The State of Modern War: Palantir & Anduril Execs on Drones, AI, and the End of Traditional Warfare

Why does it matter? Because 8 days of munitions isn't a stockpile — it's a countdown.

The US has 8 days of munitions on hand for a major conflict with China. Eight days versus the 800 needed. This conversation between Palantir's Sean Sankar and Anduril's Trey Stevens cuts to the structural rot underneath that number — how 30 years of globalization deliberately dismantled the factories that made deterrence real, and what it will take to rebuild them before the window closes.

• The stockpile was never the deterrent — the factory was; the US lost the ability to regenerate weapons at wartime consumption rates
• From 94% dual-purpose defense spending in 1989 to 86% pure-play specialists today — wartime mobilization capacity is essentially gone
• US regulatory overreach handed China the drone market; semiconductors are following the same path
• Defense innovation has always required a single pathological founder — the F-35 has 400 congressional districts and no owner

Ukraine burned through 10 years of production in 10 weeks — and nobody called the fire alarm

When Ukraine consumed a decade's worth of munitions in 10 weeks of fighting, Sankar says that "probably should have been a five alarm fire that we got the fundamental calculus on deterrence wrong." It wasn't.

The misconception is that stockpiles deter adversaries. Stevens and Sankar agree: it was always the factory. "It was the ability to generate and regenerate the stockpile." Without manufacturing capacity to refill what you expend, even a large inventory is just a fixed clock ticking toward zero.

The numbers make this concrete. A 10,000-to-1 drone production gap versus China. A 23x shipbuilding capacity disadvantage. And that 8-days-versus-800-days munitions figure that Sankar has cited publicly. Meanwhile, the US keeps shooting $2 million interceptors at $20,000 drones — a cost-exchange ratio that, as Sankar puts it bluntly, you "can't have that math work very long."

The deterioration is structural. In 1989, when the Berlin Wall still stood, only 6% of major weapons spending went to pure-play defense specialists — 94% went to dual-purpose companies like Chrysler (Minuteman ICBMs and minivans), General Mills (torpedoes and inertial guidance), and Ford (satellites until 1990). Today that figure has inverted: 86% flows to defense specialists. The WWII mobilization model — flipping commercial factories to wartime production — took 18 months even when those factories existed. The factories are gone. The model is a fantasy.

America's industrial base didn't erode — it was sent away, deliberately, and Tesla is the only proof anyone tried to rebuild it

Stevens grew up in Ohio. His grandparents and aunts worked at GM, Ford, Frigidaire, National Cash Register, Armco Steel. Every single one of those factories closed. None still exist.

That personal history is the macro story. Thirty years of globalization gutted the communities — and the manufacturing knowledge base — that would be essential in any serious conflict. The skills didn't disappear entirely; they went underemployed. Anduril's Arsenal One factory campus in Columbus is deliberately tapping into that dormant knowledge, workers who know how to build things at scale but haven't had anywhere to do it.

The broader indictment: for at-scale manufacturing from a company founded this century, Stevens says there's "really only one that comes to mind — just Tesla. That's it." The US hasn't built serious new manufacturing muscle in a generation.

Re-industrialization, in this framing, isn't an economic preference — it's a national security imperative. The 1993 "Last Supper" dinner, where the Pentagon told 15 of 51 prime contractors to consolidate or die, compressed the defense industrial base down to five primes. The Justice Department stopped a further Lockheed-Northrop merger in 1999, but the damage was already done. What remained was optimized for peacetime procurement, not wartime generation.

The drone market was America's birthright — and regulators handed it to China

Abe Kareem invented the Predator. General Atomics built it. The drone was an American creation. Then the government classified it as a flying missile, slapped ITAR controls on it, and the FAA banned beyond-line-of-sight operations. The domestic market died before it could form.

Sankar is precise about the counterfactual: there was a world where General Atomics had a consumer subsidiary called DJI, where the consumer drone market was entirely American, where US companies rode the price-production curve down and carried that advantage into national security applications. Instead, China owns the market and the manufacturing scale that comes with it.

Now the same dynamic is playing out with semiconductors. Stevens is direct: "almost no amount of money is going to fix this problem for us certainly on a timeline of relevance as it pertains to the risk that Taiwan has in 2027." Once TSMC got ahead of steam, the US didn't invest in a domestic competitor — it let the capability migrate and calcify overseas.

The pattern is consistent enough to be a law: over-regulate a dual-use technology in its consumer phase, cede the production volume and learning curves to a competitor, then discover the strategic deficit a decade later when the cost to recover exceeds the political will to spend it.

Arsenal One is built like a contract manufacturer — because the next war's weapon requirements can't be predicted in advance

The Stinger and Javelin crisis of 2022 is the design brief for Arsenal One. When Ukraine needed those weapons, the US burned through inventory — and then discovered the assembly lines no longer existed. The primes were literally calling retirees back to teach workers how to rebuild them from scratch.

Anduril's answer is modular production capacity rather than optimized single-product lines. Stevens describes it explicitly: "we're thinking about this more like contract manufacturers think about building assembly capacity" — the way a Foxconn builds optics systems for Facebook, Samsung, and Apple on the same floor. Arsenal One is designed so that if the conflict demands Roadrunners, you ramp Roadrunners; if it needs Barracudas, you pivot to Barracudas.

