The Seeing Stone
In August 2025, Palantir Technologies reported its first billion-dollar quarter. The stock had already climbed more than 1,700% since the company's direct listing on the New York Stock Exchange in September 2020, and on the day of the earnings call, it surged past a $430 billion market capitalization — making Palantir the twenty-third most valuable company on the planet, nestled just behind Johnson & Johnson, a firm with more than twenty-three times its revenue and thirty-five times its headcount. Alex Karp, Palantir's co-founder and CEO, a man who holds a doctorate in neoclassical social theory from Goethe University Frankfurt and keeps a wooden tai chi sword in his office, opened the call with characteristic restraint. "I've been cautioned to be a little modest about our bombastic numbers," he told analysts, "but honestly, there's no authentic way to be anything but have enormous pride and gratefulness about these extraordinary numbers." Then he turned to the retail investors who had made the stock a meme-era phenomenon and a conviction trade all at once: "Maybe stop talking to all the haters — they're suffering."
The haters have always been part of the Palantir story. The company provokes — by design, by temperament, by the nature of what it sells. Named for the "seeing stones" of Tolkien's Middle-earth, dark orbs that allowed their users to perceive events across vast distances, Palantir builds software that integrates siloed data into a unified operating picture, enabling its customers — the CIA, the U.S. Army, ICE, the NHS, Airbus, BP, and dozens of others — to find patterns that would otherwise remain invisible.
Peter Thiel, the company's co-founder and chairman, once described the philosophical divide between his two most famous investments: Facebook reveals networks of people who want to be seen; Palantir reveals networks of people who do not want to be seen. The same analytical engine that helped track roadside bomb networks in Iraq can, it turns out, optimize gum placement at checkout counters, or identify targets for deportation using Medicaid data.
That duality — between the sublime and the sinister, the defensive and the invasive, the patriotic and the authoritarian — is the crack that runs through Palantir's foundation. It is the source of the company's moat and the reason for the protests outside its offices, the $10 billion Army contracts and the "No Tech for ICE" campaigns, the 555% annual stock appreciation and the discomfort of anyone who thinks too carefully about what a seeing stone actually does.
By the Numbers
Palantir at a Glance
$2.87BFY2024 total revenue
$1.0B+Q2 2025 quarterly revenue (first time)
~$430BMarket cap (August 2025 peak)
55%Revenue from government customers
93%U.S. commercial revenue growth (Q2 2025 YoY)
~3,800Employees worldwide
1,700%+Stock appreciation since 2020 direct listing
50+Sectors with deployed solutions
The PayPal Cosmology
To understand Palantir, you must first understand the intellectual ecosystem from which it emerged — not Silicon Valley writ large, but a very specific corner of it, a tightly networked group of contrarians who came together at PayPal in the late 1990s and then scattered like seeds across the technology landscape, founding or funding Tesla, SpaceX, LinkedIn, YouTube, Yelp, and Palantir itself. The so-called "PayPal Mafia" was not a mafia in any organizational sense but a shared sensibility: that technology companies should pursue monopoly, not competition; that the most valuable businesses solve problems no one else will touch; and that the conventional wisdom of established industries is usually wrong.
Peter Thiel sat at the center of this constellation. A German-born, chess-prodigy, Stanford Law–educated libertarian who co-founded PayPal, made a legendary $500,000 angel investment in Facebook, and turned it into more than $1 billion, Thiel was by 2003 the most intellectually ambitious investor in the Valley — and the most ideologically unorthodox. He had come to believe that the information age was producing diminishing returns, that Silicon Valley's obsession with consumer internet apps represented a failure of ambition, and that the most important problems — in defense, energy, biotech, transportation — were being systematically ignored by venture capital. "We wanted flying cars, instead we got 140 characters," read the credo of his venture firm, Founders Fund.
