Palantir

Seventeen years of losses became the AI era's defining software stock: +2,469%, with an 85% crash and a dip below the listing price along the way

Palantir's direct listing opened on September 30, 2020 against a $7.25 NYSE reference price and closed its first day at $9.50, up 31%. What followed was exactly the retail-driven trading its own risk factors had predicted: the stock became a retail and meme-era favorite, roughly quadrupling to about $34 by late January 2021 while the company was still losing over half a billion dollars a year ($520M net loss in FY2021).

Then the S-1's warnings cashed in one by one. Lock-ups released in early 2021, option-expiration selling pressed on the float, revenue growth decelerated from 41% (FY2021) to 24% (FY2022) toward 17% (FY2023), and the 2022 rate shock repriced every unprofitable software company. The stock collapsed to $6.29 on December 18, 2022 — 13% below the direct-listing reference price and roughly 85% below its early-2021 levels. At that moment, more than two years in, every public buyer at any point in the stock's history was underwater.

The turn came almost immediately. Q4 2022 delivered Palantir's first GAAP-profitable quarter ($30.9M of net income), and in April 2023 the company launched AIP, its Artificial Intelligence Platform, catching the post-ChatGPT wave with an actual deployed product inside government and enterprise networks. FY2023 became the first profitable year ($209.8M of net income on $2.23B of revenue), and growth reaccelerated in a way almost unseen at scale: FY2024 revenue of $2.87B (+29%) with $462M of net income, then FY2025 revenue of $4.48B (+56%) with $1.63B of net income. Government concentration — the 2020 bear case — became the moat: a U.S. Army enterprise agreement worth up to $10 billion over ten years (July 2025), a nearly $1 billion Navy contract (November 2024), and NHS England's £330 million Federated Data Platform.

Recognition followed. Despite the S-1's warning that the multi-class structure could bar index inclusion, S&P announced Palantir's addition to the S&P 500 on September 6, 2024 (the shares rose 14% the next trading day), and the listing moved to Nasdaq that November. The stock peaked at $200.47 on October 26, 2025 — up 2,665% from the reference price, with a market capitalization above $400 billion — while becoming the most expensive stock in the S&P 500 at 85 times expected forward sales; The Economist mused it might be 'the most over-valued firm of all time' at over 600 times 2024 earnings. As of August 28, 2026 it trades at $186.29, up 2,469% from the reference price: one of the best outcomes in this archive, achieved along one of its most violent paths.

Palantir arrived at the NYSE in September 2020 as a seventeen-year-old company that had never turned a profit, listing directly rather than through an underwritten IPO. Its pitch: two software platforms — Gotham, born in the intelligence community, and Foundry, its commercial counterpart — that integrate an institution's fragmented data into a single operating picture. The numbers were strange for a company its age: $742.6 million of 2019 revenue from just 125 customers, net losses of roughly $580 million in each of 2018 and 2019, and a sales model built on months-long unpaid pilots that either die or convert into deeply entrenched multi-year contracts (top-twenty customers averaged $24.8 million of revenue and 6.6 years of tenure). The filing is openly, almost proudly, anti-Silicon-Valley: the company declares its mission is to support 'Western liberal democracy and its strategic allies,' refuses to do business with the Chinese communist party, and defends its work for defense and intelligence agencies that other software companies shunned. The governance matches the ideology: a novel Class F share held in a founder voting trust gives Karp, Thiel, and Cohen up to 49.999999% of the voting power regardless of how much stock they sell. The bet offered to public investors was that government-grade data software would become indispensable to Western institutions, that pilot losses were deferred margin rather than a broken model, and that the founders deserved permanent control while they proved it.

