Palantir and NVIDIA deploy Nemotron open models in air-gapped US government systems

Palantir Technologies and NVIDIA announced on June 29, 2026 a joint system that runs NVIDIA Nemotron open models inside sovereign, air-gapped environments for US government agencies and critical-infrastructure operators, letting agencies train models on their own data and retain full ownership. The partnership, detailed by NVIDIA, pairs Nemotron with Palantir’s Sovereign AI Operating System.

The selling point is control. Agencies handling classified or regulated data have been wary of routing it through closed commercial models; this system keeps the data and the model inside the agency’s perimeter, with no call out to a vendor’s cloud. Palantir said the architecture enforces explicit data authorization, architecturally enforced isolation and full auditability, built on its AIP, Ontology, Foundry and Apollo platforms.

The framing from both companies leans on national-security stakes. Both NVIDIA and Palantir argued the design strengthens US AI leadership by giving agencies frontier-level open models they can customize and audit while keeping operational knowledge in-house, rather than ceding it to closed commercial models.

The companies disclosed no financial terms, no named agency customer and no deployment timeline, so the announcement is a product and a pitch rather than a signed contract. Palantir stock rose 3.85% intraday after the news.

Whether agencies adopt the system at scale will depend on accreditation and procurement, neither of which the announcement addressed. The security properties are Palantir’s own description and have not been independently audited.