Palantir was Thiel's attempt to build the kind of company he believed the world actually needed: one that operated at the intersection of technology and national security, that brought Silicon Valley's engineering talent to bear on the hardest data problems faced by intelligence agencies and militaries, and that did so as a commercial software company rather than a government contractor. The initial investment came from In-Q-Tel, the CIA's venture arm — not because Thiel needed the money, but because the imprimatur legitimized the mission.
The founding team, assembled in 2003 and incorporated in 2004, was quintessentially PayPal-adjacent. Stephen Cohen, who had been an undergraduate at Stanford when he began contributing to PayPal's fraud detection algorithms, became the technical co-founder. Joe Lonsdale, another Stanford undergraduate who had worked at Clarium Capital, Thiel's hedge fund, brought the operational energy. Nathan Gettings provided early engineering. And for the CEO role — the person who would be the face of a company that sold surveillance tools to spy agencies — Thiel chose the most improbable candidate imaginable.
The Philosopher's Gambit
Alexander Karp grew up in Philadelphia, the son of a Jewish pediatrician and a Black artist, in a left-leaning household that produced a child who was severely dyslexic, fiercely intellectual, and constitutionally contrarian. He attended Haverford College, a small Quaker liberal arts school outside Philadelphia, then earned a JD from Stanford Law School — where he met Thiel, and where the two bonded over a shared contempt for law school and an appetite for political argument. Karp considered himself a socialist. Thiel was an arch-libertarian. Out of these debates, an improbable friendship crystallized.
After Stanford, Karp didn't practice law. He went to Frankfurt, Germany, where he studied under Jürgen Habermas, one of the most important philosophers of the twentieth century, and earned a doctorate in neoclassical social theory. He had no background in computer science. No experience running a company. No obvious reason to lead a defense-technology startup. But Thiel understood something about Karp that few others did: the same intellectual restlessness that made him a philosopher also made him a systems thinker, and the same moral seriousness that drew him to critical theory would give him credibility with customers who needed to trust that their data was being handled responsibly. A self-described socialist running a company that sells tools to intelligence agencies — the contradiction was the point. It was, in Thiel's parlance, a secret: a truth that almost nobody would believe.
He had no background in computer science or business when he joined Palantir in the early 2000s, and yet 20-plus years later, he has turned this company into a $400 billion colossus.
Karp joined Palantir around 2005 as CEO and has held the role ever since — a tenure that now stretches past two decades. His management style is flamboyant and unorthodox: he practices tai chi, Nordic skis obsessively, has famously unruly hair, and once told an interviewer he would love to spray his critics with "light fentanyl-laced urine." He describes Palantir as "a rare cult with no sex and very little drugs and we're not poisoning anyone." Employees sometimes refer to each other as "hobbits," in homage to the company's Tolkien nomenclature. The culture is deliberately strange, deliberately intense, and deliberately at odds with the agreeableness that dominates most of corporate America. Michael Steinberger's
The Philosopher in the Valley, which chronicles Karp and Palantir's ascent, captures the essential paradox: a man whose intellectual formation was in the Frankfurt School tradition of critiquing power structures now runs one of the most powerful surveillance infrastructure companies on Earth.
Intelligence Augmentation, Not Artificial Intelligence
The founding insight of Palantir was not about artificial intelligence — at least not in the way the term is used today. It was about intelligence augmentation: the idea that the most valuable thing technology could do was not replace human analysts but make them vastly more effective. In the early 2000s, the U.S. intelligence community was drowning in data. The failure to prevent the September 11 attacks was, at root, a failure of data integration — the CIA, the FBI, the NSA, and other agencies each possessed fragments of the puzzle, but no system existed to assemble them into a coherent picture. Information was siloed not just by agency but by data type, by classification level, by the sheer incompatibility of legacy systems.
