Palantir (PLTR) explained: what it does, how it makes money, and why it trades like a meme stock

The standard $PLTR story is "a secretive big-data company that spies for the government." The story is half-right and a decade stale: the government work is real, but the reason the stock is a $438B company is that Palantir sold Wall Street on being the pure-play ontology layer for the agentic-AI era, and then, in 2026, actually started to grow into the pitch.

The catch sits in the tape. Palantir does not trade like enterprise software. On QuantAbundancia's 252-day correlation map its closest names are $HOOD, $COIN and $SOFI, not the SaaS complex it is nominally part of. It is priced like a story and it moves like one. This piece walks through what Palantir actually does, how it makes money, where it sits in the AI stack, and why the single hardest thing about the name is that the business and the valuation are telling two different stories at once.

What does Palantir do?

Palantir sells four products, but the mental model is two engines and a layer on top. Gotham is the original: intelligence and defense software that fuses feeds (signals, imagery, logistics, human reporting) into a single operational picture. Foundry is the commercial twin: the same data-integration spine pointed at a company's supply chain, factories, claims, or trades. AIP, the AI Platform launched in 2023, is the layer that matters now: it lets large language models run on top of that integrated data as governed agents, so a soldier or an operations manager can ask a question in plain language and get an action, not a chart.

The word that unlocks the thesis is ontology. Most "AI" software is a chatbot bolted onto documents. Palantir's bet is that the value is not the model, it is the map underneath it: a live, permissioned model of every asset, person, and process in an organization, wired together so an agent's answer is grounded in the real state of the business rather than in a PDF. Building that map is slow, expensive, and sticky. That stickiness is the entire moat, and it is why Palantir sells "bootcamps" instead of demos: get an ontology stood up inside a customer, and ripping it out later is a re-platforming project, not a vendor swap.

Who buys it: governments that cannot afford to be wrong (the US Department of Defense, the intelligence community, the US Army, the UK NHS) and large enterprises where the data is a mess and the stakes are high (BP, Airbus, Stellantis, Cleveland Clinic). It is not a self-serve tool. It is a platform you commit to.

How they make money

Two reported segments: US Government and US Commercial, plus the international remainder. Revenue is recognized from multi-year software subscriptions and usage-based deployments, and the shape of the business changed materially in 2026.

In Q2 2026 (reported 2026-08-03), total revenue grew 93% year over year to $1.94B. US revenue grew 115%, split between US commercial (+149% to $764M) and US government (+90% to $809M). Palantir raised full-year 2026 guidance to roughly $8.15B in revenue. Just as important for a name priced this richly, the profitability is real GAAP profitability, not adjusted-only: 47% GAAP operating margin in the quarter, with a trailing net margin near 49%.

The contract book is the visibility moat. Recent, named deals:

• May 2025 - DoD Maven Smart System ceiling raised to $480M (from $153M), a four-year expansion across Army, SOCOM, CENTCOM and INDOPACOM.
• Apr 2025 - NATO adopts Maven Smart System, its first alliance-wide AI deployment.
• 2025 - US Army TITAN ground-station program, where Palantir is the software prime under a Northrop Grumman ($NOC) hardware contract.
• 2024 - UK NHS Federated Data Platform, a five-year deal worth about GBP 330M, the largest commercial Foundry win to date.
• 2024 - US Special Operations Command Mission Manager, a $463M five-year ceiling.

The concentration to keep in front of you: US Government is still a majority of revenue. That is a strength (renewals are sticky and budget-backed) and the single biggest risk (procurement is political and lumpy). The bull case is that US Commercial, growing three-figures off AIP bootcamps, is the leg that turns Palantir from a government contractor into a software company. That transition is the whole investment, and it is not finished.

Where it sits in the AI stack

Palantir is the awkward one in the taxonomy, and that is the interesting part. On QuantAbundancia it is mapped to the Enterprise SaaS AI bubble, but explicitly as a singleton: it does not co-move with the SaaS trio, so it is shown as thesis-adjacent and excluded from the cluster's correlation stats. In plain terms, Palantir is classified as software but does not trade as software.

What it trades with instead is the tell. Its highest 252-day correlations are to $HOOD, $COIN, $SOFI and $AXON: three retail-momentum fintechs and a defense-tech name. That is not a coincidence of sectors, it is a signature of ownership. Palantir is a retail-beloved, narrative-driven stock whose marginal buyer behaves like the marginal buyer of Robinhood, not like the marginal buyer of an infrastructure-software index. When risk appetite is on, the whole cohort runs together; when it breaks, it breaks together. If you own Palantir for the fundamentals, the thing that will actually move your position on any given week is retail-momentum beta.

