Silicon Valley Defense Tech:
New Powers and a Redeemed Sample

The starting point of this storyline is not any valuation announcement, but a contrast in which history nearly completely reversed itself.

In 2018, Google participated in the Pentagon's "Project Maven," using AI to analyze drone surveillance footage to assist in targeting; hundreds of employees signed a protest letter, and the company was forced to abandon the contract and pledge not to use AI for weapons. Eight years later, history nearly completely reversed: in 2026, Google allowed Gemini to access military classified networks for "any lawful purpose," nearly 600 employees protested, but management did not back down this time—having quietly removed the weapons ban clause from its public AI principles in February 2025. Around the same time, OpenAI quietly revised its usage policy in January 2024, removing clauses that had explicitly prohibited "weapons development" and "military and warfare" applications.

Numbers are more honest than attitudes: in 2025, total global defense tech VC investment reached $49.1B, up 80% YoY; defense tech equity financing was up 145% YoY; VC exit volume grew from $18.2B to $54.4B, +199%. This great pivot sets the tone for the entire storyline: the valuation frenzy in every act below is not an abstract industry cycle, but a capital reaction after an entire industry collectively tore up its moral red lines from the previous decade.

After the great pivot, the protagonist of the first act is capital itself—it needed a narrative framework to repackage "making munitions" into something honorable.

The "American Dynamism" investment thesis, established by a16z in 2022 and led by partner Katherine Boyle, is the most important driver of this narrative. The annual "American Dynamism 50" list in March 2025罕见地以"Taiwan Strait conflict" as its narrative scenario, featuring Anduril, Castelion, Saronic, and others, with a blunt assessment: "The question is not whether America should build, but whether America builds first, or cedes the power to set the rules of war to Beijing."

The most dazzling specimen of this narrative is Anduril—founded by Oculus creator Palmer Luckey in 2017, its valuation rose from approximately $250M in 2018 to $61B in the May 2026 Series H, and then to $100B under negotiation in July 2026, up over 3x in one year; in March 2026, it won a US Army enterprise-level agreement worth up to $20B over ten years. Across the Atlantic, the Russia-Ukraine war彻底改写了Europe's perception of "strategic autonomy," the EU launched the approximately €800B "ReArm Europe" plan, and Helsing—founded in Munich in 2021 by computational biologist Torsten Reil and others—raised $1.8B in its July 2026 Series E, pushing its valuation to $18B, the largest single financing round in European defense startup history.

Mach Industries founder Ethan Thornton dropped out at 19 to start his company; in 2023, a self-funded prototype test exploded, injuring team members with shrapnel; the company's June 2026 Series C valuation rose to $1.8B, up 4x in one year. His own definition of this race is clear-eyed: "I don't think we can outcompete China in manufacturing volume... This is a chess game against an opponent, with hundreds of different products on the board." But controversy follows just as closely—a former project manager criticized the company for securing massive financing "with no actual product, no test flights, no demos"; even Helsing's co-founder publicly warned of "a defense tech bubble in Europe," estimating "80% won't survive."

The valuations in Act I are still "under negotiation" or "projected." Act II examines what this story truly looks like once it lands on the ground.

In 2026 Q2, Palantir's revenue was $1.936B, up 93% YoY, US commercial revenue up 149% YoY, adjusted operating margin 62%, Rule of 40 hit 155%—this marks a transition from "a stock that tells stories" to "financials that crush the narrative." After earnings, the stock surged 29.5% in a single day, and market cap briefly hit $380.7B, momentarily surpassing Lockheed Martin; the skepticism of short-sellers Michael Burry and Citron Research grew increasingly isolated against these numbers. Karp himself put it this way: "For the first time, people believe us."

Underpinning these financials are two playbooks that the entire industry is now imitating. Ontology connects clients' digital assets to real-world objects; Palantir's commercial head Ted Mabrey stated it bluntly: "If you want to build the next Palantir, you should build it on Palantir." The Forward Deployed Engineers (FDE) model embeds engineers directly at client sites to write production code—this clumsy method invented in 2005 evolved into the "boot camp" playbook after the 2023 AIP launch, with 1,000+ sessions and a近75% conversion rate, now being密集copied by OpenAI and Anthropic.

And the starting point for all of this was a company that almost no one wanted to invest in: the idea was first rejected by Sequoia and other VCs, and it was ultimately the CIA's venture arm In-Q-Tel that made the first seed investment, with the CIA becoming the sole client from 2005 to 2008. Twenty years later, this path that started from an intelligence agency culminated in the August 2025 Army enterprise-level agreement worth up to $10B over ten years—consolidating 75 previously independent contracts.

"Building an empire" does not come without costs. Act III examines the two types of pressure this business must endure once it has redeemed its promises: domestic controversy and industry infighting.

In May 2025, The New York Times reported that Palantir was working with the "Department of Government Efficiency" (DOGE); critics argued this was constructing a cross-agency "master database," though the company's privacy director publicly denied it. In June 2025, Senator Wyden and Representative Ocasio-Cortez co-signed a letter to Karp accusing the relevant activities; there are currently 12 independent lawsuits alleging the related plans violate the Privacy Act. Around the same time, ICE successively used three sets of tools—ICM, ImmigrationOS ($30M contract), and ELITE—for immigration enforcement and deportation; in May 2025, 13 former Palantir employees—including members of the privacy and civil liberties team—co-signed an open letter arguing that the company's Trump-era business violated its original ethical principles.

The rifts within the industry are even more intriguing. In July 2025, Anthropic won a two-year, $200M Department of Defense contract with an explicit red line: Claude must not be used for "fully autonomous weapons" or "large-scale domestic surveillance targeting Americans." In March 2026, Secretary of Defense Hegseth declared Anthropic a "supply chain risk," ordering that "no contractor may engage in commercial dealings with Anthropic"; Anthropic promptly sued the Department of Defense, and a federal judge initially granted a preliminary injunction blocking the designation, but a federal appeals court then denied Anthropic's request for relief, and the conflict continues to escalate. According to TechCrunch, Palantir is caught "in the middle"—Claude is reportedly still embedded in Palantir's Maven system, "suggesting hundreds of targets and providing precise coordinates" for US operations against Iran, but regulatory uncertainty has already prompted a number of clients, including Lockheed Martin, to开始转向other model providers.

At this point, the four acts of the storyline have answered "what happened." The finale asks the final question: why Palantir, and why now.

On the August 2026 earnings call, Karp uttered the sharpest line in this storyline: "There is something Marxist about our business—many of the companies building large language models, whether intentionally or not, want to seize the means of production of their partners... They are superior to you, and you deserve to be colonized." This is not merely a provocation, but the core narrative of Palantir's "AI sovereignty" positioning: the profit pool只剩two layers—compute and the application layer; the pure model layer is being commoditized.

Behind this statement lies Karp's own backstory of rejection by mainstream academia. While pursuing his doctorate at Goethe University Frankfurt, he initially hoped to study under Frankfurt School philosopher Jürgen Habermas, but according to Karp, Habermas wrote in 2000 to refuse继续指导: "You simply cannot compete with the literary critics and theorists who have recently expressed opinions." In February 2025, Karp and the company's chief legal officer co-authored The Technological Republic: Hard Power, Soft Belief, and the Future of the West, whose opening sentence is: "Silicon Valley has lost its way." The core argument is that Silicon Valley abandoned its World War II/Cold War-era partnership with the state, turning instead toward "technological escapism" in consumer markets—reviews were polarized; The New York Review of Books讽刺它"like a LinkedIn post stretched to 300 pages," but几乎no reviewer questioned Karp's own deeply learned philosophical training.