The best case in the filing is a compounding land-and-expand machine pointed at the largest, stickiest buyers on earth. Palantir absorbs upfront losses on pilots ('Acquire'), proves value inside a customer's real operations ('Expand'), and then scales into contracts that renew and grow for years — the top three customers had been with the company an average of eight years, and contribution margin had already swung from 17% to 48% in a single year as the model matured. On the government side, a federal statute (FASA) requires agencies to buy commercial products where practicable, and the company says its enforcement had already driven a significant increase in U.S. federal business; rising national-defense, cybersecurity, and healthcare technology spending would keep pulling demand. On the commercial side, sales to one customer had historically seeded sales to similar customers across industries. If both motions kept running, the upfront-loss optics would invert into software economics at scale, and Palantir would become default infrastructure for institutions that cannot afford for their software to fail.

With hindsight: The filing's best case — pilots converting into entrenched, expanding contracts; FASA and defense spending pulling government revenue; contribution margins maturing into software economics — was not just realized but dwarfed. Revenue grew six-fold from $742.6M (2019) to $4.48B (FY2025), the company swung from $580M annual losses to $1.63B of GAAP net income, and the government alignment the S-1 defended became the centerpiece of a $10B Army agreement. What the filing never imagined was the mechanism: AIP and the post-2022 AI boom turned Palantir's ontology-and-deployment model into the market's favorite AI story, driving the stock up 2,469% versus its reference price and into the S&P 500's most expensive multiple. The only respect in which reality lagged the prospectus dream was the first two years, when decelerating growth and the 2022 selloff took the shares below the listing price.

• Existing customers reliably expand, and growing deployments within the installed base is the core growth engine.

  “We derive a significant portion of our revenue from existing customers that expand their relationships with us. Increasing the size and number of the deployments of our existing customers is a major part of our growth strategy.” source ↗

• A federal commercial-procurement mandate had already driven a significant increase in U.S. government business, with more to come.

  “The enforcement of FASA has resulted in a significant increase in our business with the U.S. federal government.” source ↗

• Winning one customer historically seeds sales to similar customers, powering expansion into new commercial verticals.

  “Historically, sales to a new customer have often led to additional sales to the same customer or similarly situated customers.” source ↗

• Government technology spending on defense, cybersecurity, and healthcare mandates is a rising demand driver for the platforms.

  “increased government spending on technology aimed at national defense, financial or policy regulation, cybersecurity, or healthcare mandates may drive customer demand at different times throughout our year” source ↗

1. Seventeen years of losses with no assurance of profitability

   partly came true

   profitability · structural

   Founded in 2003, Palantir had lost money every single year, including back-to-back annual net losses of ~$580 million, and the filing concedes it may never achieve or maintain profitability. Costs of the Acquire-Expand-Scale model are incurred upfront while revenue is recognized over multi-year terms, so growth itself deepens near-term losses.

   “We have incurred losses each year since our inception, including net losses of $580.0 million and $579.6 million for the years ended December 31, 2018 and 2019, respectively” source ↗

   What happened: Losses persisted and even deepened at first — a $520M net loss in FY2021 and $374M in FY2022 — but the terminal fear proved wrong: Q4 2022 was the first GAAP-profitable quarter ($30.9M), FY2023 the first profitable year ($209.8M), and by FY2025 net income reached $1.63B on $4.48B of revenue.

2. Class F structure locks founder control at 49.999999% of voting power

   came true

   governance · structural

   A bespoke Class F share with variable voting rights, held in a founder voting trust, guarantees Karp, Thiel, and Cohen up to 49.999999% of the vote so long as they hold a modest ownership threshold — meaning they can sell most of their stock and keep control. The arrangement can persist until the death of the last living founder, and control concentrates further among survivors if any founder exits.

   “Our Class F common stock will have a variable number of votes that, together with the Founder Voting Trust Agreement and the Founder Voting Agreement, ensure our Founders who are then party to the Founder Voting Agreement will retain up to 49.999999% of the Company’s voting power” source ↗

   What happened: The structure works exactly as designed: the founders retain effective control regardless of economic ownership, and public shareholders have no meaningful vote. It simply never mattered to returns — the controlled company delivered a 25x outcome, and Karp remained CEO throughout.