Palantir's first product, Gotham, was built to solve this problem. It could ingest data from radically disparate sources — bank records, phone logs, satellite imagery, field reports, social media — and weave them into a single, searchable, relational graph. An analyst could start with a phone number and, through a series of associative queries, surface a network of connections that would have taken weeks of manual work to construct. The system didn't tell you what to think; it showed you what was there, in a way that made patterns legible to human cognition. In Iraq, the Pentagon used Palantir's software to track patterns in roadside bomb deployments and determine that garage-door openers were being used as remote detonators — a prediction derived from the computational correlation of thousands of incident reports.
This approach — the human-in-the-loop model of decision-making — became Palantir's philosophical and commercial foundation. It was, in many ways, the anti-thesis of the fully autonomous AI systems that would dominate Silicon Valley discourse a decade later. Palantir's bet was that the world's most consequential decisions — battlefield targeting, counterterrorism, pandemic response, supply chain management — could not be handed off to black-box algorithms. They required human judgment, augmented by technology that could process information at scales no human could manage alone.
The concept that unified all of this was what Palantir would come to call the "ontology" — the systematic mapping of data, logic, and action to meaningful semantic concepts. Rather than simply storing data in tables or processing it through models, Palantir organized it into a representation of the real world: people, places, events, transactions, relationships. This ontology layer was the moat. Once a customer had built an ontology of their operations in Palantir's platform, switching costs became enormous — not because Palantir locked in data, but because it locked in meaning.
The Government Incubation
For most of its first decade, Palantir was essentially a government company. The CIA, through In-Q-Tel, was the first customer. The U.S. Special Operations Command followed. Then the Army, the Marine Corps, the Air Force, the NSA, the FBI, the IRS, the CDC, the NIH. The company's engineers were embedded with military units in Iraq and Afghanistan, deploying software in forward operating bases, iterating in real time based on the operational needs of soldiers and analysts.
This was not a typical government contracting relationship. Traditional defense primes — Lockheed Martin, Raytheon, Northrop Grumman — built hardware on cost-plus contracts, with timelines measured in decades and incentive structures that rewarded overruns. Palantir built software on commercial terms, deployed it in weeks rather than years, and measured success by whether operators actually used the product. The culture clash was immediate and bitter.
Palantir's battle against the defense establishment
2003Palantir founded with In-Q-Tel backing from the CIA.
2004–2010Gotham deployed across intelligence agencies and special operations forces in Iraq and Afghanistan.
2012U.S. Army launches the Distributed
Common Ground System (DCGS-A), a legacy intelligence platform that soldiers in the field widely criticized as dysfunctional.
2016Palantir sues the U.S. Army, alleging it was improperly excluded from competing for DCGS-A contracts under the Federal Acquisition Streamlining Act.
2016Federal judge rules in Palantir's favor, forcing the Army to consider commercial alternatives.
2019Palantir wins a $800 million Army contract for the Army Vantage program.
2024
The lawsuit against the Army — in 2016, Palantir actually sued the U.S. government for the right to compete — was a defining moment. It was the kind of move that no traditional defense contractor would have attempted, both because suing your customer is commercially suicidal in most industries and because the defense procurement establishment viewed Silicon Valley companies as interlopers who didn't understand how the Pentagon worked. Palantir's argument was simple: the Federal Acquisition Streamlining Act required the government to consider commercially available software before funding custom-built alternatives, and the Army was ignoring the law. A federal judge agreed. The ruling didn't just open the door for Palantir; it established a precedent that would benefit an entire generation of defense-technology startups, from Anduril to Shield AI.
The Seventeen-Year IPO
Palantir stayed private for seventeen years — an almost comically long time for a Silicon Valley company of its scale. By December 2015, when it raised $880 million at a $20 billion valuation in its eleventh funding round, it had accumulated roughly $2 billion in total capital. Investors included Bridgewater Associates, Tiger Global Management, and Morgan Stanley. Karp had repeatedly expressed reluctance to take the company public, and for good reason: Palantir's business was secretive, its revenue was concentrated in classified government contracts, its culture was intentionally opaque, and the scrutiny of public markets would expose all of it to a level of examination that Karp found distasteful.
