political polarization prevents cooperation even against an existential threat, remains an open question.