“Please stop talking to Palantirians about how big Foundry is going to be,” Karp recalled Ted Mabrey saying. “It’s depressing, and everybody knows you’re out of touch.”

Karp spoke Thursday with Zeta Global CEO David Steinberg at Zeta Live ’26, Zeta’s annual AI marketing conference. The Palantir CEO said Mabrey, who now runs the company’s commercial business, told him the constant hype was coming off a tad aggressive for employees who had to hear him boast about Foundry constantly.

The idea stuck—albeit for a little bit: “I had stopped talking to people for like six months,” Karp said. Instead, the CEO said he has a particular way of making decisions, and that often comes at odds with what people think is possible.

“I have an esthetic idea of what could work and what won’t work,” Karp said. “About once a month, I have some idea that I think is important, and then I know it’s important. Everyone tells me we can never do that.”

‘Tell me something I want to hear’

He said he relies on colleagues he has worked with for years to tell him the truth, and that those relationships are “very harsh.”

“In fact, I often tell them, ‘Can you tell me something I want to hear, please?'” he added.

Foundry, Palantir’s platform for pulling a company’s data into one place and controlling who can see it, is now at the center of its fastest-growing business. In the second quarter, U.S. commercial revenue grew 149% year over year to $764 million, according to the company’s earnings release. Palantir raised its full-year U.S. commercial revenue guidance to more than $3.4 billion, a growth rate of at least 134%. After those results, Karp told investors “for the first time people believe us.”

Zeta is one of the companies building on it. In June, the two firms announced a deal under which Zeta’s Data Cloud will be rebuilt on Palantir’s Foundry. Steinberg told Adweek the seven-year partnership would bring in “over $100 million a year in sales” within “the next year or so.”

‘Where’s the parasite?’

During the talk, Karp explained that getting Palantir to the point where it is now took plenty of convincing, simply because investors didn’t get the business. He said Palantir struggled to raise money in its early years because it refused to build software the way investors wanted.

“Normal software is built to force you to adopt something that’s parasitic,” he said. “They take every one of those things, turn it into a commodity, and sell it to everyone else.”

According to Karp, VCs wanted to hear that clients would be unhappy but couldn’t leave. “‘Sticky’ is a metaphor for parasitic,” he said. “I get your data. I get it into something you can’t get rid of. The thing turns into something I can sell to everybody else.”

So when he pitched making special operators “more lethal” and bringing them home, Karp said, investors asked, “But where’s the parasite?”

He said Palantir was also mocked for sending engineers to work inside client companies, a model it calls forward-deployed engineering. It has since become one of the fastest-growing jobs in Silicon Valley. Steinberg said Zeta has copied it.

“We pivoted a huge part of our business to follow you guys on this because it’s so freaking efficient,” he said.

Karp said Palantir has about 4,500 employees and is “not everyone’s cup of tea.” Some of its best references, he said, come from former staff who left after “another crazy lecture from Karp” and later wanted to come back. In March, Mabrey publicly called on “former hobbits” to return to “the Shire,” making more references to its namesake from the Lord of the Rings.

Second-rate AI is costing businesses

Also during the AI-focused lecture, Karp argued the tech has changed what poor execution costs a business. In the past, a second-rate software rollout cost a company time and money but rarely sank it, since a firm with a strong moat could get by.

He used defense as an example. A German government will buy a German company’s missile “even if it’s B minus because it flies and it’s German,” he said. But a missile that can’t hit targets against Russian weapons is no longer good enough, and getting it there now depends on an AI application layer.

“Your B-minus missile is a triple-plus with this, and your A-plus missile is useless,” he said. “So that’s just binary.”

He said the same logic applies to every company. Asked what one decision the CEOs in the room should make, Karp told them to identify what makes their business special, get full visibility from the factory floor to the cost of goods, and use AI to strengthen it without letting a model company or a rival learn how it works.

“If your person down the street gets it right, even if they’re inferior to you, just like the missile, they will outperform you,” Karp said.