Palantir Technologies and NVIDIA have deepened their push into government AI, unveiling a new intelligent engine that brings NVIDIA's open Nemotron models into the secure, air-gapped environments that U.S. federal agencies demand. Announced on June 29, 2026, and detailed on the NVIDIA blog, the partnership frames open-source AI as a tool for American "technology leadership" — letting agencies customize, train, and own their models outright rather than depend on a cloud-hosted frontier system. For more on how governments are adopting AI, see our AI industry coverage.
Open models behind closed doors
The core of the announcement is a marriage between NVIDIA's open Nemotron models and Palantir's existing data infrastructure. Palantir will use Nemotron open models to build custom frontier-quality models that serve the U.S. government, running them inside air-gapped environments — secure setups completely isolated from unsecured networks — on NVIDIA accelerated computing.
That architecture matters for agencies handling classified or sensitive information, which often cannot send data to a public-cloud AI service. With the new engine, agencies and operators can run customized Nemotron models on their own infrastructure, train them on their own data, and retain full ownership of the resulting models, including the weights that encode operational knowledge.
"Today's Palantir announcement brings NVIDIA Nemotron open models into air-gapped environments on NVIDIA accelerated computing," the companies said, positioning the combination as a way to get frontier-level capability without surrendering control of data or model weights. The emphasis on inspectability is deliberate: open models can be audited and adapted in ways that closed, API-only systems cannot, which the partners argue is a prerequisite for trust in national-security settings.
Built on Palantir's sovereign AI stack
The Nemotron models plug into what Palantir calls its Sovereign AI Operating System, a platform built on four of its core products: AIP, Ontology, Foundry, and Apollo. That stack handles the operational and data-authorization layer that makes deployment feasible in sensitive environments, providing explicit data authorization, architecturally enforced isolation, and full auditability — features that are already central to Palantir's pitch to defense and intelligence clients.
Each layer plays a distinct role. Foundry organizes an organization's data into a governed, searchable foundation; the Ontology maps that data into a digital model of how the enterprise actually operates; AIP layers AI and agent capabilities on top so models can act within defined guardrails; and Apollo manages continuous delivery and updates across distributed, even disconnected, environments. Together they are designed to let a model do useful work without bypassing the rules that govern who can see and do what.
For larger, enterprise-grade deployments, the companies said NVIDIA's AI Enterprise software suite can support production rollouts, adding optimization, reliability, and support tooling on top of the open model layer.
Why the U.S. government is the target
The partners framed the federal government as a uniquely large and complex customer. With roughly 3 million civilian employees, the U.S. government is "essentially one of the world's largest enterprises," and its operations mirror those of private-sector companies across commerce, energy, healthcare, agriculture, education, and transportation. Providing critical services across so many disciplines is complicated, and the companies argue AI can help streamline that complexity and boost productivity — from food safety to maintaining interstate highway infrastructure.
The partnership also leans heavily on a cost argument. The companies cited figures suggesting that about two-thirds of companies are already using open models and reporting on their cost efficiency, framing those savings as a reason open-weights AI can scale inside government budgets that are under constant scrutiny. Lower cost, they argue, fuels broader adoption and economic development.
The politics of open models
The timing is politically charged. Open-weights models have become a flashpoint in the broader debate over AI policy, with proponents arguing that transparency and inspectability make them safer for government use, and critics warning that openly released weights can be fine-tuned by adversaries for harmful purposes. By emphasizing "open models, closed environments," Palantir and NVIDIA are trying to occupy a middle ground: the customizability and cost advantages of open weights, combined with the security of an air-gapped, government-controlled deployment.
For Palantir, the deal reinforces its ambition to be the default operating layer for government AI, extending a franchise built on defense and intelligence contracts. For NVIDIA, it widens the reach of its hardware and software stack into the federal market at a moment when competition over who supplies AI to the public sector is intensifying. And for the agencies themselves, it offers a path to frontier capability that does not require handing their most sensitive data to a third-party API — a proposition that aligns neatly with the growing demand for technological sovereignty. Whether open-weights systems can deliver frontier performance reliably enough for mission-critical government work remains an open question, but the partnership is a clear signal that the largest AI vendors now treat the public sector not as an afterthought but as a strategic frontier in its own right.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
