Palantir Technologies

Every large organisation is drowning in data it cannot use: the army knows where its trucks, fuel and spare parts are, but in forty different systems; a hospital has its beds, staff and waiting lists in a dozen more. Palantir sells the software that pulls all of it into one live model of the organisation — it calls this the "Ontology" — and then lets people, and now AI agents, make decisions and take actions on it. It began by doing this for American intelligence agencies, still earns half its revenue from governments, and since 2023 has been selling the same idea, rebranded around artificial intelligence, to companies at astonishing speed. A ten-year-old could understand the pitch: we connect everything you know, so you can act on it.

The numbers are as unusual as the company. In the quarter to June 2026 Palantir's revenue was $1.94bn, up 93% on a year earlier — and the growth rate has risen every quarter for two years, which almost never happens to a company this size. It kept 85% of revenue as gross profit and 47% as operating profit under standard accounting, needs almost no capital, carries no debt and holds $9.2bn of cash. In the quarter it made more net profit ($1.06bn) than it had revenue in the same quarter a year before.

At $190.04 on 1 October 2026 the market values it at about $457 billion — some 73 times its sales of the last twelve months and about 162 times its earnings. That is the whole report in two sentences: one of the best businesses we have studied, at one of the highest prices we have seen. The question is whether the first can justify the second.

Two lessons for an owner. The first is that Palantir's business and Palantir's share price have been only loosely connected. The shares fell 87% while revenue grew; they fell 49% while growth went from 63% to 93%. An owner who cannot sit through a halving should not own this stock, whatever he thinks of the company — it has happened twice in five years, and nothing in the price today makes a third less likely.

The second is that the people who know it best have been steady sellers. A listing designed for insiders to sell, $2bn of sales by the chief executive in 2024, and about $1.9bn by insiders across 2025 and 2026 so far (Part VIII). Much of it is tax on vesting shares, and founders are entitled to diversify. But it is a fact, and it sits beside another: the company has bought back almost none of its own stock.

How the sale is made. Palantir's most effective salesman is a five-day "bootcamp", in which its engineers solve a real problem on a prospect's own data — "generally at our own expense and without a guarantee of future returns," as its 10-Q puts it. It is an expensive way to sell and a powerful one: when it works, the customer has seen the software change an operation before signing anything. Sales and marketing has nevertheless fallen from 24% to 18% of revenue in a year, because existing customers are doing the buying.

What the contracts are, and are not. The headline numbers — "up to $10bn", "remaining deal value" — are ceilings. The Army agreement is at most $1bn a year and consolidated work Palantir already had. The legally binding backlog is $4.9bn, and the company itself warns that it "allows many of its customers to terminate contracts for convenience prior to the end of the stated term." That is normal for government software; it means the revenue is renewed by performance every year, not locked in.

Buffett's test is that "a truly great business must have an enduring 'moat' that protects excellent returns on invested capital" (2007 letter). The returns are not in doubt — Palantir needs almost no capital at all. The moat is real, too: a retention rate of 157% is something customers do with their money, not something a company says. The question is endurance. The method that built it — engineers embedded in the customer — is now being copied by the richest companies in the world, and the AI labs are selling their own "context layers" to the same chief executives. We score the moat 8: wide today, and the trend at home is still widening; but it is being attacked from more directions than any moat we have scored that high.

Buffett's way of valuing a business is the one John Burr Williams set down in 1938: "The value of any stock, bond or business today is determined by the cash inflows and outflows — discounted at an appropriate interest rate — that can be expected to occur during the remaining life of the asset" (1992 letter). So instead of arguing about a multiple, we ask what Palantir would have to become for a buyer at $190 to earn 10% a year for ten years.

Start from the company's own 2026 guidance, $8.15bn of revenue. Assume it compounds for a decade; that by 2036 Palantir earns a 35–40% net margin — about Microsoft's today — and that it is then valued at 25–30 times earnings, a premium for a mature company; dilute shareholders by 2% a year, as stock-based pay has done; and discount at 10%. At 20% a year the shares are worth about $54 today. At 25%, about $112. To justify $190 you need about 32% a year for ten straight years — revenue of some $131bn in 2036, roughly twice what Oracle sells today, at margins far above Oracle's.

Is that impossible? No — and that is the honest difficulty. Palantir is growing at 93% now, on a rising curve. But very few companies in history have compounded revenue at 30% for a decade from a base of $8bn, and the price requires Palantir to be one of them with no stumble long enough to matter.

★ The Microsoft lesson, not the Cisco one. The danger at Palantir is not that it fails; on everything we can see, it is a superb business. The danger is that it succeeds and you still earn nothing for a decade, because the price had already counted the success. That is what happened to buyers of Microsoft in 1999, and Microsoft did everything right.

The pattern. On today's evidence Palantir is beating everyone it meets: US commercial +149%, against +35% at Snowflake and +80% at Databricks, while C3.ai shrinks. The competition that should worry an owner is not on that scoreboard yet. It is the AI labs and the cloud giants moving up from the model to the "layer that acts" — exactly where the Ontology sits — with balance sheets that dwarf Palantir's, and with the method Palantir invented. The irony is too good to leave out: the company that coined the "forward-deployed engineer" now competes with labs that hire them by the hundred, and calls their strategy Marxist.

Abroad, the moat runs the other way. The same American identity that wins work in Washington is costing it work in Europe, and the numbers show it: rest-of-world revenue grew 34% against 115% in the United States. A company that sells "sovereignty" to its customers is discovering that Europe wants its own.

Integrity: no veto. The disclosures are thorough, the selling is mostly pre-planned and tax-driven, and founders who built a business over twenty years are entitled to diversify. Capital allocation: excellent in the business, indifferent to shareholders. Palantir reinvests brilliantly — it needs almost no capital and grows at 93% — but returns nothing, buys back nothing of consequence, and has paid its staff in stock worth some $5.2bn since 2019, against cumulative profits of about $1.6bn. Owner mentality: divided. The founders have a tenth of the economics and half of the votes, and the shareholders paid the chief executive's antitrust filing fee. We score management and capital 5: a great operating team, run for its founders first.

Two adjustments. Palantir's favourite measure, the "Rule of 40" — growth plus margin — scored 155% in the second quarter, against 40% for a good software company. But the margin in it is the adjusted one, which leaves out $265m of stock-based pay. Buffett asked the question in 1992 and repeated it in 1998, having "not yet received an answer": "If options aren't a form of compensation, what are they? If compensation isn't an expense, what is it? And, if expenses shouldn't go into the calculation of earnings, where in the world should they go?" On the GAAP margin the score is 140% — still extraordinary, and the honest number.

The second adjustment is smaller: the 55% net margin is above the 47% operating margin because Palantir earns interest on its $9bn of cash and booked gains on investments. Neither is wrong; neither is the business.

For once, the three lenses agree. At Microsoft and Alphabet we have argued that free cash flow understates earning power because it charges growth investment as if it were upkeep. Palantir has almost no capital spending to argue about, so earnings, owner earnings and free cash flow land in the same place — between about 145 and 195 times. There is no accounting distortion hiding a cheaper company underneath. The price is simply high.

Our three paths in Part VI put a value of about $54 a share on 20% growth for a decade and about $112 on 25%, using margins as high as Microsoft's. We take $112 as the top of a fair range for one of the best businesses we have studied, and our own rule (playbook §5.8, born in our Microsoft follow-up) says that for a business of this quality the buy zone is where the price becomes fair, not where it becomes a bargain. The shares touched $106.37 in June.