Palantir and NVIDIA are shipping AI for US agencies that runs isolated from the internet, with weights under the agency's own control. If the data can't go out - the model comes to it.
- NVIDIA and Palantir are shipping the Nemotron models for US agencies and critical infrastructure, incl. air-gapped environments (June 29).
- Agencies keep their own weights and train on their own data; sensitive material doesn't leave the perimeter.
- Built on Palantir AIP, Ontology, Foundry and Apollo: the model goes to the data, not the other way around.
The data doesn't go anywhere. The model comes to it. NVIDIA and Palantir announced a solution where the open Nemotron models (NVIDIA's family of open models) run inside Palantir for government and critical infrastructure, including in air-gapped environments cut off from the internet. Agencies keep their own weights and train on their own data.
What matters here goes beyond government. This proves serious AI can live in isolation, under the control of whoever uses it. For a long time the equation was simple - upload everything to someone's cloud and trust them. Palantir and NVIDIA flip it. The model comes down to the data, the door closes, the key stays with the agency. For you, if you carry sensitive data, isolated AI is now a real option, not a dream.