Karp takes another jab at OpenAI and Anthropic: the model isn't the business, the application is

By TheStreet (via Anthropic News/Google News) · July 12, 2026.

The material available on this story is a headline and a description without the body of the article, so it is worth being cautious: we do not have the verbatim quote of the "blunt verdict" that TheStreet attributes to Alex Karp about OpenAI and Anthropic, and we are not going to invent it. What we can do, honestly, is situate the context in which it is likely framed.

Karp has for years maintained, in earnings calls and interviews, a recognizable thesis: the labs that train the large foundation models —OpenAI, Anthropic, Google DeepMind— are waging a very expensive and increasingly commoditized infrastructure race, while the real value for enterprise and government clients is captured at the application layer: whoever connects that model with an organization's concrete data, processes and workflows. It is, in essence, Palantir's commercial argument with its AIP platform and its "Ontology": we don't sell the brain, we sell the nervous system that connects it to operational reality.

If the original article follows that line, it would not be a technical critique of the performance of OpenAI's or Anthropic's models —both remain among the most capable on the market, with Anthropic consistently leading in our own quality indexes— but a critique of the business model: the suspicion that training frontier models is a bet on thin margins and commoditization over the medium term, while vertical integration in regulated sectors (defense, healthcare, banking) is where pricing is defended.

Our reading, with the caveats that come from not having the full text, is that this kind of statement is part of a positioning contest more than an objective verdict: it suits Palantir narratively for the market to perceive the big labs as providers of interchangeable infrastructure, just as it suits OpenAI and Anthropic for the opposite to be true. In general, the industry is experiencing this debate over where value accumulates —model, platform or application— as one of the most relevant of 2026, and the answer is probably not exclusive: there will be margin at all three levels, with different winners depending on the sector. What does seem to be consolidating is that not even the best model on its own guarantees market dominance if it does not solve a real deployment problem, an idea that connects with our underlying thesis: the abundance AI promises will arrive sooner through practical application than through the mere scale of the model.