The State of Modern War: Palantir & Anduril Execs on Drones, AI, and the End of Traditional Warfare
Summary
The binding constraint on US military power is increasingly the factory, not the quality of its best weapons. Sankar contrasted Ukraine consuming “10 years of production in 10 weeks” with an industrial structure that shifted from 6% of major-weapons spending going to defense specialists in 1989 to 94% today. Friedberg added the stark relative-capacity figures: a 10,000-to-one drone-production gap and 223x shipbuilding disadvantage versus China.
Anduril’s wager is that a modular, privately financed manufacturing platform can restore the volume traditional procurement lost. Its five-million-square-foot Arsenal-1 campus in Columbus is designed to pivot among Roadrunner, Barracuda, Furies and future systems like a contract manufacturer, avoiding the Ukraine-era spectacle of calling retirees back to reconstruct Stinger and Javelin assembly lines. Stephens said the model requires tremendous capital and probably cannot support “a hundred new primes.”
The required weapons mix is shifting from exquisite scarcity toward attritable mass, but not toward an all-drone force. Sankar’s simplest economic warning was that “you can’t keep shooting $2 million interceptors at $20,000 drones”; Stephens still called B-2s, Patriots and existing precision weapons exceptional, but said the high-low mix must change massively over five to 10 years. Even with unlimited cash, he estimated it would take roughly 18 months to put the country on track for a sustainable industrial base.
Defense innovation remains a monopsony fight in which products must often be validated from the battlefield backward. Palantir had to sue its customer for the right to compete, while deployed users bent rules to obtain software that worked; Anduril then reached $10 million of annual revenue in 22 months versus roughly five years for Palantir by reusing that institutional knowledge. Sankar’s governing archetype is the “heretic” founder who builds what the buyer does not yet know to request.
Defense-tech capital will follow a power law, making valuation discipline as important as access to money. Stephens warned founders that they “could raise less at a lower price” and said Anduril deliberately lowers its revenue multiple with each round, including from Series A to Series G, ahead of a medium-term IPO. Public capital can help most where supply chains bottleneck—brushless motors, critical minerals and refining—not by “peanut butter” funding every entrant equally.
Autonomy does not eliminate human responsibility; the speakers’ preferred model is accountable command rather than a universal human-in-the-loop rule. Stephens cited CIWS, which must respond to incoming threats autonomously, while leaving someone on the vessel accountable for its actions. He argued AI can improve discrimination and reduce civilian casualties, and that refusing to participate is not morally neutral: “You are making a moral decision when you decide to abstain.”
The 2040 fork is economic as much as military: dependency could produce a Chinese-led order, while reindustrialization could rebuild both deterrence and the middle class. Stephens highlighted that 80% of APIs for generic drugs are produced by China and warned that semiconductor dependence is difficult to reverse before the cited 2027 Taiwan risk window. Sankar’s downside is a Chinese century of “vassal states” playing by rules America no longer helps set; Stephens’s upside is reindustrialization, trusted institutions and a middle class that believes its children’s future will be better.
Deep dive
1. Deterrence, not enthusiasm for war, anchors the mission
Sankar traced Palantir’s founding premise to a false post-9/11 tradeoff: why must society choose privacy or security when technology could push out the frontier of both? The company began narrowly, solving counterterrorism problems for a handful of institutions while pairing that mission with a technical approach to privacy and civil liberties.
Stephens’s moral baseline was categorical: “War is awful. War is bad.” Some adversaries will nevertheless use might to define the rules, so the purpose of preparation is to make conflict unthinkable—or, if deterrence fails, to win “decisively and quickly.”
His test was the general who must call parents after a combat death: nobody seeks that responsibility. Palantir’s and Anduril’s stated objective is therefore not more warfare, but capabilities sufficiently credible that an adversary decides a challenge is not worth attempting.
Sankar called Silicon Valley’s former hostility to defense a “beautiful consequence” of peace. The Valley itself grew from defense—Lockheed, Corona satellites and Cold War urgency—but the post-Cold War “end of history” made threats feel abstract and the military-industrial-complex critique seem sufficient.