The alternative — what Anduril is explicitly avoiding — is a factory campus locked into one product. "If we build out the factory campus and we said we're just going to build furies here, that provides a tremendous limitation in a moment of conflict." The next war's specific consumption pattern is unknowable in advance. Flexible capacity is the only honest answer to that uncertainty.

The business model difference matters too. The legacy primes respond to government requirements and invest little ahead of capability. Anduril does private R&D and sells the output as a product — a fundamentally different risk posture that requires significantly more private capital but preserves the product discipline that government cost-plus contracting systematically destroys.

Every major defense breakthrough was an act of heresy — and the F-35 has 400 congressional districts but no owner

Churchill built the tank because the British Army couldn't see past horses. Benny Schriever drove the ICBM program. Admiral Rickover built the nuclear navy. Kelly Johnson constructed 41 airframes — including the SR-71, still the fastest manned aircraft — and his explicit operating rule was that he had to play defense to keep government bureaucrats out of his own program.

Sankar's framework: every one of these breakthroughs was "an act of heresy" driven by "a founder-like figure who is so committed pathologically to a different heretical concept" that they see it through despite institutional resistance. The validation only comes in combat, or when the moment finally arrives that proves them right.

Today that figure has been bureaucratized out of existence. Stevens asks: if you go to the Pentagon and ask who's responsible for the F-35, "I'm not sure they even know." The F-35 has components manufactured in 400 congressional districts — it's not an aircraft program, it's a jobs program with aerodynamic properties.

When the US built ICBMs, it ran eight competing programs simultaneously. In peacetime that would be condemned as wasteful. But that competitive redundancy is exactly what produced the breakthrough. The free market logic the US claims to believe in — variance in execution quality, competition of ideas, concentrating on what actually works — was abandoned in defense procurement precisely when it mattered most.

Abstaining from defense AI is not a neutral moral position — it's an active choice that makes wars more deadly

Anthropic refused to let Claude be used in Project Maven without human oversight constraints. The Pentagon labeled them a supply chain risk. Sankar's response to this episode is not diplomatic.

"I don't think abstention from participating in the building of technology for national security is a morally neutral decision. You are making a moral decision when you decide to abstain." The argument runs directly through precision targeting: AI as a command center for better decisions means better discrimination, fewer civilian casualties. Dropping AI-guided precision munitions is ethically superior to dropping dumb bombs on city blocks. Withholding that capability doesn't prevent war — it makes war more destructive.

The harder edge of the argument invokes Theodore Hall, the 18-year-old Manhattan Project physicist who in 1944 walked into the Soviet trade mission in New York and handed over critical bomb secrets — convinced he was securing world peace through mutual deterrence. Sankar's verdict: "every death from communism since 1949 is actually on his hands. And there's no accountability for that."

The structural critique is equally sharp. When a small number of tech executives impose ethics frameworks that constrain what democracies can deploy — with no accountability to voters — Sankar calls it by name: "tyranny by tech bro." Elected representatives authorize these policy decisions. Companies that unilaterally overrule them are substituting their own geopolitical judgment for democratic accountability, with no mechanism for being proven wrong.

Anduril deliberately prices each round below the last — and thinks the current defense-tech seed mania ends badly

When Trey Stevens got his Palantir offer letter, the equity scenarios topped out at $20 billion. People around the company said that was ridiculous — they'd never be worth $20 billion. Now, he says, "people are going out and they're like, our seed round will be priced at $20 billion."

Anduril's explicit counter-discipline: the company never wants to raise the next round at a higher revenue multiple than the prior one. The Series H came in "down pretty significantly from the Series G" — not because investors wouldn't pay more, but because Stevens believes the discipline is strategically essential heading into a medium-term IPO. The fragility created by inflated multiples — the pressure to hit numbers that government sales cycles structurally cannot support — is an existential risk that founders are choosing to load onto themselves.

The advice to the field is blunt: "your product might be awesome, but you can also raise less at a lower price." Defense-tech companies face a specific version of this trap: long procurement cycles, budget unpredictability, and a single buyer who can renegotiate or delay. Raising at consumer-tech multiples against that revenue profile isn't optimism — it's playing chicken with math that doesn't work.

The factory was always the deterrent — and 2027 may arrive before the rebuild does

What this episode reveals is a race with a fixed endpoint. The 2027 Taiwan window isn't a talking point — it's the timeline against which every factory, every production ramp, every procurement reform has to be measured. Stevens says that with unlimited capital starting today, the US could get to a sustainable industrial base in 18 months. But the capital isn't unlimited, the political will is episodic, and the semiconductor gap is already beyond recovery on any relevant timeline.

The adversary's strategy isn't to outspend the US military — it's to wait until the factories are gone and the stockpiles run out in week two. The deterrent was never the missiles in the warehouse. It was always the line that could make more.

---

Topics: defense technology, industrial policy, autonomous weapons, AI ethics, drone warfare, manufacturing, venture capital, national security, China, deterrence