1 hour ago
21
The US government is breaking up with closed AI models, and Palantir’s CEO wants everyone to know about it.
Alex Karp told CNBC’s Squawk Box on July 1 that multiple US government clients have started transitioning away from proprietary AI models in favor of Nvidia’s open-weight Nemotron models. The shift, Karp emphasized, is driven by a growing demand among government agencies for full control over their data, compute resources, and analytics pipelines.
What the Palantir-Nvidia deal actually looks like
The expanded partnership, formally announced on June 29, pairs Nvidia’s Nemotron family of open-weight models with Palantir’s full platform stack, including AIP, Ontology, Foundry, and Apollo. Together, the companies are calling the integration an “intelligent engine” designed to run inside sovereign and classified environments.
Government agencies and critical infrastructure operators can now customize and deploy AI models using their own sensitive data without shipping it off to some third-party cloud controlled by a closed-model provider.
Karp put it plainly during the interview. “Many of our US clients are already using these models, including multiple supporting critical US infrastructure,” he said.
The two companies haven’t disclosed which agencies are making the switch, how many contracts are involved, or the financial terms of the arrangement.
This isn’t a cold start, either. The partnership builds on a collaboration established back in October 2025, when Palantir and Nvidia first unveiled what they called a Sovereign AI Operating System Reference Architecture. That earlier effort laid the groundwork for what’s now becoming an operational deployment across classified networks.
Why open-weight models are winning the government contract game
Closed proprietary models, the kind offered by labs like OpenAI and Anthropic, typically operate on token-based pricing structures. You send data in, you get results back, and you pay per use. For a defense agency processing classified intelligence, it’s a non-starter.
Karp was notably critical of those token-based pricing models during his CNBC appearance, arguing they simply don’t meet the operational requirements of enterprise and government customers.
Nvidia’s Nemotron models sidestep this problem entirely. Because they’re open-weight, meaning the model parameters are freely available for modification and deployment, organizations can run them on local hardware without external dependencies.
What this means for investors watching the AI arms race
For Palantir shareholders, this partnership reinforces the company’s positioning as the connective tissue between raw AI capability and the highly regulated environments where that capability actually matters.
For Nvidia, open-weight model adoption in government settings drives demand for the company’s hardware, and it validates Nvidia’s strategy of investing in open-source AI models as a business development tool. Every government agency that deploys Nemotron is another customer locked into Nvidia’s silicon ecosystem.
OpenAI and Anthropic have both been pursuing government contracts aggressively, but their closed-model architectures face inherent friction in classified environments.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.
English (US) ·