Palantir is benefiting from the AI boom, but its CEO is also using that boom to draw a sharper line between his company and the frontier labs building large language models.
After Palantir reported a record second quarter, CEO Alex Karp argued again that enterprises should be wary of how they work with AI labs. His message was not that AI demand is slowing. It was that the structure of some AI partnerships may leave companies paying for tools while helping future competitors learn from their operations.
A record quarter gave Palantir a bigger platform
The timing matters. Palantir reported $1.9 billion in revenue for its second quarter, up 93 % over the year-ago quar ter. The company also reported $1.1 billion in profit.
Karp framed the profit figure in stark terms, saying it was "more profit in a single quarter than we did in total revenue in the same period the year before." In other words, Palantir is not making this argument from a position of weakness.
The growth of AI use has helped Palantir, according to the source article. The company sells model-agnostic AI and analysis software to governments and enterprises. That positioning lets Palantir present itself as a layer that can work across models, rather than as a frontier lab trying to own the model relationship itself.
That distinction is central to Karp's argument. Palantir says its approach lets organizations control their data as well as their AI "exhaust," described in the source as prompts, orchestration and context. For enterprises, that framing turns AI governance into a question of business control, not just technical performance.
Karp's warning is about control, not demand
In a quarterly shareholder letter, Karp wrote that "There are Marxist overtones and undertones to our business." He then aimed his criticism at companies building large language models, saying that others "intend, knowingly or otherwise, to capture the means of production of their purported partners."
The language is intentionally provocative. But the business issue underneath it is more practical: when a company uses an AI lab's tools, what happens to the knowledge created through that use?
Karp's concern is that enterprises may transfer more value than they realize. The source article describes his view as a warning that businesses could be paying for access while also allowing AI labs to absorb intellectual property, know-how and expertise. In that framing, the risk is not simply that an outside vendor provides software. The risk is that the vendor learns enough from customers to build products that compete with them.
During the quarterly conference call with Wall Street analysts, Karp extended that argument with language tied to defense technology and national competition. He questioned whether companies want to buy into a future where their work helps adversaries win, while a small group captures the means of production.
His remarks also linked enterprise AI adoption to moral confidence among AI builders. He argued that some AI companies believe they are superior to the enterprises they serve and therefore entitled to absorb what those enterprises know.
The enterprise AI question is changing
The source article notes that Karp's broader point is appearing elsewhere, including from Microsoft CEO Satya Nadella. The concern is not limited to Palantir's competitive messaging.
Many companies have partnered with or paid Anthropic and OpenAI while AI labs have launched similar businesses. The source lists areas such as design tools, healthcare operations, legal and drug discovery. That pattern is what gives Karp's warning its practical force.
For an enterprise, the key issue is not whether AI tools are useful. Palantir's own results show that AI demand can support major growth. The issue is how companies evaluate the bargain they are making when they bring outside AI systems into workflows that contain proprietary context.
That bargain can include several layers:
- Data control: who has access to the information used in AI workflows.
- Prompt control: whether the instructions and interactions with AI systems remain under the organization's control.
- Context control: how business-specific processes and expertise are handled.
- Strategic control: whether a vendor relationship could later support a competing business.
Palantir's pitch is that organizations should keep control over those layers while still using AI. Its model-agnostic position is part of that pitch, because it separates enterprise AI deployment from dependence on one model builder.
No simple villains in a fast-moving market
The source article ends on a more balanced point: these companies are not simply economic villains or heroes. They are for-profit companies operating in a fast-changing market.
That matters because Karp's language can overshadow the underlying commercial reality. AI labs are trying to build large businesses around powerful models. Enterprises are trying to use those systems without giving away too much of what makes their own businesses valuable. Palantir is trying to sell software into that tension.
The result is a new kind of vendor risk conversation. Traditional enterprise software questions focused on cost, security, integration and reliability. AI adds another question: whether the system improves the customer's business while also helping the vendor understand enough to move into the customer's market.
Palantir's second-quarter results show there is strong demand for AI software across governments and enterprises. Karp's comments show that the next stage of enterprise AI may be less about whether companies adopt AI and more about who controls the knowledge created when they do.