Palantir grew revenue 85% YoY last quarter, throws off a 57% free-cash-flow margin, and trades at 68× sales. This is a deep-dive on what it does, where the money comes from, what Alex Karp is promising — and whether a Druckenmiller-style macro lens says you should own it here.
Palantir is, on the numbers, the best-performing software company at this scale in history — and the market knows it. At $156.54 the stock discounts roughly 42% annual revenue growth sustained through 2031. The bet is no longer "is the company good." It is "can it clear a bar this high, and is there any margin of safety if it can't." Our answer: own it, don't chase it.
The bull and bear here are arguing about completely different things, which is why sell-side targets span $70 to $255 — a 3.6× spread on the same asset, one of the widest in large-cap tech. The bulls (Wedbush's Dan Ives at $230, Rosenblatt at $225) underwrite an AIP flywheel compounding off a near-monopoly position in operational government AI. The bears (Jefferies' Brent Thill at $70, RBC's Rishi Jaluria at $90) see consulting-flavoured economics trading at WeWork-era multiples with $6B of insider selling since 2024.
Both are partly right. The fundamentals are genuinely extraordinary; the valuation genuinely leaves no room for error. That combination resolves to HOLD with a buy-the-pullback bias — accumulate aggressively below ~$120, trim into euphoric spikes above ~$200.
→ US commercial revenue compounding at 133% YoY on a 75% bootcamp-to-contract conversion engine; net dollar retention 150%.
→ Maven Smart System is now a DoD Program of Record across all five service branches — a multi-year, no-credible-competitor government annuity.
→ Rule-of-40 score of 145 — a class shared only with Nvidia, Micron and SK Hynix.
→ 68× EV/Sales and ~160× trailing P/E: any deceleration triggers violent multiple compression.
→ Insiders sold ~$6B since 2024; Druckenmiller and Renaissance have exited.
→ ICE / ImmigrationOS surveillance work is a live ESG and headline overhang.
Forget the spy-movie reputation. The clearest way to understand Palantir is to picture any large consumer-goods company — and the daily problem of data that lives in forty different systems that don't talk to each other. That is the problem Palantir sells against. Here is the whole thing, explained as a report rather than a brochure.
Palantir builds the single screen a large organisation runs its operations from — one place where all your scattered data is fused into a live picture of the business, and where you (or now, an AI agent) can not just see what's happening but act on it. It is sold as software, on multi-year contracts, to two kinds of customer: governments/militaries (the Gotham product) and companies (the Foundry + AIP products).
Imagine you want to answer a simple question: "Which of our products are about to stock out at retail this week, in which regions, and what should we ship to fix it?"
To answer that today, someone pulls inventory from SAP, shipment status from the logistics system, sell-through from Nielsen/retail-EPOS feeds, production from the plant systems, and stitches it together in Excel. It takes two analysts three days, the numbers are already stale by the time they're reconciled, and the "decision" is a slide deck that someone then has to manually turn into instructions for the supply-chain team.
Palantir's pitch is: what if all of that lived in one connected model, updated live, and the fix was a button you press — not a deck you email? That is the entire product, and everything else is detail.
The one piece of jargon worth learning is the Ontology. Think of it as a digital twin of your business: every real-world thing you care about exists once, as a digital object, connected to every other.
A product is an object. A distribution centre is an object. A shipment, a purchase order, a store, a supplier, a factory line — each is an object, with its real properties (this DC has 4,200 units on hand) and its real relationships (this shipment is going from that DC to those stores). Crucially, objects also carry actions — the verbs — like "re-route shipment" or "approve reorder", and those actions are wired back into the real systems. Press the button in Palantir and the truck actually gets re-routed in SAP.
That distinction — modelling the nouns and the verbs of your business, not just the data — is why Palantir keeps saying a chatbot alone is useless. A language model can write you a paragraph about your stock-outs; it can't fix them. "Your agents can go nowhere without ontology," as Karp puts it.
What it actually looks like on a screen · Foundry + AIP
Illustrative representation of the Foundry/AIP interface — not an actual Palantir screenshot. Built to show the three things that make the product distinctive: a list of live business objects (left), one object with its real data and action buttons (centre), and an AI agent that proposes an action you approve in one click (right).
There are really only two products — one for governments, one for companies — plus an AI layer on top and the plumbing that delivers it all. Same engine, different customers.
Exactly the screen above, but the objects are people of interest, vehicles, locations, sensor and drone feeds, and the actions are operational responses. This is the software behind the Army's TITAN targeting node and the Pentagon's Maven system. Getting cleared to run on classified networks takes years — which is precisely why, once it's in, it almost never gets removed.
The commercial twin of Gotham and the growth engine. Foundry becomes the "central operating system" a company runs on — one governed place where finance, supply chain and operations all work off the same live Ontology instead of 40 disconnected systems. Customers include Stellantis, GE Aerospace, BP and Airbus.