But by 2020, the calculus had shifted. The company needed liquidity for employees who had been holding equity for over a decade. The COVID-19 pandemic had demonstrated the value of Palantir's platforms — the NHS in the U.K. used Foundry to manage pandemic logistics, and the CDC deployed it for contact tracing — and the public markets were euphoric about technology stocks. On September 30, 2020, Palantir went public via direct listing on the NYSE, referencing a price of $7.25 per share and closing its first day of trading at $9.50, giving it a market capitalization of roughly $16 billion.
The S-1 filing, submitted on August 25, 2020, was remarkable for its candor. It disclosed that Palantir had never been profitable. It disclosed total revenue of $742.5 million in 2019, with 56% from government and 44% from commercial customers. It disclosed that the top twenty customers accounted for 67% of revenue, and that three customers individually accounted for 10% or more. It disclosed stock-based compensation that was staggering even by Silicon Valley standards. And it included a letter from Karp that read less like a corporate filing and more like a philosophical treatise: a meditation on the relationship between software companies and Western democratic institutions, a defense of working with the U.S. military, and a rebuke of Silicon Valley companies that refused to do so.
Our software is used to target combatants and to try to help plan for a potential invasion of a foreign country. We have chosen sides, and we know that our products are being used for good.
The direct listing was a statement of independence — no underwriters, no roadshow, no lock-up concessions. It was also, in hindsight, terrible timing for anyone who sold early. The stock drifted below $10 through much of 2022 and 2023, cratered to the $6 range in late 2022, and appeared to validate every skeptic who had ever called Palantir overvalued, unprofitable, and structurally dependent on government spending. Then the generative AI wave arrived.
The Ontology Meets the Large Language Model
When ChatGPT launched in November 2022, most enterprise software companies scrambled to figure out what AI meant for their products. Palantir had a different problem: it had been building AI infrastructure for twenty years, but nobody had called it that. The ontology layer that mapped customer data to real-world concepts was, functionally, the semantic scaffolding that large language models needed to operate in enterprise environments. The data integration pipelines that Palantir had painstakingly built for the CIA were exactly what Fortune 500 companies needed to connect their own siloed systems to AI capabilities. Palantir didn't need to pivot to AI. AI pivoted to Palantir.
The company's response was the Artificial
Intelligence Platform, or AIP, launched in 2023. AIP layered generative AI capabilities — including integration with models from Anthropic, OpenAI, and Meta's Llama — on top of Palantir's existing ontology infrastructure. The key insight was that large language models, unconstrained, were unreliable — they hallucinated, they lacked context, they couldn't be trusted with consequential decisions. But LLMs grounded in an ontology, connected to real operational data, governed by the access controls and audit trails that Palantir had spent two decades perfecting — those could be deployed in mission-critical environments.
The commercial impact was immediate. In Q2 2025, U.S. commercial revenue grew 93% year-over-year. The company's "boot camp" sales model — intensive, multi-day workshops where Palantir engineers worked directly with potential customers to build a working prototype on their own data — drove adoption at a pace the company had never previously achieved. Boot camps reduced implementation timelines from months to days, turning skeptics into customers and customers into evangelists. The strategy was almost anti-enterprise: instead of selling to procurement committees through eighteen months of RFPs, Palantir sold to operators by showing them what their own data could do.
The Prime Contractor
The TITAN contract, awarded in 2024, was a watershed. TITAN — the Tactical Intelligence Targeting Access Node — is a next-generation ground station designed to process sensor data from satellites, aircraft, and drones and feed targeting information to artillery and missile units. It is, in essence, the nervous system of the future Army. And Palantir won it not as a subcontractor to a traditional defense prime, but as the prime contractor itself — the first time a software company had ever served in that role for a major Pentagon program.