2. Geopolitical shocks ended Silicon Valley’s defense holiday
Russian tanks crossing the Ukrainian border were Sankar’s backtest for the cultural turn: many people protesting national-security work also displayed Ukrainian flags, exposing a “policy mismatch” between desired outcomes and opposition to the tools required to produce them.
Stephens added that global technology companies did not necessarily understand themselves as American institutions. Some internal protests were signed by non-US citizens whose obligations and assumptions naturally differed from those of a Cold War-era domestic industrial workforce.
The work became markedly less controversial after 2017–18, but Friedberg resisted complacency. He listed Crimea in 2014, militarization of the Spratly Islands in 2015, Iran reaching breakout capability in 2017, a pogrom in Israel and Houthi disruption of Red Sea trade as empirical evidence that deterrence is eroding.
Friedberg framed the reversal through market value: Palantir at a cited $400 billion, Anduril reportedly raising at a $60 billion valuation after landing a $20 billion Army contract. The guests’ answer to “Silicon Valley taking over defense” was that civilian industry historically underwrote defense—and only recently stopped doing so.
3. The stockpile never deterred alone; the factory regenerated it
Sankar’s historical contrast was “missiles and minivans.” Chrysler built the Minuteman ICBM, General Mills applied grain-processing research to torpedoes and inertial guidance, and Ford built satellites until 1990. America’s industrial base served economic prosperity and the security that made that prosperity possible.
In 1989, only 6% of major-weapons spending went to pure-play defense specialists and 94% to dual-purpose companies; today, Sankar said, roughly 94% goes to defense specialists. Losing the broader industrial base meant losing volume, supplier depth and unexpected cross-industry inventions.
Ukraine burning “10 years of production in 10 weeks” should have been a five-alarm warning. The mistaken deterrence theory centered on stocked weapons; Sankar’s correction was that “it was always the factory”—the capacity to generate, expend and regenerate those stocks.
The force’s high end remains formidable, but its economics are brittle. “You can’t keep shooting $2 million interceptors at $20,000 drones,” Sankar argued, especially amid Friedberg’s cited 10,000-to-one drone-production gap and 223x shipbuilding disadvantage versus China.
4. Arsenal-1 is a platform for changing the production mix
Stephens tied deindustrialization to his own Ohio family: relatives had worked at GM, Ford, Frigidaire, National Cash Register and Armco Steel, yet every referenced factory closed. Anduril’s response is a five-million-square-foot Arsenal-1 campus in Columbus that can draw on underemployed manufacturing knowledge.
The Arsenal software platform, including work with Palantir Foundry, is intended to reduce automation and process overhead. Its more important design choice is modularity: locking the campus to only Furies, Roadrunners or Barracudas would make it unresponsive to the specific demands of the next conflict.
Stephens compared the model to contract manufacturers that build optical systems across competing VR headsets. Shared skills, scaled component purchasing and a reusable supply chain should let Arsenal “pivot on a dime” when demand shifts among products.
Ukraine supplied the counterexample: after warehouse inventories of Stingers and Javelins were depleted, viable assembly lines and experienced workers no longer existed. Primes literally called retirees back to teach a new workforce how to build them.
5. Private product bets challenge a customer accustomed to writing specs
Anduril funds its R&D privately and sells the resulting product; traditional primes largely respond to government requirements. Stephens said this model is capital-hungry, and he doubts the market can sustain “a hundred new primes,” even though more competition would be healthy.
Sankar described defense as a monopsony: one buyer holds extraordinary power, even when it is wrong. Churchill, as head of the Royal Navy, developed the tank because the British Army failed to see that horses would not define the next battle; important innovations repeatedly begin as “an act of heresy.”
The 1993 dinner formalized consolidation after the Cold War: the department brought 15 of 51 major contractors together, told them budgets would fall and gave permission to merge. By 1999, the Justice Department blocked a proposed Lockheed–Northrop combination that would have reduced the remaining five primes to four.
6. Palantir proved new entrants from the field backward
Palantir was not welcomed by the acquisition system and ultimately sued its customer merely for the right to compete. Its validation came from deployed users saying, “What I’m being given doesn’t work. I’d like to come home,” then bending rules to obtain software that did.
