Public transcript

Palantir CEO Alex Karp says 'something has gone completely wrong' with how AI is sold

By: CNBC Television
Length: 8 min
Language: English
Text from: the video's captions
Transcribed: 13 October 2026

Read the full transcript Open on YouTube

Notes in English.

Summary

In this CNBC interview, Palantir co-founder and CEO Alex Karp discusses the company's new NVIDIA partnership and argues that something has gone completely wrong with how frontier AI labs sell large language models to enterprises. He contends that clients are paying for tokens without getting value while risking the transfer of their data, IP, and business alpha. Karp positions Palantir's ontology application layer and agnostic model approach as the solution that lets customers control their compute, models, weights, and data.

Key points

Alex Karp says the NVIDIA partnership grew out of questions about who controls models, weights, and the value of the business.
Palantir's ontology application layer makes LLMs safe and useful by preventing them from touching, caching, or transferring client data and IP.
Karp claims enterprises are unhappy with frontier labs because they pay for tokens, get no value, and risk losing their business alpha.
Palantir now sells an agnostic product that lets customers switch from model to model.
Karp argues value comes from the model plus the application layer plus compute, and that ontology and compute are where Palantir makes money.
He claims Palantir can bring an open model to frontier-level performance while clients control the weights.
Karp says trust must be rebuilt around questions like who owns the data, where it is cached, and whether prompts are secure.
He calls the frontier models irresponsibly oversold and says the country's battlefield cannot be outsourced to Silicon Valley's consensus view.

Questions it answers

00:00NVIDIA partnership

How did the NVIDIA deal come together?

Karp says the partnership arose from technical questions about who controls the models, weights, and value of the business, combined with client discomfort and loss of trust toward the frontier labs.

Karp cites technical issues around who controls models, weights, and business value.
Palantir sits on critical infrastructure in America, Ukraine, and Israel.
Everyone using LLMs on the battlefield runs on Palantir's ontology.

01:20The ontology layer

What is Palantir's ontology and why does it matter?

Karp explains that large language models need an application layer to be valuable in battlefield, regulated, or manufacturing contexts. Palantir's ontology makes LLMs safe and useful by preventing them from touching underlying data, caching data, or transferring IP.

Ontology makes LLMs safe, useful, and precise.
It prevents the model from caching data and replicating the client's business.
Everyone is now copying the ontology concept.

02:50Criticism of frontier labs

What has gone wrong with how AI is sold?

Karp says enterprises believe they are wasting time and money on tokens, getting no value, and handing over their IP. He argues the models have been irresponsibly oversold, which he calls dangerous for everyone.

Karp says enterprises feel they get no value and lose their IP.
He claims enterprises privately trust and love Palantir.
He asks why labs charge for tokens if the value is so high.

03:20Control and trust

What do customers actually want from AI providers?

Karp says technical customers want control over their compute, models, data stack, and alpha — they want to own the means of production. Palantir now sells an agnostic product that lets customers switch between models, and trust must be rebuilt around basic questions like who owns the data and where it is cached.

Customers want to own the means of production.
Palantir's product lets customers switch from model to model.
Trust requires answering who owns the data and where it is cached.

04:50Open models and true cost

Can an open model compete with frontier models?

Karp claims it is the model plus an application layer plus compute that creates value, and that Palantir can take an open model in a classified or non-classified context and get it to the level of a frontier model while the client controls the weights.

Value comes from model plus application layer plus compute.
Ontology and compute are where profit and free cash flow come from.
Palantir can bring an open model to frontier-level performance.

06:20Anger at overselling

Why is Karp so angry about the AI market?

Karp says he is channeling the voice of American business, claiming enterprises are livid that they pay for tokens that create no value while their business alpha is at risk. He argues the country's battlefield cannot be outsourced to the consensus view in Silicon Valley and that the models have been irresponsibly oversold.

Karp says enterprises are livid about paying for tokens with no value.
He argues the battlefield cannot be outsourced to Silicon Valley consensus.
He calls the models irresponsibly oversold and dangerous.

Notes

Overview

Alex Karp, Palantir's co-founder and CEO, sat down with CNBC to discuss the NVIDIA partnership and his critique of how frontier AI labs sell large language models.
He argues that "something has gone completely wrong" with AI sales to enterprises and government.

The NVIDIA partnership

The deal arose from technical questions: who controls the models, the weights, and the value of the business.
Palantir sits on critical infrastructure across America, Ukraine, and Israel; everyone using LLMs on the battlefield runs on Palantir's ontology.
What aligns Palantir with NVIDIA is giving technical customers control over their compute, models, data stack, and alpha.

The ontology application layer

LLMs need an application layer to be valuable in battlefield, regulated, or manufacturing contexts.
Palantir's ontology makes an LLM safe, useful, and precise: it doesn't touch underlying data, prevents the model from caching data and replicating the business, and doesn't transfer IP.
Karp notes everyone is now copying the ontology concept, which was once dismissed.

Critique of frontier labs

Karp says enterprises feel they are wasting time and money on tokens, getting no value, and losing their IP and business alpha.
He asks why labs charge for tokens if the value is so high — suggesting a revenue-share would make more sense.
He claims enterprises privately trust and love Palantir, especially those in critical infrastructure, public and private.

Palantir's answer

Value comes from the model plus the application layer plus compute; ontology and compute are where Palantir makes profit and free cash flow.
Palantir sells an agnostic product that lets customers switch from model to model.
Karp claims Palantir can take an open model, in classified or non-classified contexts, to frontier-model performance while the client controls the weights.

Rebuilding trust

Trust requires answering basic questions: Who owns the data? Where is it cached? Are prompts secure? Is IP being transferred?
Karp says enterprises are livid and that the models have been irresponsibly oversold, which he calls dangerous for the country.

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