Thematically the cleaner homes are AI Applications (software productizing AI for end users) and, more distinctively, AI + Battlefield Software, the theme where Maven, TITAN and the NATO deployment live. Palantir also anchors the White House Basket watchlist, QuantAbundancia's tracker of policy-favored, national-security-adjacent names, which is the cleanest one-line summary of the government leg of the thesis.

The numbers

Read the table top to bottom and the debate resolves into a single tension. The top half is a genuinely rare business: a company adding revenue at 90%+ while running a near-50% GAAP margin, a Rule-of-40 score comfortably above 65 at a multi-billion revenue scale. Almost nothing at this size does that. The bottom half is a valuation that has already priced most of it: at roughly 70x sales and 145x earnings, the stock is not asking whether the ontology thesis works, it is assuming it works flawlessly for years and asking what else you have. That is why the same print can send the stock up 12% and then see it give it back a month later on nothing but profit-taking. The fundamentals justify a premium. The size of the premium is the open question, and no amount of contract news settles it.

The bull case

• US Commercial revenue is compounding at three-figure rates off AIP bootcamps, the leg that turns a government contractor into a software platform.
• The contract book (Maven at a $480M ceiling, NATO, TITAN, the NHS platform) gives multi-year, budget-backed visibility that most high-growth software cannot match.
• Rule-of-40 above 65 at this scale is genuinely rare: growth and real GAAP margin at the same time, not one funded by the other.
• The ontology is a structural switching cost. Once an organization runs its operations on a Palantir data model, the exit is a re-platforming, not a renewal decision.
• AIP positions Palantir as the governed-agent layer exactly as enterprises move from AI demos to AI in production, and the Anthropic and AWS integrations widen the model options inside it.

The bear case

• Valuation is the whole risk. At roughly 70x sales and 145x earnings the stock is priced for flawless multi-year execution, and any deceleration re-rates it hard.
• US Government is still the majority of revenue: procurement is political, lumpy, and exposed to budget-cycle and administration-shift risk.
• Stock-based compensation has historically run above 20% of revenue, and the dilution is a real drag on per-share economics even as GAAP margins improve.
• The AIP bootcamp-to-paid conversion rate is the unverified hinge of the commercial story; a stall there breaks the "becoming a software company" narrative.
• Insider selling has run into the billions across the founders over the trailing year, and prominent bears have made the overvaluation-plus-competition (notably from Google) case directly.
• Because it trades as a retail-momentum name, a rotation of narrative capital toward the next hot cohort (quantum, for instance) can drain the stock regardless of the fundamentals.

How to access

Palantir is US-listed (NYSE, ticker PLTR), so there is no ADR friction: it is a clean, directly buyable name for a US-retail or LLC brokerage account. To trade it that way, see /stack/ibkr; the non-resident LLC route is at /stack/us-llc.

For indirect exposure, Palantir's cleanest thematic ETF home is the ARK Autonomous Technology & Robotics ETF ($ARKQ), where it is a roughly 3.8% position. It also sits inside broad-market and large-cap-growth index funds by virtue of its size, so most diversified US-equity holders already own some Palantir whether they meant to or not.

What to watch

• US Commercial growth rate. This is the whole thesis. As long as it stays well ahead of government, the "becoming a software company" story is intact; a deceleration is the first crack.
• AIP conversion. Watch for disclosed bootcamp-to-paid conversion and net dollar retention. The bull case needs the funnel to keep converting.
• Government renewal cadence. Maven, TITAN and the intelligence contracts are the visibility moat; a renewal miss or a procurement-shift headline is the government-leg risk made real.
• The multiple, on any print. A 90%-growth quarter that still sells off tells you the bar is priced in. How the stock reacts to good news is the cleanest read on how much optimism is already in it.
• The cohort. If $PLTR breaks correlation with the $HOOD / $COIN retail-momentum bloc, the ownership base is changing, and that matters as much as any contract.

Bubble-correlation shifts and rule-based alerts on $PLTR, the kind that fire when a name breaks from its cohort or crosses a curated level, are part of /pro.

Live data on this ticker: /stocks/pltr - price, ETF holdings, bubble correlation, curated Fibonacci levels, and bot positions.

Bubble context: /bubbles/ai-software - why Palantir is mapped to Enterprise SaaS AI as a singleton, and how the cluster is moving.

QuantAbundancia is educational research. Nothing here is investment advice. See /disclosures.