Palantir and Nvidia’s Nemotron: A Real Signal Inside a Self-Interested Claim

Alex Karp, Palantir's CEO, has announced that US government customers are abandoning proprietary AI models from Anthropic in favour of Nvidia's open-weights Nemotron. Because nothing says 'independent validation' like a CEO of a company that just launched a Nemotron deployment platform the very same week. The timing is impeccable—if your definition of impeccable includes 'so transparently self-serving it could double as a press release.'

Open-weights models are suddenly the darling of defence and intelligence, apparently. Never mind that Nemotron is still an Nvidia product with all the corporate entanglements that implies; the label 'open' is what matters. Karp's claim suggests government customers are fleeing Anthropic's proprietary models—perhaps because they've realised the chatbot's opinions on defence spending are less than hawkish. Or perhaps because Palantir has a platform to sell. The charitable interpretation is that government agencies woke up one morning and decided they'd rather trust a model whose weights are public but whose safety testing is even more opaque than the classified projects it might serve.

This is the kind of news that gets libertarian futurists excited: the government embracing open-source AI! But in practice, Palantir deploying open-weights models for government use means the same surveillance-adjacent company gets to pick which open model to put its stamp on. The 'shift away from proprietary' is a shift towards a different flavour of vendor lock-in, where the model is open but the integration platform is decidedly not. Karp's claim may be a real signal of government intent, but it's also a textbook example of turning a self-interested product launch into a narrative about market evolution. Cheers, Alex—the pint's on the taxpayer.