3. Extreme customer concentration

   didn't happen

   operations · structural

   With only 125 customers in H1 2020, the top three accounted for roughly three-tenths of revenue, and the filing admits significant customers have previously cut spend, terminated agreements, and even forced refunds of amounts already paid. Losing or shrinking a handful of relationships would move the whole income statement.

   “Our top three customers together accounted for 33% and 28% of our revenue for the years ended December 31, 2018 and 2019, respectively, and 31% and 29% of our revenue for the six months ended June 30, 2019 and 2020, respectively.” source ↗

   What happened: No concentration blowup occurred. The customer base expanded far beyond the 125 of H1 2020 as the U.S. commercial business scaled, revenue grew six-fold to $4.48B by FY2025, and the largest relationships (U.S. Army, Navy, NHS) grew rather than churned.

4. Dependence on government customers and public-sector budgets

   partly came true

   market · structural

   A significant share of revenue comes from U.S. and allied government agencies, exposing the business to budget cycles, continuing resolutions, sequestration, changes of administration, shifting procurement priorities, and the political perception of Palantir itself — including blowback abroad from its perceived closeness to the U.S. government.

   “We derive a significant portion of our revenue from contracts with federal, state, local, and foreign governments and government agencies, and we believe that the success and growth of our business will continue to depend on our successful procurement of government contracts.” source ↗

   What happened: The dependence persisted and arguably deepened — but it inverted from liability to moat. Instead of budget cuts, Palantir won consolidation: a July 2025 Army enterprise agreement worth up to $10B over ten years, a ~$1B Navy award in November 2024, and NHS England's £330M platform contract. The feared spending decline never came for its programs.

5. Contracts terminable for convenience; full deal value may never be realized

   didn't happen

   regulation · serious

   The majority of contracts — government and commercial — can be terminated for convenience on three to six months' notice, sometimes with pro rata refunds of deposits. The headline $1.2 billion of remaining government deal value includes unexercised options, and the company concedes it has historically not realized the full deal value of its government contracts.

   “As of June 30, 2020, the total remaining deal value of the contracts that we had been awarded by government agencies in the United States and allied countries around the world, including existing contractual obligations and contractual options available to those government agencies, was $1.2 billion.” source ↗

   What happened: No wave of convenience terminations materially damaged the business; government deal value expanded enormously beyond the $1.2B disclosed at listing, culminating in the Army's up-to-$10B enterprise agreement consolidating 75 prior contracts.

6. Direct listing with no price discovery could produce a volatile, rapidly declining stock

   came true

   market · serious

   No underwriters, no book-building, no stabilization, no over-allotment support: the opening price would be set by a market maker matching orders, historical private-market prices might bear no relation to public trading, and the filing repeatedly warns of unusual volatility — including that retail investors could have outsized influence on the price.

   “Consequently, upon listing on the NYSE, the public trading price of our Class A common stock may be more volatile than in an underwritten initial public offering and could decline significantly and rapidly.” source ↗

   What happened: Among the most accurate self-forecasts in the archive: retail buyers drove the stock from a $9.50 first close to ~$34 by late January 2021, it then fell roughly 85% to $6.29 by December 2022 — below the $7.25 reference price — before a 30x recovery to a $200.47 peak in October 2025. Volatility defined the entire holding experience.

7. Long, expensive sales cycles built on free pilot deployments

   partly came true

   growth · serious

   Sales require months of senior-management-intensive effort and no-cost or low-cost pilot deployments that may never convert; the company had a limited direct sales force and had only begun hiring dedicated salespeople in late 2018. Large enterprise and government evaluations can run past twelve months.

