Palantir CEO Karp trashes OpenAI, Anthropic token model on CNBC

The AI-rant-plus-partnership-launch is a familiar CEO move, but Alex Karp turned it into something louder than usual on Wednesday. According to Forbes, the Palantir CEO went on CNBC to unveil an expanded partnership with Nvidia and then spent the segment attacking OpenAI and Anthropic hard enough that co-anchor Becky Quick interrupted with 'You sound pretty angry.' Karp's answer: 'This is the voice of American business that is being channeled through me.'

The specific charge was that token-based pricing is a bad deal for enterprises. Karp said 'something has gone completely wrong,' accused frontier labs of 'stealing the weights and alpha of my business,' and called it 'a wealth tax that does not help the poor, it just punishes.' His other line, aimed at national security, was that outsourcing 'the battlefield of this country to the consensus view in Silicon Valley' was 'effing insane.'

Behind the rhetoric was a product pitch. Palantir announced an intelligent engine that runs Nvidia's Nemotron open models on its own AIP, Ontology, Foundry and Apollo stack, aimed at U.S. government agencies and critical-infrastructure operators that can't route sensitive data through commercial cloud providers. Karp's framing is that customers keep their data and model weights inside their own perimeter; Nvidia's Jensen Huang added that open models 'can strengthen America's leadership in AI.' Markets liked the package: PLTR jumped roughly 9% to $127.31 in midday trading, leading a broader AI-software rally.

The honest caveat is that Karp has an obvious commercial interest in reframing pay-per-token AI as extractive, since his own product is the alternative he is selling. The reporting doesn't quantify what enterprises actually pay versus what they claim to get back, and it doesn't detail pricing, latency or model-update cadence for the Nemotron engine against frontier competitors. That is the part to watch. If Palantir can turn the sovereign-deployment pitch into a real budget line at agencies wary of the OpenAI or Anthropic stack, the token-vs-weights argument becomes something more durable than a soundbite.