Palantir posts record profits, then its CEO calls the AI industry 'Marxist'

Key points

• Palantir reported $1.9 billion in revenue for Q2 2025, up 93% year-on-year, and $1.1 billion in profit for the quarter.
• CEO Alex Karp accused frontier AI labs of migrating clients' intellectual property into their own models to build competing businesses.
• Palantir sells model-agnostic software, meaning it works across different AI systems and lets organisations keep control of their own data.
• A similar concern has been raised publicly by Microsoft CEO Satya Nadella about the relationship between big enterprises and AI labs.
• The underlying tension Karp describes is real, even if his language was extreme: several companies that funded Anthropic and OpenAI later found those labs launching services that competed directly with them.

Palantir, the data-analytics company best known for its work with governments and defence agencies, just posted one of the strongest quarters in its history. Revenue for the second quarter of 2025 hit $1.9 billion, up 93% on the same period last year. Profit came in at $1.1 billion. CEO Alex Karp noted in the shareholder letter that the firm made more profit in a single quarter than it once made in total revenue over the same period.

Then he called much of the AI industry Marxist.

What is Karp actually arguing?

Stripped of the colourful language, the point is straightforward: when a company plugs into a frontier AI lab's model, every prompt, every document and every workflow it feeds through that system teaches the lab's model something. Karp argues companies are, in effect, paying to train a competitor.

"You are paying for the right for them to migrate your IP, your know-how, your expertise to their model," he told analysts on the quarterly earnings call, as first reported by TechCrunch AI. He described this as the labs believing they are "superior" and intending to "colonise your enterprise."

His use of Marxist language carries a specific sting. Karp, who holds a PhD in social theory, wrote in the shareholder letter that frontier labs intend, "knowingly or otherwise, to capture the means of production" of their supposed partners. In Marxist economic theory, controlling the means of production means controlling who benefits from everyone else's work. The accusation is that AI labs are doing exactly that with their customers' data.

Is this a real risk for businesses?

The concern has genuine substance, even if the rhetoric is extreme. Several large companies paid to partner with or invest in OpenAI and Anthropic, only to watch those labs later launch tools in healthcare operations, legal services, design and drug discovery, markets those same partners operate in.

Satya Nadella, CEO of Microsoft, which has invested heavily in OpenAI, has raised a quieter version of the same concern publicly.

Palantir's pitch to enterprise customers is different by design. Its software, called the Palantir AI Platform, sits on top of whichever AI model a client chooses and keeps the organisation's data, prompts and outputs under the client's own control. Palantir does not train its own frontier model, the term for large, general-purpose AI systems like those built by OpenAI or Anthropic.

What does this mean for ordinary businesses?

If your company uses an AI service, it is worth reading the terms carefully. Many AI providers do reserve the right to use inputs to improve their models. Some enterprise agreements explicitly opt out of this, but the default varies.

Karp's broader point is that the AI market is moving so fast that the rules of who owns what are still being written. That does not make any one company a villain. Palantir's own record quarter shows there is plenty of commercial room for multiple approaches right now.

Common questions

Does using ChatGPT or similar tools mean my company's data trains their AI?

It depends on the plan. Consumer versions of most AI tools do use conversations to improve models by default. Business and enterprise tiers typically offer an opt-out or a contractual guarantee that your data stays private. Always check the settings and the contract.

What is 'model-agnostic' software and why does it matter?

Model-agnostic means the software works with many different AI systems rather than locking you into one provider's model. For a business, this means you can switch AI suppliers without rebuilding your whole workflow, and it reduces dependence on any single lab.

Should smaller businesses be worried about this?

The risk Karp describes is most acute for large enterprises feeding significant proprietary data into AI systems at scale. For smaller businesses using standard AI tools for everyday tasks, the practical exposure is lower, but checking your provider's data policy is still good practice.