Palantir has launched a Sovereign AI Operating System for deploying NVIDIA Nemotron open models in controlled environments, including air-gapped systems used by U.S. agencies and critical infrastructure operators. NVIDIA and Microsoft source material frames the move as part of a broader push to support customized en...
Palantir announced a strategic initiative with NVIDIA to deploy NVIDIA Nemotron open models in sovereign and air-gapped environments.
The announcement, published by Business Wire for Palantir Technologies, says the company’s new Sovereign AI Operating System is designed to run customized Nemotron models where organizations need tight control over data, infrastructure and model operations. Palantir said the offering is aimed at U.S. agencies and critical infrastructure operators, including environments that require isolation from public networks.
According to Palantir’s announcement, the system combines Palantir’s existing software stack — including AIP, Ontology, Foundry and Apollo — with NVIDIA Nemotron open models. The stated goal is to let customers deploy, operate and improve AI models within their own controlled environments rather than relying solely on externally hosted services.
NVIDIA’s blog described the approach as a way for agencies and operators to “run customized models on their own infrastructure” while retaining control over data and model behavior. The blog specifically highlighted support for secure and disconnected environments, a key requirement for some government and defense users.
Palantir said the system supports customer-specific isolation, telemetry-based model improvement and operational control. In practice, that means the company is positioning the product around the full lifecycle of enterprise AI deployment: preparing data, adapting models, evaluating results, applying security controls and operating the resulting systems in production.
NVIDIA’s Nemotron family is part of the company’s broader enterprise AI model and agent stack. NVIDIA Newsroom said its enterprise agent infrastructure includes Nemotron models along with other components such as NeMo-related tooling, CUDA-X skills and AI-Q. In the same announcement, NVIDIA referenced Palantir’s Nemotron-based AI field deployment engineer platform for air-gapped enterprise systems.
The important point is not only that Palantir is supporting a particular model family. It is that open models can be adapted and deployed inside environments where data movement is limited by law, contract or security policy. For regulated customers, that can be more practical than sending sensitive operational data to an external model endpoint.
Palantir’s announcement also fits a broader enterprise pattern: customers want generative AI systems that can be customized against internal data, evaluated against internal standards and governed under existing security requirements. The company is presenting its platform as a way to manage those steps in environments where sovereignty and operational control are central requirements.
Microsoft is not the issuer of Palantir’s launch, but Microsoft’s official blog provides useful context for the wider NVIDIA ecosystem. In a March post about Microsoft at NVIDIA GTC, Microsoft said NVIDIA Nemotron models are available through Microsoft Foundry. Microsoft also said its Foundry Agent Service and observability tools support production AI agents, and that Microsoft-NVIDIA infrastructure is being extended for sovereign and regulated customer-controlled environments.
That context matters because the sovereign AI market is not just about one vendor’s deployment product. NVIDIA, Palantir and Microsoft are each describing components of an enterprise AI stack: foundation models, infrastructure, agent tooling, observability, deployment controls and regulated-environment support.
The sources consistently point to the same challenge: moving from AI prototypes to operational systems in sensitive environments. Palantir’s Business Wire release emphasizes air-gapped deployment, operational control and customer-specific isolation. NVIDIA’s blog emphasizes secure AI for U.S. agencies using open models. Microsoft’s blog emphasizes production agent services and infrastructure for regulated customers.
For organizations in government, defense, energy, healthcare or other critical sectors, those requirements are often not optional. They need to know where data is stored, how models are customized, how outputs are evaluated, what telemetry is collected and who can operate or update the system.
Palantir’s launch is therefore best understood as an infrastructure and governance move rather than a consumer AI product announcement. The company is using NVIDIA’s Nemotron open models as part of a controlled deployment architecture for customers that need AI systems inside sovereign, regulated or disconnected environments.
The next test will be adoption and operational proof. The source announcements describe the capabilities and intended markets, but they do not provide customer deployment metrics or independent performance evaluations. For buyers, the key questions will be how well customized Nemotron deployments perform on domain-specific tasks, how model updates are governed in isolated environments, and how telemetry-based improvement works when data cannot leave the customer’s control.
Still, the launch reflects a clear direction in enterprise AI: large organizations are seeking AI systems that can be customized, evaluated and operated under their own security and sovereignty constraints, rather than treating model access as a simple cloud API decision.
Palantir announced a strategic initiative with NVIDIA to deploy NVIDIA Nemotron open models in sovereign and air gapped environments.
agencies and critical infrastructure operators, including environments that require isolation from public networks.
What Palantir is launching According to Palantir’s announcement, the system combines Palantir’s existing software stack — including AIP, Ontology, Foundry and Apollo — with NVIDIA Nemotron open models.
Continue reading