Those facts on the ground eventually overcame closed Washington doors, but the path took years. Palantir needed roughly five years to reach $10 million in annual revenue; Anduril did it in 22 months because Stephens, Matt Grimm and Brian Schimpf had already lived through Palantir’s learning curve.
Hardware initially looked easier because officials understood bills of materials, production costs and acceptable margins. Software was judged through lines of code or cost-plus accounting, which failed to reflect the R&D amortized across government and commercial customers and the small marginal price relative to that R&D.
At Founders Fund, Stephens searched federal bidders and met hundreds of GovTech companies over three years, finding only one investment—KDM, later renamed Expanse and acquired by Palantir Networks. The absence of credible companies produced Anduril’s founding thesis: build a next-generation prime that is “software-defined and hardware-enabled,” not hardware-defined with software attached.
7. Defense-tech returns will concentrate, so round discipline matters
Sankar rejected “innovation theater” that spreads funds evenly among companies. Like venture portfolios, defense programs follow a power law: experimentation should remain broad initially, but capital must eventually concentrate behind the products and teams demonstrating superior performance.
Stephens offered the category test: space investors without SpaceX, crypto-infrastructure investors without Coinbase and social-media investors without Facebook probably lost money. Defense-tech allocators should expect the same concentration rather than forgetting how prior booms resolved.
Yet category success has pushed expectations from Palantir employees doubting a $20 billion outcome to seed companies behaving as though $20 billion were the starting point. Stephens’s advice was the Silicon Valley line founders rarely hear: “You could have raised less at a lower price.”
Anduril deliberately aims to lower its revenue multiple each round; Stephens said the Series A multiple was materially below Series G despite investors’ willingness to pay more. That discipline reduces the risk of “playing chicken” with implausible operating targets before a medium-term IPO.
8. Strategic capital belongs at bottlenecks, not everywhere
Sankar recalled that 96% of integrated circuits sold in 1968 went to Apollo, yet Bob Noyce kept government funding below 4% of Fairchild’s R&D. By retaining control of the roadmap and chasing Moore’s law, industry reached the price-performance later required for precision-guided munitions.
The cost-plus counterexample was the shuttle at roughly $50,000 per kilogram to orbit; Sankar contrasted it with heavy-reuse Starship at “under $20 a kilogram.” A contractor paid for costs destroys profit by deleting costs, while a product company can turn lower costs into margin and lower prices.
Friedberg cited $200 billion that Emil Michael wants to deploy. Stephens saw value in “Deal Team 6” and the Office of Strategic Capital when directed at structural constraints: many drone makers can exist, but all may bottleneck on brushless motors, domestic mineral refining or another thin supplier layer.
Stephens compared that role with Air Force investment in the titanium supply chain during the 1950s and ’60s. The first customer may be uneconomic, so strategic offtake agreements and a one-time capital bolus can establish capacity that later customers sustain.
9. Production capacity creates the next innovation cycle
Stephens called drones “an American birthright,” beginning with Abe Karem and the Predator at General Atomics. ITAR treatment of drones as flying missiles and FAA restrictions on beyond-line-of-sight operations then suppressed the domestic market; his counterfactual was a General Atomics consumer subsidiary called DJI.
Pharmaceuticals illustrate why capacity affects national will, not merely weapons supply. With China producing 80% of APIs for generic drugs, Stephens asked whether Americans would defend sovereignty if the price were a five-year-old dying from a previously trivial ear infection.
Stephens also connected production to research. Google’s 2017 “Attention Is All You Need” work arose from pursuing a 3% improvement in Google Translate—“something more banal leading to something more revolutionary.” He connected that mechanism to 50% of new clinical trials now occurring in China.
Stephens put semiconductors in the same category: after TSMC gained momentum, the US allowed semiconductor capacity to run away elsewhere. In his assessment, almost no amount of money can fully repair that dependence on a timeline relevant to the cited 2027 Taiwan danger window.
10. Decision advantage requires flexible buying and accountable autonomy
Stephens and Sankar framed AI and software as the third offset: after nuclear weapons, then stealth and precision-guided munitions, the next advantage is the ability to “outthink and outexecute” an adversary. Hardware and software must interoperate, but that architecture must be earned through tests, exercises and combat.