The Artificial Intelligence Platform plugs large language models into the Ontology — that's the chat box on the right of the mock. Because the agent is wired into real objects and real actions, it can do things, not just describe them. This is the product behind 133% US-commercial growth and a 150% net-dollar-retention figure.
The unglamorous part that quietly installs and updates all of the above anywhere — a classified cloud, an air-gapped Navy ship, an on-prem datacentre. It's why Palantir can serve regulated, data-sovereign customers that ordinary cloud software simply cannot reach, and it's what the new Nvidia and Dell partnerships rely on.
Palantir doesn't hand you a blank tool and a login. It sends forward-deployed engineers to sit with your team and build a working version on your actual data — through a 5-day "bootcamp" that turns a vague idea into a live use-case by Friday. Roughly three in four bootcamps convert into a paying multi-year contract, which compresses what used to be a 12–18 month enterprise sales cycle into a week.
And once your whole company is running off that single live model, the switching cost is brutal: ripping out Palantir means re-disconnecting everything you just connected. That stickiness — not the spycraft — is the real moat, and it's what the 84% gross margins and 150% net retention are telling you.
Palantir is increasingly a US story. In Q1 2026, US revenue grew 104% YoY to $1.28B — 78% of the total — while international commercial managed just 26%. The chart below shows the revenue base since IPO, split by segment. Toggle the view.
The geographic skew is the thesis and the risk in one number. US revenue compounded 75% while international grew a fraction of that. Karp has been blunt: "we really don't have the bandwidth to do anything that's difficult outside of America." Bulls read this as untapped optionality; bears read it as a business that only works in one country.
Government (~52% of mix) — Gotham-led, multi-year, sticky, accredited. Q1’26 US gov +84% YoY to $687M. Maven now a Program of Record; $13.7B+ in contract ceilings awarded since 2025. Lumpy by quarter, but structurally an annuity.
Commercial (~48% of mix) — Foundry + AIP, bootcamp-driven, faster-growing, higher-beta. US commercial +133% YoY to $595M. Trailing-12m US commercial bookings (TCV) +115% to $4.7B. This is where the multiple lives.
Remaining deal value (backlog) reached $11.8B (+98% YoY), with RPO at $4.5B (+134%) — forward visibility is improving even as growth accelerates, a rare combination.
On 4 May 2026 management lifted FY2026 revenue guidance to $7.65–7.66B — a 71% growth rate, up ~10 points from the 61% guided just three months earlier. US commercial guidance went to ≥120% YoY. This is what a "beat-and-raise" looks like when demand outstrips the company's ability to deliver.
| Metric | Q4’25 | Q1’26 |
|---|---|---|
| Revenue YoY | 70% | 85% |
| US commercial YoY | 137% | 133% |
| Rule of 40 | 127 | 145 |
| Net dollar retention | 139% | 150% |
| Adj. operating margin | 57% | 60% |
| Adj. FCF margin | 56% | 57% |
| Customers | 954 | 1,007 |
Every single line accelerated or held at a record. Growth and margins expanding simultaneously at 85% scale is essentially unprecedented in software.
"Our financial results now demonstrate a level of strength that dwarfs the performance of essentially every software company in history at this scale."— Alexander Karp, Q1 2026 Shareholder Letter
"How can a company grow 100% in the US with functionally a non-existent salesforce? We are doing what a normal company would do with 7,000 salespeople with seven people."— Q1’26 earnings call, on operating leverage
"The appearance of software working is not software working… your agents can go nowhere without ontology."— Q1’26 call, on AIP vs. commodity LLMs
"We just cannot meet demand. We are at our limit doing 100% this year in the US."— Q1’26 call, on the deployment ceiling
"Silicon Valley owes a moral debt to the country that made its rise possible… this next era of conflict will be won or lost with software."— The Technological Republic, 2025
Pentagon designates Maven Smart System a program of record across all five branches. $13.7B+ ceilings since 2025; effectively a no-bid annuity on core military AI.
Joint reference architecture on Blackwell Ultra GPUs — agentic AI for data-sovereign / on-prem customers that can't touch public cloud.
$300M BPA for the Landmark platform; delivered $4.4B in farmer assistance in its first five days — expansion beyond defence into civilian government.
Revenue $1.633B (+85%), adj. EPS $0.33 vs $0.28 est. Guidance raised. Stock initially fell 7% on a single downgrade — "priced for perfection" in action.
Dell's blowout (AI servers +757%, $51.3B backlog) validates the joint Dell–Palantir on-prem AI platform. The catalyst behind today's +9.2%.
Navy submarine modernisation ($448M), military-aviation sustainment, and a five-year Stellantis renewal — commercial AIP land-and-expand at work.
A conventional DCF spits out a fair value near $11 — useless, because the entire value is terminal. The honest question is a reverse-DCF: what growth must Palantir deliver to justify today's price? Move the sliders. The model projects FY2026 guidance of $7.66B out five years, applies a terminal FCF margin and exit multiple, discounts at a 10% WACC, and adds $8B net cash across 2.30B shares.
