NVIDIA said on June 29, 2026 that Palantir combined NVIDIA Nemotron open models with its Sovereign AI Operating System to unveil a secure AI engine for U.S. government agencies and critical-infrastructure operators. Agencies run Nemotron on their own infrastructure, train it on their own data, and retain full ownership of the resulting model — including the weights. ASAP summarizes NVIDIA's official announcement in a direct-answer format.

What was announced

Palantir unveiled a secure AI engine on June 29, 2026 that integrates NVIDIA Nemotron open models into its Sovereign AI Operating System. The operating system comprises four components — AIP, Ontology, Foundry, and Apollo — and targets U.S. government agencies and critical-infrastructure operators first. The announcement was posted on the NVIDIA blog under Justin Boitano's byline.

What the core structure is

Owning the weights is the core of this engine. Agencies run customized Nemotron on their own infrastructure, train it on their own data, and fully retain the resulting model, including the weights that encode their operational knowledge. They can operate frontier AI inside the agency without handing data to an outside cloud.

Why air-gapped

An air-gapped environment is a secure setup completely isolated from unsecured networks. The U.S. federal government employs roughly 3 million civilian workers and handles sensitive data, so AI must be able to run on closed networks cut off from the outside for adoption to be feasible. Nemotron being an open model enables "trust through transparency," where independent review checks for vulnerabilities and biases.

Why it matters

Open-model adoption is already heading mainstream. According to NVIDIA, about two-thirds of companies already use open models, and the ability to own and audit weights lowers the barrier to adoption in the heavily regulated, security-conscious public sector. Sovereign AI that self-operates frontier models even in closed environments is emerging as a baseline requirement for government procurement.

Wrap-up

Palantir and NVIDIA unveiled a secure AI engine built on Nemotron open models on June 29, 2026. A structure where agencies run models on their own infrastructure and own the weights targets the public and infrastructure sectors that require air-gapping. It shows open models and sovereign AI settling in as standard requirements for government AI.

Source: NVIDIA Blog, "Open Models, Closed Environments: Palantir Brings Secure AI to US Agencies With NVIDIA Nemotron" (2026-06-29, by Justin Boitano).