NVIDIA and Palantir Deploy Nemotron Open AI Models in Secure Air-Gapped Environments for US Government

Imagine running a sophisticated AI system that handles classified national security data — but the machine doing the thinking is physically cut off from the internet, locked inside a secure facility, and every piece of knowledge it learns belongs entirely to you. That is precisely what NVIDIA and Palantir Technologies are now offering to US government agencies and critical infrastructure operators through a newly announced collaboration built around NVIDIA’s Nemotron family of open AI models.

The partnership was announced via NVIDIA’s official channels and marks a meaningful step in what both companies are calling “sovereign AI” — the idea that governments and critical operators should own and control their AI systems outright, rather than depending on shared cloud services run by third parties.

What Is an Air-Gapped AI System?

For anyone unfamiliar with the term, an air-gapped system is a computer or network that has been physically and logically separated from any unsecured external network — including the internet. Think of it as a machine that cannot be reached from outside the building, full stop. These environments are standard in defence, intelligence, and critical infrastructure settings precisely because they make remote hacking or data theft extraordinarily difficult.

Running advanced AI inside such an environment has historically been a challenge, because most modern AI services rely on cloud connectivity to function. What NVIDIA and Palantir have built is designed to change that.

How the Joint Engine Works

Palantir has launched what it describes as an intelligent engine that runs NVIDIA AI and Nemotron open models inside sovereign environments. The Nemotron models are part of NVIDIA’s broader push towards open AI — meaning they can be accessed, customised, and deployed by the organisations using them, rather than being locked inside a provider’s proprietary system.

The engine sits on top of Palantir’s Sovereign AI Operating System, which brings together four of the company’s existing products: AIP (its AI platform), Foundry (data integration and analytics), Ontology (data modelling), and Apollo (deployment and update management). Together, they allow an agency or operator to train models on their own data, run everything on their own hardware, and keep full ownership of the resulting models — including the model weights, which are essentially the encoded intelligence the system develops over time.

Palantir says the offering includes explicit data authorisation, secure perimeter enforcement, customer-specific isolation, data portability, the right to erasure, and full auditability. That last point matters. In sensitive government contexts, being able to trace exactly what an AI system did and why is not a nice-to-have — it is a requirement.

Who Is This For?

The intended customers are US government agencies and US critical infrastructure operators, covering areas such as defence, intelligence, energy, transport, communications, and healthcare. Neither NVIDIA nor Palantir have published a customer list, and the scale of any current deployments has not been confirmed publicly.

NVIDIA frames the Nemotron models in these environments as supporting national security, corporate sustainability, and industrial innovation. The collaboration extends prior work between the two companies, where Palantir had already integrated NVIDIA accelerated computing into its platforms for defence and industrial customers.

Ryan Taylor, a senior figure in Palantir’s defence and government division, has previously described Palantir’s approach to sovereign AI as ensuring that “the data never leaves your environment and the model is yours.” That principle is central to what this engine is designed to deliver.

The Broader Picture

This announcement sits inside a wider competition among major AI and defence-technology suppliers to offer secure AI capabilities to governments. Countries around the world — including the UK — are actively debating how to develop AI infrastructure that keeps sensitive data on home soil and under domestic control, rather than routing it through foreign or multinational cloud platforms.

There are questions worth asking here. Some observers have raised concerns about the opacity of AI systems deployed in classified settings, where public oversight is limited by definition. Others have pointed to potential risks around algorithmic bias when powerful models are used in high-stakes security contexts. And a few voices in the technology policy space have questioned whether concentrating these capabilities in a small number of defence-technology firms raises longer-term competition and governance issues. No specific criticisms of this particular engine have been documented, but those broader debates are live and ongoing.

The design of the system — on-premises, air-gapped, owner-controlled — does align with regulatory pressures that restrict sensitive data from being processed in shared or public cloud environments. For the agencies involved, that is likely a feature, not a limitation.

What This Means for Kent Residents

This collaboration is focused specifically on US government agencies and US critical infrastructure, so there is no direct impact on Kent residents right now. But the sovereign AI model it represents — where operators own their systems, their data, and their models — is increasingly the direction UK government and public-sector bodies are moving in too. Should Palantir and NVIDIA extend similar offerings to UK entities, organisations such as NHS trusts, local authorities, or port and energy operators here in Kent could eventually find themselves evaluating this kind of on-premises, secure AI infrastructure. For now, it is a story to watch rather than one that changes anything on the ground.