What Palantir Actually Does

Imagine your company's data lives in twelve systems, one of them written in a programming language older than your parents. Nobody can answer a simple question like "where is that order right now."

That's not imagination, that's my last job.

Right. And that gap is the entire business Palantir is built on. In the quarter ending June thirtieth, twenty twenty-six, they reported about one point nine three five billion dollars in revenue, split almost evenly between government and commercial.

Okay but what are people actually buying? Because I've genuinely never been able to tell.

Four platforms, and one idea underneath them. Let's do the idea first, because it's the thing that makes the rest make sense.

My assumption, honestly, is they sell data. Like they've got a giant database of people.

No. Their twenty twenty-five annual report says it flatly: "We are not in the business of collecting, mining, or selling data." The customer brings the data. Palantir brings the plumbing on top of it.

On top of... so they don't even fix the mess underneath?

Deliberately not. Their software sits over your existing systems and reads from them. WIRED described it as a technical band-aid, and that's fair. But it's why a government agency with sixty-year-old code can buy it on Tuesday and use it.

A band-aid you pay millions for.

So here's the core concept, and it's the one word worth learning: the Ontology.

Sounds like a philosophy seminar.

It's simpler than it sounds. Your raw data is rows and columns in spreadsheets. The Ontology turns rows into objects, like a Customer or an Order or an Aircraft. Columns become properties. And connections between them are called links.

So instead of "table seven, column four", it's "this plane."

And "this plane is assigned to that mission, which is waiting on that part." Palantir's own docs say the Ontology has four pieces: data, logic, action and security. The action part is the one people miss.

Meaning what, you can press a button and something happens in the real world?

You reroute the shipment in the app, and it writes back into the source system. Most analytics tools show you a chart and stop. This one changes the record you came from.

Okay, that's a meaningful difference. A dashboard that argues back.

And security travels with it. Fine-grained access controls propagate from the source data all the way through to a shared analysis, so the nurse and the general don't see the same fields.

Alright, so where do the four products come in?

Foundry is the foundational one: data management, building the Ontology, analytics, workflow apps. Commercial customers and civil agencies. Palantir says all of its commercial customers use it.

And Gotham is the government one.

Investigations, targeting, mission planning. Militaries, intelligence services, police, immigration agencies. A former Palantir staffer told WIRED that in minutes, an analyst can map who's in a person's network and pull the documents linking them.

That's the part people get uneasy about, and I get why.

It's the same engine. Gotham just ingests crime reports, booking logs, subpoenaed records instead of inventory and invoices. The power and the discomfort come from the same place.

What are the other two?

Apollo delivers the software. It pushes updates across cloud, on-premises, classified networks, even disconnected environments, thousands of zero-downtime upgrades a day. Unglamorous, but it's why they can ship to a submarine.

And AIP is the AI one, I'm guessing.

Launched April twenty twenty-three. It connects large language models to the Ontology, plus tools to build agents and an evaluations framework for governing them in production.

Why does connecting a chatbot to your data need a whole product? I can paste a spreadsheet into one for free.

Because a model that only sees text can hallucinate a part number. A model pointed at the Ontology sees real objects with real permissions, and it can only take actions you've defined.

So the Ontology is the leash.

Try this one. You run a factory. Your maintenance logs are in one system, parts inventory in another, shift rosters in a third. A machine starts failing. What does the Ontology get you that a report doesn't?

Um. The machine's an object, linked to... the part it needs, and whoever's on shift. So I don't just see "machine broken", I see whether the part's in the building. And then I'd want to order it right there.

And that order writes back into the inventory system. Palantir sells that as a product, Warp Speed, a manufacturing operating system, first deployed in December twenty twenty-four with Anduril, L3Harris, Panasonic Energy of North America and Shield AI.

So the pitch really is "buy one thing, delete twelve dashboards."

That's it. Foundry to model and operate, Gotham for defense and intelligence, Apollo to deliver it anywhere, AIP to put models on top. And the Ontology is the common layer under all four.

And the critics' worry isn't the tech, it's that the same tool works just as well on people as on parts.

Yes. Markets like it, for what that's worth. On October twenty-ninth, twenty twenty-six, JPMorgan started coverage with an overweight rating and a two hundred sixty-two dollar price target.

Still weird that the whole empire rests on the fact that nobody's data is tidy.

If everyone's data were tidy, they'd have a much smaller company.