   “Our sales cycle often lasts six to nine months but can extend to a year or more for some customers.” source ↗

   What happened: The cost structure of the model remained real, but it stopped constraining growth: after AIP's April 2023 launch, commercial adoption accelerated and total revenue growth reaccelerated from 17% (FY2023) to 56% (FY2025) — the pilot machine ultimately converted at scale.

8. Intense competition from larger, better-resourced software and defense companies

   didn't happen

   competition · serious

   Palantir faces enterprise software giants, defense contractors, system integrators, and in-house IT departments that favor internally built tools. Larger rivals can bundle, sell at zero margin, and leverage existing relationships; the filing concedes customers may not buy even when Palantir's platforms are objectively better.

   “A significant number of companies are developing products that currently, or in the future may, compete with some or all aspects of our proprietary platforms.” source ↗

   What happened: No incumbent or startup displaced Palantir in its core government and enterprise deployments. By 2025 it had instead taken share from traditional defense contractors (the Army consolidated 75 contracts onto Palantir) and became one of the world's most valuable software companies, with a market capitalization above $400B.

9. Controversial customer relationships invite public criticism and activist pressure

   came true

   other · serious

   Work for defense, intelligence, and immigration agencies had already drawn media criticism and physical protests at company property. The filing frames a double bind: yielding to activists could alienate government customers, while refusing to yield sustains reputational attack — either path can cost customers, employees, and standing.

   “Our relationships with government customers and customers that are engaged in certain sensitive industries, including organizations whose products or activities are or are perceived to be harmful, has resulted in public criticism, including from political and social activists, and unfavorable coverage in the media.” source ↗

   What happened: The controversy never stopped — immigration enforcement work, military AI, the NHS data platform, and campus protests kept Palantir a lightning rod throughout 2020-2026. But as the S-1 implicitly bet, the criticism cost it nothing commercially: the customers being criticized kept expanding their contracts.

10. Self-imposed mission constraints, including refusal to do business with China

    didn't happen

    other · serious

    Palantir voluntarily walks away from revenue that conflicts with its stated mission of supporting Western liberal democracy, and categorically forgoes the Chinese market — a large addressable market its competitors may pursue. The company acknowledges these choices may limit growth.

    “We generally do not enter into business with customers or governments whose positions or actions we consider inconsistent with our mission to support Western liberal democracy and its strategic allies.” source ↗

    What happened: Forgoing China did not visibly limit growth — revenue compounded to $4.48B by FY2025 without it. The pro-Western positioning became a commercial asset as geopolitical tension rose, aligning the company with exactly the defense and allied-government budgets that expanded fastest.

11. Immediate share supply from expiring options and vesting RSUs

    partly came true

    financing · serious

    The direct listing came with a partial lock-up: a large tranche of shares was saleable on day one, ~55.5 million RSUs vested with the listing, nearly 180 million options faced expiration through December 2022 (forcing exercises and likely sales), and the rest unlocked shortly after full-year 2020 results — a structural supply overhang in the stock's first eighteen months.

    “179,881,132 options will expire through December 2022 if not exercised prior to their respective expiration dates, and we expect many holders will elect to exercise such options prior to expiration.” source ↗

    What happened: Early supply pressure was real: after the February 2021 lock-up release the stock slid from its ~$34 January 2021 level to the mid-$20s and kept falling through 2022, and heavy stock-based compensation diluted holders throughout. But supply never impaired the long-term outcome, which ended +2,469% versus the reference price.

12. Dependence on Alexander Karp

    didn't happen

    key person · serious

    Sales had historically depended on the direct involvement of senior management, and the filing singles out Karp — a founder and CEO of seventeen years — as difficult to replace. All executives are at-will employees.

    “We are highly dependent on the continued contributions and customer relationships of our management and particularly on the services of Alexander Karp, our Chief Executive Officer.” source ↗

    What happened: Karp remained CEO through the entire period — still running the company (and its headcount philosophy) as of 2025-26 — so the key-person loss never occurred; his idiosyncratic public persona arguably became part of the retail shareholder base's attachment to the stock.

