Palantir CEO Alex Karp went on CNBC to call OpenAI and Anthropic, the makers of ChatGPT and Claude, 'effing insane' over their pricing and data practices, and Palantir's stock jumped 9% the same morning the company announced a partnership with chipmaker Nvidia for secure, on…
The U.S. government's most consequential AI vendor decision of the year did not arrive in a procurement memo. It arrived on CNBC on Tuesday morning, with Palantir CEO Alex Karp calling the country's leading AI companies "effing insane" and declaring himself "the voice of American business that is being channeled through me" (Forbes). The outburst was theatrical, but the market did not treat it as theater. Palantir shares jumped more than 9% the same morning — a move that came as Karp's interview and the announcement of a new Nvidia partnership to bring secure, on-premises AI to U.S. agencies landed on the same day (Stocktwits).
The wire reading of the moment is that Karp lost his temper. The structural reading is that the U.S. government, the largest single AI customer in the world, is in the middle of a quiet vendor decoupling, and Palantir, the data and AI software firm that sells to U.S. intelligence and defense agencies, is now its most visible alternative to OpenAI and Anthropic.
Karp's specific complaint, made on CNBC, was that OpenAI and Anthropic, the makers of ChatGPT and Claude, have built a pricing model that functions as a "wealth tax" on businesses while quietly harvesting customer data to improve their own models (CNBC). "Something has gone completely wrong," he said of how AI is sold. The sharper line, though, was reserved for defense and national-security buyers: "Are we really going to outsource the battlefield of this country to the consensus view in Silicon Valley? That is effing insane."
That is the line investors, defense officials, and AI competitors will remember, because it lands inside a policy backdrop that was already moving in Karp's direction. In March, the Pentagon placed Anthropic on its rare "supply chain risk" list, a designation historically reserved for foreign adversaries rather than domestic software vendors, over concerns that its models might be used for autonomous weapons and mass surveillance. In June, President Trump signed an executive order asking federal agencies to review new AI models before public release, effectively inserting the U.S. government into the AI release cycle.
Read together, those moves describe a buyer that is no longer willing to treat commercial frontier AI labs as neutral infrastructure. The Palantir-Nvidia partnership is the on-ramp for agencies that want to use modern AI without ceding data, inference control, or release timing to OpenAI or Anthropic. According to Nvidia's announcement, the architecture centers on Nvidia's Nemotron family of open models deployed inside Palantir-managed, closed environments, meaning the weights, the data, and the inference traffic stay within the agency's own perimeter. That is the structural answer to the "wealth tax" critique: customers own the model, the data never leaves, and there is no per-token meter running back to a frontier lab.
When the CNBC host told Karp "you sound pretty angry," he replied that he was not angry, he was channeling "the voice of American business that is being channeled through me," and that other CEOs would say the same thing if they were willing to say it on television (Mediaite). The 9% stock move is the cleanest test of whether investors agree. They did, at least for a day.
Two things to watch next. First, whether the Pentagon's "supply chain risk" designation of Anthropic is followed by an explicit decision to route sensitive workloads to Palantir-Nvidia rather than to any frontier-lab API. Second, whether OpenAI and Anthropic respond to the "wealth tax" framing with a pricing or data-handling concession aimed squarely at government and enterprise buyers. Neither side has treated Karp's complaint as cheap noise. The structural decoupling was already underway; Karp just made it the most expensive TV interview of the year.