Palantir CEO Alex Karp attacks frontier AI labs as untrustworthy

Palantir reported $1.9 billion in second-quarter revenue and $1.1 billion in profit, and CEO Alex Karp used the moment to attack what he cast as the AI industry’s ideological drift. On an analyst call and in the company’s shareholder letter, Karp argued that leading AI labs are too untrustworthy for enterprise customers even as Palantir sells itself as the safer intermediary between models, data and decision-making.

According to TechCrunch, Karp framed his critique in unusually political language, writing that there are “Marxist overtones and undertones” to Palantir’s business and warning that some large language model builders aim to “capture the means of production” of their partners. The claim is less about slogans than about the mechanics of modern AI procurement: companies pay to use models, then feed them prompts, workflows and domain knowledge that can be turned into training data and product features. Karp told analysts that enterprises end up paying “a real cost” for what he called token spending, while also paying “for the right” for labs to migrate a customer’s intellectual property and know-how into the model.

The dispute is sharpened by the direction many frontier labs have taken. TechCrunch notes that firms that once sold model access have also moved into design tools, healthcare operations, legal work and drug discovery—areas where their customers and partners already operate. That expansion creates a familiar tension: the vendor that sees everything in the workflow can decide whether it is merely supplying infrastructure or quietly learning the business well enough to compete with it.

Palantir’s pitch is that it is “model-agnostic” and that customers can keep control of their data and AI “exhaust,” including prompts, orchestration and context. That message overlaps with a line Microsoft CEO Satya Nadella has been making: enterprises are wary of locking themselves into a single lab, and they want a layer that can swap models while keeping corporate data and processes inside the customer’s perimeter. In practice, the companies offering that layer—Palantir in software and Microsoft in cloud platforms—also gain leverage by sitting between customers and the model providers.

Karp’s ideological branding arrives alongside a balance-sheet reality: Palantir’s growth is closely tied to government and enterprise contracts, where procurement cycles reward compliance, auditability and a clear chain of responsibility. The company’s results suggest that, at least for now, many buyers are willing to pay for a system that looks less like a chatbot and more like a controlled interface for analysis and action.

Karp is warning customers about partners who may one day replace them. He is doing it from a company whose strongest quarter was built on selling AI into institutions that do not get to fail quietly.