The significance was structural, not just financial. As a prime contractor, Palantir controlled the architecture. It determined what hardware the system used, which subcontractors participated, and how data flowed through the network. This was the position that Lockheed, Raytheon, and Boeing had monopolized for decades, and from which they extracted enormous margins. Palantir's argument was that in an era where software defined the capability, the software company should define the program.
By 2025, the defense partnerships had proliferated. Palantir and Anduril Industries — founded by
Palmer Luckey, another product of the PayPal-adjacent network — agreed to integrate their respective platforms, allowing battlefield data collected by Anduril's Lattice software to flow into Palantir's systems. Palantir partnered with Shield AI to deploy its "Warp
Speed" manufacturing operating system for autonomous drone production. Amazon Web Services and Palantir jointly contracted to provide U.S. intelligence and defense agencies access to Anthropic's AI models. Each deal expanded the surface area of Palantir's integration layer, making it harder for any single competitor to displace.
Palantir is here to disrupt and make the institutions we partner with the very best in the world, and when it's necessary to scare our enemies and, on occasion, kill them.
The rhetoric was blunt to the point of provocation. But the business logic was precise: Palantir was positioning itself as the operating system of the Western military-industrial complex — the layer that connected sensors to shooters, data to decisions, AI models to kinetic outcomes. If the ontology was the moat in the commercial world, the integration layer was the moat in defense. And both moats were deepening simultaneously.
The Uncomfortable Customer
The same capabilities that made Palantir indispensable to the Army made it incendiary to civil liberties organizations. The company's contract with U.S. Immigration and Customs Enforcement — a relationship that predates both Trump administrations but intensified dramatically during them — became the focal point of a broader debate about the role of technology in state power.
The details, revealed through FOIA requests obtained by Just Futures Law and reported by The Guardian and 404 Media, were granular and unsettling. Palantir's tools were deeply embedded in the day-to-day operations of Homeland Security Investigations: agents used Palantir platforms to track air travel, analyze driver's license scans, search across databases containing 4.9 million records of non-immigrant students and exchange visitors, and locate individuals using cell phone records. A tool called ELITE, developed by Palantir, reportedly ingested data from Medicaid databases to generate dossiers and "leads" on people ICE believed to be deportable, assigning a "confidence score" to each individual's current address. In 2025, Palantir won a $30 million contract to build ImmigrationOS, a platform designed to "streamline" the identification and deportation of immigrants.
Palantir's defense was consistent and legalistic: the company is a data processor, not a data controller. It builds the software; customers decide how to use it. Data integrated into Palantir's platforms is "entirely collected, owned, and controlled by the customers themselves, not by Palantir." This is technically correct and emotionally insufficient. The Electronic Frontier Foundation argued that Palantir was "helping consolidate vast troves of government data into a single searchable, AI-enabled interface" — turning discrete, disconnected databases into a surveillance panopticon. The "No Tech for ICE" campaign organized boycotts. Foxglove, a tech-justice nonprofit in the U.K., launched a parallel "No Palantir in Our NHS" campaign when the company's Foundry platform was deployed for pandemic data management.
The controversy was not incidental to Palantir's business. It was constitutive of it. The company built the most powerful data integration platform in the world precisely because it was willing to work with the customers that other Silicon Valley companies refused to touch — defense agencies, intelligence services, law enforcement organizations operating in morally ambiguous domains. That willingness was the source of two decades of engineering iteration in the hardest data environments on Earth. Sanitize the customer list and you lose the product advantage. This is not a bug. It is the architecture.
The Valuation Paradox
By late 2024, Palantir's market cap had surpassed RTX — the parent company of Raytheon, Collins Aerospace, and Pratt & Whitney — at $174 billion, on revenue of approximately $2.87 billion. Lockheed Martin, with $71.3 billion in 2024 revenue, was worth $121.6 billion. Northrop Grumman, with $39.3 billion in revenue, was valued at $69 billion. Palantir, with a fraction of their revenue, dwarfed them all.
