Palantir and NVIDIA team up to deliver an AI platform for U.S. government in air-gapped settings

Palantir Technologies and NVIDIA are partnering to build an AI platform aimed at U.S. government agencies, according to a report published June 29. The effort is designed around a common hurdle in public-sector AI deployments: many agencies require systems to operate in tightly controlled networks, sometimes fully disconnected from the public internet, often referred to as “air-gapped” environments.

The companies plan to combine NVIDIA’s Nemotron open models with Palantir’s software. Nemotron refers to NVIDIA’s open-model technology, meaning the models are provided in an open format that can be run and adapted within an organization’s own infrastructure. Palantir’s contribution is intended to be the software layer that supports how agencies train, manage, and deploy AI workflows inside secure settings.

Under the planned approach, agencies would be able to train and deploy AI while maintaining network isolation. The report characterizes the target use case as supporting secure development and deployment in environments where connectivity restrictions are enforced, including air-gapped operations. In practice, that means agencies can use AI models without having them depend on continuous external access to model providers or cloud services.

The announcement indicates a continued push by major AI vendors to adapt to procurement and compliance constraints in government contracting. For NVIDIA, aligning its model technology with enterprise deployment software can make its models easier to use where security requirements shape architecture choices. For Palantir, the pairing reinforces its focus on mission software that can be installed and operated under strict governance rules.

Palantir’s platform is often described in terms of software that helps organizations operationalize data and analytics, including in regulated environments. In this case, the partnership frames Palantir as a key integration and orchestration partner, while NVIDIA supplies the model backbone through Nemotron. The report does not provide additional technical details on the exact deployment stack, such as which specific runtime components will be used or how agencies will handle model updates and governance over time.

The companies also did not provide, in the report, a clear list of which agencies are expected to pilot the system, whether the work will start as a small program before broader rollout, or the expected timeline for availability. Likewise, it is unclear from the report what contract vehicle or procurement pathway will be used, and whether existing government contracts will be expanded or whether a new agreement will be pursued.

What is disclosed points to an industry pattern: buyers want AI capabilities that can be executed in their own networks, with auditable controls, rather than relying on external, continuously connected services. Open models can help address that by allowing organizations to host models internally, while integration software can help connect data workflows and deployment operations to those models.

Looking ahead, the key question for the partnership will be execution. Agencies will likely scrutinize how the platform performs when models are run locally, how security controls are maintained during training and inference, and how updates are managed once the system is deployed. The market will also watch for whether NVIDIA and Palantir clarify partner implementation details, pilot commitments, and timelines beyond the initial description of a combined air-gapped AI offering.