Public transcript

Alex Karp Leaves Audience Speechless on Palantir's Q2 Earnings Call

By: David Carbutt
Length: 22 min
Language: English
Text from: the video's captions
Transcribed: 13 October 2026

Read the full transcript Open on YouTube

Notes in English.

Summary

This is Palantir's Q2 earnings call, featuring CEO Alex Karp celebrating 93% aggregate growth and just under 150% US commercial growth. Karp frames the results around Palantir's sovereign AI stack, warning enterprises against handing their IP and alpha to frontier model providers. He also discusses the sovereign boot camp, Palantir's outsider identity, and model lock-in risks before closing with remarks to individual investors.

Key points

Karp cites 93% aggregate growth, just under 150% US commercial growth, and 115% aggregate growth in America.
Palantir is extending the AIP stack for sovereign AI, including fine-tuning models on Nvidia infrastructure where customers own the weights and the alpha.
Karp warns that enterprises using frontier models without protection are transferring their IP, know-how and expertise to third parties who may compete with them.
A sovereign boot camp drew an unexpectedly heterodox group of executives, including older customers who now want to migrate across the stack beyond Foundry.
Karp attributes Palantir's positioning to full alignment with partners and to the company being a 'colony of believers and artists' making non-playbook bets.
Palantir already offers a product across the US government that lets customers switch out models, avoiding lock-in to any single lab.
Ryan and Shyam emphasize converting tokens to value, building benchmarks that reflect the customer's actual business, and automating model evaluation.
Karp tells individual investors their support was crucial and that Palantir will become a much larger company.

Questions it answers

00:52Q2 results and origins

What results did Palantir post and how does Karp explain them?

He cites 93% aggregate growth, just under 150% US commercial growth, 115% aggregate growth in America, a rule of 40 of 155, and 62-63% free cash flow margins, and traces the story back to building products for US government partners.

Just under 150% US commercial growth at significant scale.
Origins in serving US government partners with NLP before AI.

02:52Sovereign AI stack

What is Palantir's sovereign AI offering and why does Karp warn enterprises about frontier models?

Palantir is extending AIP for sovereign AI, fine-tuning models on Nvidia infrastructure so customers own the weights and the alpha. Karp warns that unprotected use of frontier models transfers enterprise IP and expertise to third parties who can compete against them.

Fine-tuned models on Palantir's Nvidia stack outperform frontier models.
Enterprises will either adopt this or avoid unprotected frontier model interaction.

07:16Sovereign boot camp

What happened at the sovereign boot camp?

A boot camp organized after the AI revolution drew an unexpectedly heterodox group of CEOs and operational leads, including people Palantir hadn't worked with, plus older customers who now want to migrate across the stack beyond Foundry.

Massive educational demand about contracts, open vs closed weight models, and ontology.
NDR expected to become even more positive.

12:15Why Palantir got AI right

Why did Palantir position itself differently on AI than other software companies?

Karp credits full alignment with partners, artistic insight, and outsider status, describing Palantir as a colony of believers and artists making bets in a non-playbook world.

Palantir sometimes acts against its economic interests, e.g. supporting European institutions.
Playbook-based parasitic software models don't work now.

17:08Model lock-in and switching

How does Palantir protect customers from being locked into one model?

Palantir already offers a product across the US government that lets customers switch models. Shyam adds that benchmarks often have nothing to do with a customer's business, citing vanilla Nemotron Ultra beating frontier models on real tasks, and says automation should be used to evaluate models empirically.

Lock-in enables monopoly pricing and reduced quality.
Focus is converting tokens to value with the right models for the right purposes.

21:00Closing remarks

What did Karp say to individual investors to close the call?

He thanks them, says their support was crucial, calls it one of the most exciting times to be involved in the Palantir mission, and says the company will become much larger.

The sovereign frame is described as inclusive of everyone who wants a better world.
Karp invites everyone to engage with the mission.

Notes

Results

93% aggregate growth, just under 150% US commercial growth, 115% aggregate growth in America.
Rule of 40 cited at 155; 62-63% free cash flow margins; NDR described as anomalously strong.

Origins and philosophy

Palantir began serving US government partners, working with NLP before AI existed.
Karp says Palantir rejected the Silicon Valley "parasitic" model of extracting client money without adding value.
Products built over the years: PG, Foundry, Gaia, Maven, Ontology, AIP.

Sovereign AI

AIP stack is being extended for sovereign AI: orchestrating and fine-tuning models so partners get a completely sovereign stack.
Partnership with Nvidia; fine-tuned models in classified environments outperform frontier models, and customers own the weights and the alpha.
Karp warns that unprotected interaction with frontier models lets third parties migrate your IP, know-how and expertise into their models to build a business that doesn't require yours.

Sovereign boot camp

Organized after the AI revolution exploded; drew CEOs, operational leads, and people Palantir hadn't worked with before.
Big educational demand: contracts, open vs closed weight models, how ontology works with the compute stack.
Older Foundry-only customers now want Ontology and the sovereign AI stack.

Why Palantir got it right

Karp credits full alignment with partners, artistic insight, and outsider status; calls Palantir a "colony of believers and artists."
Supporting some European institutions is against Palantir's economic interest but continues out of belief.

Model lock-in

Palantir already has a product across the US government to switch out models.
Shyam: benchmarks have little to do with customers' businesses; vanilla Nemotron Ultra beat frontier models on real tasks within 24 hours.
Focus is converting tokens to value and automating model evaluation on an empirical basis.

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