Palantir and NVIDIA have introduced a sovereign AI stack for complex supply-chain operations, with the first deployment running inside NVIDIA’s own supply chain.
The system combines NVIDIA Nemotron open models with Palantir Foundry, AIP and the Palantir Ontology. Organizations can post-train Nemotron on proprietary operational data, use NVIDIA cuOpt for optimization and scenario planning, and deploy the stack on premises, in colocation facilities or in cloud environments.
NVIDIA says the initial deployment is focused on materials-allocation decisions across a supply chain that spans millions of parts and thousands of suppliers. The companies describe a governed feedback loop in which operational outcomes can be used to improve specialized models while supply-chain experts retain control of final decisions.
A material part of the announcement is deployment sovereignty. The joint Palantir Sovereign AI Operating System Reference Architecture is designed to let organizations retain control of models, data and infrastructure rather than relying exclusively on a hosted frontier-model service.
The companies have not published independent performance benchmarks or quantified savings for the system. AiToolMap therefore treats the operational benefits described in the launch as vendor claims, while the architecture and NVIDIA deployment are confirmed product facts.