Public transcript

Alex Karp on Why Palantir Is WINNING | Zeta Live '26

By: More Future
Length: 25 min
Language: English
Text from: the video's captions
Transcribed: 13 October 2026

Read the full transcript Open on YouTube

Notes in English.

Summary

Alex Karp, CEO of Palantir, sits down with a Zeta executive at Zeta Live to discuss why Palantir is winning in AI. He argues the AI revolution is uniquely split between America and China, that AI has made software advantage binary, and that companies must protect their unique "alpha" through data sovereignty rather than parasitic software models. He closes with advice for CEOs: know what makes your business special, get full visibility across your stack, and shorten execution timelines because delays signal something parasitic.

Key points

Karp argues the AI revolution is uniquely concentrated in America and China, with original research done in America and derivative advancement in China.
He says we are in both the first and second phases of AI: easy use cases work, but complex ones require an application layer with a model.
Karp uses a missile example to argue AI is now binary: a B-minus missile with the right AI layer beats an A+ missile without it.
He explains data sovereignty as augmenting your unique capabilities with models trained on your data without exposing your alpha to others.
Karp says Palantir couldn't get VC financing because VCs wanted a parasitic, sticky model, while Palantir wanted to make clients happy.
Forward deployed engineers let Palantir bespoke-build products for each enterprise and train clients' people to be their own FDEs.
Karp's advice to CEOs is to identify what makes their business special and get total visibility up and down the stack.
He argues anything taking too long signals something parasitic in the stack: wrong person, wrong part of the stack, or wrong technology.

Questions it answers

00:00AI's unique geography

Why is the AI revolution different from past technological revolutions?

Karp says the AI revolution is uniquely split between America and China, with original research done in America and derivative advancement in China, unlike past revolutions that produced multiple national versions.

The first wave of AI happened in the military, partly classified.
There is no German or French version of AI, only America and China.

02:18Two phases of AI

What are the first and second phases of AI?

He says we're in both phases: easy use cases work with off-the-shelf models, but complex ones need an application layer with a model — his missile example shows a B-minus product with AI beats an A+ product without it.

Companies risk outsourcing their alpha to model companies in phase one.
The delta between good and second-rate software is now determinative.

06:57Data sovereignty and alpha

What does data sovereignty mean for a business?

Sovereignty means augmenting your unique capabilities with models trained on your data without exposing your alpha, sharing only what you choose at every layer of the stack.

Free cash flow depends on uniqueness that supports pricing power.
The free market drives pricing parity within about 18 months.

09:14The non-parasitic model

Why did VCs refuse to finance Palantir?

Because VCs wanted a parasitic, sticky model where clients can't leave, while Karp insisted on making clients happy — a stance that left Palantir unfunded and mocked for 17 years.

Palantir couldn't win on popularity, so it won on alignment.
Ex-Palantirians who stormed out are the company's best references.

11:51Forward deployed engineers

How do forward deployed engineers work?

Forward deployed engineers are highly technical people who understand both the product and the client's enterprise, enabling bespoke builds and training clients' own people to become their own FDEs.

Palantir has 4,500 people and won't scale to 50,000, so it works through partners.
Palantir manages uniquely shaped, brilliant, often dyslexic people.

20:59Advice for CEOs

What should CEOs do before they go home?

Identify what makes your business special and not quite understood, get complete vision from factory floor to cost of goods, and enhance it with AI without letting competitors or model companies learn exactly how you do it.

AI makes real things 10 times more real, so there's no room for error.
Anything taking too long signals something parasitic in the stack.

Notes

AI's Unique Geography

Karp notes the AI revolution is unlike past ones: instead of German, French and American versions, there is only America and China.
Original thinking and research happen in America; China does derivative advancement as a fast follower.
The first wave of AI happened in the military, partly classified, visible in places like Ukraine.

First and Second Phases of AI

We are in both phases at once. In the first, you hand data to a model, get value on easy use cases, and risk outsourcing your primitives and alpha to a model company.
The second phase is binary: Karp's missile example — a B-minus missile with an application layer plus a model becomes "triple plus," while an A+ missile without it is useless.
Europe isn't scared of AI and isn't adopting it on the battlefield because people can still point at things that don't work.

Data Sovereignty and Alpha

A business is something that will eventually produce high-quality free cash flow because it has something unique that supports pricing power.
Karp says the free market drives pricing parity within roughly 18 months if competitors can replicate any part of your stack.
Foundry brings data together and protects it granularly, so you share what you want and keep the rest sovereign; the application layer does the same with open-weight or closed-weight models.

Why VCs Rejected Palantir

VCs wanted a parasitic, sticky model: lock in clients, commoditize their data, sell it to everyone. Karp refused, saying Palantir would make clients happy.
Palantir spent 17 dark years being mocked, but everyone conceded they had the best people — still true, with recruiting "crushing it."
Forward deployed engineers extend the unique nature of each enterprise into the product; Palantir trains clients' people to become their own FDEs rather than being a body shop.

Advice for CEOs

Identify what makes your business special and not quite understood, and get complete vision from factory floor to cost of goods.
AI makes real things 10 times more real, so there's no room for error; American corporate leaders are on the front line.
Shorten timelines: anything taking too long signals something parasitic — wrong person, wrong part of the stack, or wrong technology. This is a way for non-technical executives to pentest their business intuition.

Text from the video's own captions. Summary and notes written by AI from the transcript, so check anything important against the recording.