The strategic question for government and critical-infrastructure AI has shifted from which model is smartest to who owns the weights and where they run. Palantir and Nvidia answered on June 29, 2026, launching a joint engine that lets agencies train and deploy Nvidia Nemotron open models entirely inside sovereign, classified, and air-gapped environments. The offering pairs Nvidia compute and open weights with Palantir AIP, Ontology, Foundry, and Apollo, so a customer can fine-tune a model on its own data, keep full ownership of the result, and run it behind its own security perimeter. Palantir and Nvidia emphasize explicit data authorization, customer-specific isolation, data portability, a right to erasure, and full auditability, a control surface aimed squarely at agencies that have kept workloads on premises. Using open Nemotron weights rather than a closed API is pitched as cheaper and more controllable, with customers owning self-improving, mission-specific models. The move sharpens a sovereign-AI contest in which control and provenance, not raw benchmark scores, are becoming the deciding factors for public-sector buyers.
Nvidia shareholders elect all ten directors and ditch supermajority voting at annual meeting