The company 9/11 built — and the AI engine that outgrew it

Today marks 25 years since the September 11 attacks, and it is worth remembering that the security buildout those attacks triggered did not just buy fighter jets — it created an entire durable buyer of data software. Palantir is that buyer's favorite vendor, founded in 2003 in the immediate aftermath of the attacks with a specific pitch: intelligence agencies had scattered fragments on al-Qaeda and Osama bin Laden, but no unified view, and connecting those fragments might have prevented the next attack. CIA-backed, it quietly became the connective tissue of U.S. counterterrorism data for two decades.

The reason this anniversary is an investing question rather than a news item is that Palantir has outgrown its origin. The engine that is compounding today is not government surveillance contracts — it is commercial AI selling into U.S. businesses. And while the business is every bit as strong as the narrative suggests, the market has already priced the compounding in. That split — real operating results, fully digested expectations — is what decides whether this is a great company you should own or a great company you already missed a lot of the ride on.

A company born from a failure of data

The origin matters because it explains the moat. Palantir was built around the idea that connecting fragmented intelligence could stop another terrorist attack, and that the technology might have enabled analysts to prevent 9/11 itself. It got early funding from In-Q-Tel, the CIA's venture arm, and spent years embedding its software into the intelligence community's workflows. What it sells is essentially a data-integration layer — unifying the fractured data landscape across agencies that historically did not share information.

That position produced a reliable, government-anchored revenue base. But it also pinned Palantir's reputation and its multiple to a politically charged business. A beginner investor looking at the ticker in 2023 would fairly have asked whether the company could ever escape the surveillance-story overhang and grow beyond its defense customers.

The answer arrived in the last year, and it's the most important fact in this article: the growth engine moved to U.S. commercial.

The engine moved — and de-risked itself

In the second quarter of 2026, total revenue grew 93% year over year to $1.94 billion. But listen to the mix. U.S. revenue grew 115% year over year to $1.573 billion, and within that, U.S. commercial revenue grew 149% to $764 million — faster than the overall company by a wide margin. Management raised full-year 2026 revenue guidance to about $8.15 billion, roughly 82% growth, and lifted U.S. commercial guidance to more than $3.4 billion, about 134% growth. CEO Alex Karp described the business as compounding "at an unprecedented rate and scale."

This is the part of the story that changes the risk profile for a retail investor. The political-reputation overhang around government work does not disappear, but it is no longer the marginal driver of the stock. Enterprise AI adoption is: companies signing up for Palantir's Artificial Intelligence Platform to get LLM-driven reasoning onto their own operational data, at a pace that in the second quarter beat consensus revenue by roughly $140 million.

The economics underneath that growth are unusually clean for a company this hyped. Gross margin is around 85%, operating margin around 43%, and free cash flow margin around 51% — the company generated about $3.4 billion of trailing free cash flow while spending almost nothing on plant and equipment, because this is software, not silicon. There is effectively no net debt and return on invested capital runs in the low 30s. Whatever else you want to argue about Palantir, the revenue is reaching the bottom line, and the bottom line is being reinvested. Karp's word — compounding — is earned, not aspirational.

The price has already compounded

Here is where the analysis has to stop being about the company and become about the stock. A 93%-growth software business with 85% gross margins and no debt is genuinely rare. But Palantir trades at roughly 65 times trailing sales and about 49 times even next year's raised revenue. For scale: that sales multiple is about three times CrowdStrike's and more than three times Datadog's — high-growth software peers with their own AI rides.

And the market is not confused about what it owns. The shares are down roughly 7% this year and sit about a fifth below their 52-week high even as revenue has nearly doubled. That is the tell. A 93%-growth company whose stock fell while earnings soared means the multiple has been compressing faster than the income has been growing — the market did not need to be talked into this story, and it is not being asked to re-rate it on surprise after surprise. Guidance is already "crushing consensus"; the bar keeps moving up, which is exactly the position a thesis sits in once the market has fully digested it.

I do not see the bear case as "the growth is fake." The operating results are real and traceable. The question the evidence actually raises is the one that has always decided allocation in this trade: demand is not the issue — price and opportunity cost are. At ~49 times forward sales, the near-term return curve depends on the multiple holding, not on the earnings, and there is little irregularity left to be discovered. The long-term story of the company 9/11 built, and of the commercial AI engine it built on top of that, looks intact. Whether that intact story is still the best place for the next dollar is a separate judgment — and a much harder one at this price than it was a year ago.