At these inputs, FY2031 revenue ≈ $45B and free cash flow ≈ $20B.
The market is currently pricing ~42% annual revenue growth for five straight years — at the very top of what even the bulls model.
FIG. 009-C
| Company | EV/Sales | Fwd P/E | Rev growth | Op margin* | Rule of 40 |
|---|---|---|---|---|---|
| Palantir (PLTR) | 68× | ~100× | 85% | 60% | 145 |
| CrowdStrike (CRWD) | ~25× | ~85× | 22% | 24% | 46 |
| Snowflake (SNOW) | ~15× | ~120× | 28% | 9% | 37 |
| Datadog (DDOG) | ~16× | ~70× | 25% | 25% | 50 |
| ServiceNow (NOW) | ~16× | ~55× | 21% | 30% | 51 |
| Salesforce (CRM) | ~6× | ~22× | 9% | 34% | 43 |
*Adjusted operating margin. PLTR actuals from FMP TTM data; peer multiples are approximate market estimates as of late-May 2026, labelled illustrative. PLTR trades at ~2.7× CrowdStrike's EV/Sales despite both being premier names — the premium is entirely growth-and-margin justified, which is exactly the debate.
The macro lens Stanley Druckenmiller made famous: liquidity drives markets more than earnings, ride your winners but cut hard when the thesis is priced, and never be the marginal buyer of the obvious trade. Applied to PLTR, the regime is supportive but the entry is crowded. The punchline writes itself — Druckenmiller personally owned PLTR and sold his last share at ~$88.
FIG. 009-D
Duquesne buys ~770k shares — early conviction on the AI-software thesis.
Dumps 728k shares as the stock runs. Classic Druck: take the quick win, don't overstay.
Fully exited. A 4× gain — and he then watched it run to $207.
The lesson cuts both ways. Druck's discipline (sell the priced trade) is exactly our framework. But his realised regret — leaving ~80% on the table from $88 to current levels — is the bull's best rebuttal: great compounders punish the people who sell them on valuation.
Liquidity: a 2026 easing bias and the trillion-dollar AI capex mobilisation (Dell's $51B AI backlog, Snowflake's $6B AWS deal) define a risk-on tape. Don't fight that — it lifts high-beta growth.
The trade vs. the company: Druck separates the two. PLTR-the-company scores ~85/100. PLTR-the-trade-at-$156 scores ~48 — because the edge (non-consensus insight) is gone. Everyone owns the AI-winner narrative.
Position discipline: his rule is to concentrate in asymmetric setups. At 68× sales the payoff is symmetric-to-negative near-term (bull +47% / bear −46%). That fails the asymmetry test — hence a starter weight, not a core position, until a better entry appears.
Net: the regime says own AI; the price says not this one, not here, not at size.
Strip out the index machines (Vanguard ~205M shares, BlackRock ~189M, State Street ~95M — all non-discretionary S&P 500 buying) and the discretionary picture is striking: the most famous active managers have been net sellers, even as one quant shop made a big contrarian add.
| Holder | Latest position | Recent move | Read |
|---|---|---|---|
| Stanley Druckenmiller | 0 | Fully exited (Q1’25 @ ~$88) | Sold the priced trade |
| Renaissance Tech | <7M | −85% from ~47M peak | Systematic de-risking |
| Cathie Wood / ARK | ~3.1M | Trimming; bought Apr dip | Long, cost basis ~$15 |
| Two Sigma | ~6.4M | +311% in Q1’26 | Contrarian quant add |
| Citadel (Griffin) | ~0.7M | Reduced through mid-’25 | Tactical, small |
| Coatue / Tiger Global | ~0 | Not in top holdings | Absent / exited |
13F data, Q4’25–Q1’26 cycle (as-of 31 Mar 2026 filings). Sources: Fintel, WhaleWisdom, StockCircle, InsiderMonkey, HoldingsChannel.
The tension is real but not damning: insider selling is programmatic and partly tax-driven, Thiel still holds ~55M shares, and Two Sigma's +311% add shows not everyone is bearish. Still, when Druckenmiller, Renaissance and the founders are all reducing while retail piles in via the WallStreetBets index, the burden of proof sits squarely with the bulls.
Five-year DCF, 10% WACC, off FY2026E revenue of $7.66B. The probability-weighted 12-month target lands at ~$158 — essentially today's price. That is the definition of a HOLD: the risk/reward is balanced, not skewed.
0.25×$85 + 0.50×$160 + 0.25×$230 = ~$158, a slim ~1% above spot. The asymmetry is roughly flat: you are paid the same magnitude on the upside as you risk on the downside. For a 68×-sales name, "flat asymmetry" is not good enough to chase — but it is good enough to hold and add lower.