The numbers produced a cognitive dissonance that divided analysts into two camps. The bears saw a company trading at more than 60 times forward revenue — a multiple that implied either a generational growth trajectory or a generational delusion. The bulls saw a software company with 80%+ gross margins, a government customer base with nearly zero churn, a commercial flywheel that was just beginning to accelerate, and a platform-layer position in the AI stack that no competitor had replicated. Retail investors, who poured $1.2 billion into Palantir stock in a single month in 2025, voted with the bulls.
The S&P 500 inclusion in September 2024 accelerated everything — forced buying from index funds, increased institutional attention, and the legitimacy that comes from being classified alongside the largest companies in America. Palantir's stock hadn't traded below $100 since April 2025. It hadn't been below $10 since May 2023. The journey from $6 to $160 had taken about thirty months.
Karp, for his part, seemed aware of the paradox. He had spent years telling investors that the company was undervalued, that the market didn't understand the durability of government revenue or the leverage of the ontology layer. Now the market had overshot in the other direction — at least by any conventional metric — and Karp's posture shifted from vindicated insurgent to something more complicated: a CEO who needed the stock to justify its valuation through sustained execution at a rate that was, by historical standards, almost unprecedented.
The Technological Republic
In 2025, Karp published
The Technological Republic, co-authored with aide-de-camp Nicholas Zamiska, and the book crystallized an ideology that had been implicit in Palantir's operations for two decades. The central claim: the survival of Western democracy depends on the technological revitalization of the military-industrial complex. Silicon Valley's talents should serve national purpose. National competence should restore civic pride. The West is superior, and it must build the tools to defend that superiority.
The book was, as The New Yorker's review noted, a demonstration of how Trumpian nationalism and Muskian technological ambition "might profitably fit together" — a synthesis that Karp embodied despite disliking Trump personally and admiring Musk from a distance. The argument was uncomfortable precisely because it was coherent: if you believe that AI will restructure military power, and that the West's adversaries are investing aggressively in AI-enabled warfare, then the logical conclusion is that the most important thing Silicon Valley can do is build weapons-grade software for democratic governments. The moral weight of the argument depends entirely on whether you trust those governments to use the tools wisely.
Karp's philosophical lineage — Habermas, the Frankfurt School, critical theory — made the position stranger and more compelling than it would have been from a conventional defense hawk. Here was a man trained in the intellectual tradition that interrogates power structures, who had concluded that the most important thing he could do was build the infrastructure of state power. The contradiction was not hypocrisy. It was the animating tension of the entire enterprise.
The Seeing Stone in the Dark
There is a scene in Tolkien where Pippin, unable to resist, gazes into the palantír and is seized by the Eye of Sauron. The stone does not lie; it shows what is there. But the seeing is not neutral. The viewer is changed by what he sees, and the act of seeing creates a connection that can be exploited by those on the other side.
Palantir Technologies, in August 2025, employed roughly 3,800 people and generated more revenue from the U.S. government than from any other customer category. Its software was deployed in more than fifty sectors across multiple continents. The Israeli Defense Forces used it to strike targets in Gaza. The LAPD used it for predictive policing. ICE used it to cross-reference Medicaid enrollment data with immigration databases. The NHS used it to manage a pandemic. Airbus used it to optimize supply chains. The U.S. Army used it to connect satellites to artillery batteries.
On the Q2 2025 earnings call, Karp told retail investors that U.S. revenue had nearly quintupled in five years, from $156 million to approximately $733 million. Revenue outside the U.S. had doubled, from $133 million to $271 million. The company was profitable, growing, and — for the first time in its twenty-two-year history — operating with the kind of momentum that made the valuation feel less like a bet and more like a down payment on something much larger.
At the Hill and Valley Forum in Washington in April 2025, a heckler shouted at Karp from the balcony. He responded calmly, telling the audience he believed it was her right to express her views. Then he continued making his case for the seeing stone.