13. High-value target for cyberattacks and data breaches

    didn't happen

    tech & security · serious

    Because its platforms process governments' and enterprises' most sensitive data, Palantir is an unusually attractive target — including for state actors — and its customers have an unusually low tolerance for security failure. The company discloses that it and its vendors have already experienced cybersecurity threats and that its defenses have not always succeeded.

    “we are perceived as an attractive target for attacks by computer hackers or others seeking unauthorized access, and we face threats of unintended exposure, exfiltration, alteration, deletion, or loss of data” source ↗

    What happened: No material breach of Palantir's platforms was publicly disclosed through August 2026 — notable for a company whose customers include intelligence agencies and whose brand would likely not have survived one. The security record became part of the sales pitch.

14. AI in the platforms could create reputational harm or liability

    partly came true

    tech & security · serious

    Unusually early for a 2020 filing, Palantir flags AI as a significant element of the business and a risk: flawed algorithms, biased datasets, and ethically controversial applications touching human rights or privacy could produce legal liability and brand damage.

    “AI is enabled by or integrated into some of our platforms and is a significant element of our business. As with many developing technologies, AI presents risks and challenges that could affect its further development, adoption, and use, and therefore our business.” source ↗

    What happened: The ethical controversies the filing warned about did continue — military and surveillance AI applications kept Palantir at the center of AI-ethics criticism. But AI proved to be the opposite of a liability economically: the April 2023 launch of AIP transformed the company into the AI era's flagship enterprise software stock and drove the FY2024-FY2025 reacceleration.

15. Multi-class structure may bar the stock from major indices

    didn't happen

    governance · serious

    Under index policies adopted in 2017, newly public multi-class companies were excluded from the S&P 500 and related indices, cutting off passive-fund demand for the shares; some institutional mandates also bar low-vote stock outright, potentially depressing the valuation.

    “Our multi-class capital structure may make us ineligible for inclusion in any of these and certain other indices, and as a result, mutual funds, exchange-traded funds, and other investment vehicles that attempt to passively track these indices would not invest in our stock.” source ↗

    What happened: Index policy softened after listing, and S&P announced Palantir's addition to the S&P 500 on September 6, 2024 — the shares jumped 14% the next trading day. The company moved its listing to Nasdaq that November and by late 2025 was the most expensive stock in the index at 85x forward sales.

16. Evolving privacy and data-protection regulation (GDPR, CCPA)

    didn't happen

    regulation · boilerplate

    A company whose product is the analysis of sensitive data sits directly in the path of GDPR, CCPA/CPRA, cross-border transfer litigation, and every emerging privacy regime; fines can reach 4% of worldwide revenue, and compliance obligations on customers can also suppress demand for the platforms.

    “Non-compliance with the GDPR specifically may result in administrative fines or monetary penalties of up to 4% of worldwide annual revenue in the preceding financial year or €20 million (whichever is higher) for the most serious infringements” source ↗

    What happened: No crippling privacy fine or GDPR action against Palantir materialized through 2026; privacy criticism attached instead to its government customers' use of the platforms (as with the NHS contract debate) without stopping the contracts themselves.

How the S-1 reads

Palantir's S-1 is the archive's purest specimen of founder-control maximalism: where dual-class filings settle for 10:1 super-voting shares, Palantir invented a third class with *variable* voting rights mathematically pinned to 49.999999%, decoupling control from ownership entirely. It is also a defining example of the 2020-21 direct-listing wave (after Spotify and Slack), with the characteristic self-diagnosing risk language — no book-building, retail price-setting, 'could decline significantly and rapidly.' The filing weaponizes ideology as differentiation: most S-1s sand off controversy, while this one leads with support for Western defense establishments and a refusal to serve China, converting reputational liabilities into customer alignment. Classic patterns present in extreme form: seventeen years of losses at listing (far beyond the typical 5-10), customer concentration paired with land-and-expand economics, an adjusted-profitability story resting on excluding stock-based compensation, and a lock-up/option-expiration structure that guaranteed heavy early float. Finally, its 2020-vintage AI risk factor — flawed algorithms, biased datasets, ethical controversy — reads as an artifact from before AI became the industry's entire narrative.