Sankar was skeptical that frameworks such as MOSA can pre-resolve innovation’s messiness: “The process always destroys all of the innovation.” The department’s more promising change is to buy an outcome, allowing acquisition officials to change which system achieves it instead of being trapped by a fixed quantity of a fixed product.
Treating drones and munitions as consumables would couple every order to an exercise and expenditure plan. Replenishment then creates a durable demand signal, while successive buying cycles replace aging inventory with a cheaper, improved generation.
On readiness, Sankar described an exceptionally sharp spear tip attached to a weak shaft: combatant commands experience demand, while services and industry control supply, and the two adjust slowly. Minimum-rate production merely keeps lines open; it is not deterrence, even if defense companies trading below two times revenue reveal that the business is hardly overprofitable.
11. Eighteen months could change the trajectory, but leadership must choose it
Friedberg recalled Sankar’s earlier estimate of eight days of munitions for a major China conflict against roughly 800 days needed. Stephens cautioned that readiness varies across weapons, but said even unlimited cash would require about 18 months to put the country on a sustainable industrial trajectory.
Anduril had opened Arsenal-1 “yesterday” to begin making Furies and Stephens said the company was not taking its foot off the gas over the next 18 months, yet physical run-rate limits remain. His concern was political: instead of a decisive mobilization, the country will probably trickle investment out and remain vulnerable longer.
“The wars of today are fought with the weapons of yesterday,” Stephens said. B-2s, Patriots and precision weapons remain extraordinarily capable, but are expensive and difficult to replenish; attritable mass must be built now to exist in inventory when future conflicts arrive.
He expects the high-low mix to shift substantially over five to 10 years, not for exquisite systems to disappear. Autonomous Anduril systems already operate in combat, but still represent a small share of how wars are fought.
12. Democratic accountability—not vendor abstention—is the ethical line
Stephens rejected the idea that autonomy is novel, citing CIWS systems that autonomously engage incoming threats because human reaction is too slow. Accountability remains human: someone on the vessel is responsible for the system just as a soldier is responsible for a gun.
AI’s ethical promise, in his framing, is better precision, discrimination and command decisions—continuing the move away from indiscriminate bombing. “Abstention from participating” is not morally neutral; vendors are making a moral choice whether they build or refuse.
Sankar answered the surveillance charge by saying Palantir collects no customer data: it is closer to Excel, but with much more security and auditability. If the company discovered illegal use, he said he would report it to the agency’s independent Inspector General.
His deeper objection to vendors selectively narrowing lawful policy was “tyranny by tech bro”: unelected technologists would constrain democracy without public accountability. Theodore Hall was his cautionary example—a gifted Manhattan Project physicist who gave nuclear secrets to the Soviets believing two nuclear powers would secure peace, without accountability for the consequences.
13. Institutional legitimacy determines which 2040 arrives
Stephens traced anti-defense distrust to Vietnam’s credibility rupture between academia and the military; he also added the personal divide, noting that prior-service and military-connected communities had largely disappeared from Silicon Valley. Sankar illustrated that divide by saying almost nobody in his Stanford undergraduate lectures reports an immediate family member serving. Distance leaves a cartoon model of how military decisions, doctrine and oversight actually work.
On Snowden, Sankar distinguished policy disagreement from abuse: he said investigations found fewer than 12 improper-access cases, largely technology errors, while elected governments repeatedly renewed the collection policy. The underlying dispute is therefore what law should authorize, not whether civil servants deserve effective tools.
Stephens said adversaries amplify domestic fractures, citing Soviet funding of the Vietnam peace movement worth $7 million in 2026 dollars and alleging CCP-linked money behind some Palantir protests. He called the strategy rational and noted that America has also advanced cultural values abroad, including CIA support for modern art.
Their clearest disagreement was communications: Stephens thought Palantir’s early silence surrendered the narrative, so Anduril deliberately told its story; he believes Palantir later moved toward his position. Their shared 2040 fork is starker: either Chinese rule-setting and global dependency, or American reindustrialization, trusted institutions and a middle class convinced its children’s future will be better.