Nvidia and Palantir announced on September 10, 2026 that they had developed a joint “sovereign AI” software and model stack for critical supply chains. The system combines Nvidia’s open Nemotron models with Palantir Foundry, Artificial Intelligence Platform and the Palantir Ontology. The companies said the first deployment had begun inside Nvidia’s own global supply chain. Nvidia therefore becomes both a technology provider and the initial reference customer for the combined platform.
The stated goal is to identify production constraints earlier, turn operational knowledge into software and accelerate decisions. According to Nvidia’s official announcement, the system will evaluate information about compute, memory, networking, power, cooling and mechanical components used in AI infrastructure production. Palantir’s Ontology layer is designed to represent relationships between different data sources and physical operations in a common model, while Foundry and AIP provide the operational software environment.
The first use case focuses on the supply chain for Nvidia’s Vera Rubin rack systems. Nvidia said a single rack contains about 1.3 million parts and requires coordination across thousands of suppliers. The figure comes from the company’s release and is also reported by Fast Company and Barron’s. The platform is intended to provide greater visibility into component availability, manufacturing stages and the dependencies that can slow delivery of complete systems.
Nemotron models will operate alongside enterprise data inside Palantir Foundry and AIP. The companies say customers will retain control and ownership of proprietary information. In this context, “sovereign AI” refers to organizational control over where data is processed, where systems are hosted and who manages them. It does not mean that the software makes every operational decision independently or replaces human responsibility for production planning.
Deployment flexibility is another part of the announcement. The stack can run on premises, in colocation facilities or in cloud environments. Nvidia said Cisco and Dell can support on-premises configurations, while Rackspace and Nebius are among the hosting and cloud options. The companies are positioning the architecture for industries with complex or sensitive operations, including manufacturing, pharmaceuticals, retail, agriculture, technology and government.
Fast Company reported that Nvidia is turning its own supply chain into the proving ground and reference deployment for the joint product. The Next Web described the project as a combination of Foundry, Ontology and Nemotron models within one supply-chain layer. Barron’s characterized it as an expansion of the companies’ existing AI partnership into industrial operations. These independent reports align with the central product and deployment claims in the official release.
The companies did not disclose the financial size of the arrangement, a confirmed customer count or measured efficiency gains. It is therefore too early to state exactly how much the system will reduce costs or shorten production schedules. Some claims in the announcement describe intended future benefits and should not be confused with completed performance results. The currently verifiable milestone is that the combined stack has begun deployment within Nvidia’s own operations.
The companies plan to demonstrate the technology in greater detail at Palantir’s AIPCon 11 event. Future deployments will provide evidence about integration time, decision-support accuracy and performance across organizations beyond Nvidia. For now, the confirmed development is the integration of Nemotron models with Palantir’s software layer and its initial use in Nvidia’s complex supply chain. Nvidia’s release and coverage from Barron’s, Fast Company and The Next Web support that core account.