• “individual investors, retail or otherwise, may have greater influence in setting the opening public trading price and subsequent public trading prices of our Class A common stock on the NYSE and may participate more in our initial and subsequent trading than is typical for an underwritten initial public offering” source ↗

  The filing predicted, with unusual precision, the retail-driven trading that would define the stock's first year.

• “We generally do not enter into business with customers or governments whose positions or actions we consider inconsistent with our mission to support Western liberal democracy and its strategic allies.” source ↗

  The company states its ideological alignment as a business fact — the anti-Silicon-Valley posture of Karp's letter, rendered as a risk factor.

• “We do not consider any sales opportunities with the Chinese communist party, do not host our platforms in China, and impose limitations on access to our platforms in China in order to protect our intellectual property” source ↗

  A categorical refusal of the Chinese market, framed as principle rather than compliance.

• “our Founders will be able to achieve substantial liquidity in their holdings without diminishing their voting control.” source ↗

  The founders may sell most of their stock and still control the company.

• “Each of these agreements could remain in place until the death of our last living Founder. Our Founders are currently 52, 52, and 37 years old.” source ↗

  The founder-control arrangement is designed to be effectively perpetual — the filing measures its duration in founder lifespans.

• “Our culture also prioritizes customer satisfaction over short-term financial results, and we frequently make service and product decisions that may reduce our short-term revenue or cash flow if we believe that the decisions are consistent with our mission” source ↗

  Palantir tells shareholders it will deliberately sacrifice short-term financials for mission — an explicit warning that the company is not run for the quarter.

What this one teaches

• Cumulative losses at listing are a weak signal when unit economics are inflecting: the same filing that disclosed seventeen loss-making years also showed contribution margin tripling year-over-year — the leading indicator was in the S-1 for anyone who read past the net-loss line.
• The most shareholder-hostile governance structure in the archive coexisted with one of its best returns. Founder control is not an automatic discount; it is a leveraged bet on the specific founders, and it cuts both ways.
• The direct-listing risk factors were the document's most accurate forecast: with no banked price and retail setting the open, entry point determined everything. A January 2021 buyer waited roughly three years to break even; a December 2022 buyer at $6.29 made ~30x by the 2025 peak.
• Concentrated, controversial government revenue was priced as a defect in 2020 and repriced as a moat by 2024-25. When geopolitics turned, the customer list other software companies refused to serve became the hardest asset to replicate.
• Growth deceleration is not destiny: Palantir slowed from 41% to 17% and reaccelerated to 56% at multi-billion scale on the strength of one product cycle (AIP). Writing off a platform company at its growth trough is how the 2022 sellers missed the entire outcome.
• A 25x return does not mean the stock was cheap at every point: at 85x forward sales and 600x trailing earnings in late 2025, Palantir's ending valuation embeds expectations as extreme as the 2020 skepticism was — hindsight vindicating the business does not settle the price.

The paper trail

1. 2020-08-25 S-1 filing index ↗ document ↗
2. 2020-09-03 S-1/A filing index ↗ document ↗
3. 2020-09-09 S-1/A filing index ↗ document ↗
4. 2020-09-14 S-1/A filing index ↗ document ↗
5. 2020-09-18 S-1/A filing index ↗ document ↗
6. 2020-09-21 S-1/A filing index ↗ document ↗
7. 2020-09-21 S-1/A filing index ↗ document ↗
8. 2020-09-30 424B4 filing index ↗ document ↗

Filed as Palantir Technologies Inc.. All documents are public domain, served by SEC EDGAR.