The system combines NVIDIA Nemotron open models with Palantir Foundry, AIP and Ontology. Its first deployment is supporting materials allocation and supply chain decisions within NVIDIA’s own operations
NVIDIA and Palantir Technologies have announced a collaboration to deploy a sovereign AI stack for complex supply chain operations, beginning within NVIDIA’s own supply chain.
The system brings NVIDIA Nemotron open models into Palantir Foundry and its Artificial Intelligence Platform, or AIP. The models are grounded in Palantir Ontology, which represents the operational entities, relationships and processes used to support supply chain decisions.
According to the companies, the deployment is intended to improve visibility across NVIDIA’s supply chain, identify potential constraints, capture operational expertise and help teams evaluate decisions more quickly. NVIDIA will retain control and ownership of its proprietary data.
The initial application focuses on materials allocation decisions that affect how quickly parts move through the supply chain. The system is designed to help NVIDIA teams identify constraints earlier, compare alternatives and allocate materials according to their impact on end-to-end production.
Coordinating 1.3 million parts per Vera Rubin rack
Producing a rack-scale AI system requires more than securing a sufficient supply of processors. Compute, memory, networking, power, cooling and mechanical components must all be available and coordinated before a complete system can enter production.
NVIDIA says its supply chain spans millions of parts, thousands of suppliers and a global network of manufacturing partners. Each NVIDIA Vera Rubin rack contains approximately 1.3 million parts, according to the company.
The new AI stack is intended to support the process NVIDIA describes as the journey from “wafer to first token”, connecting semiconductor manufacturing and component allocation with system production and the eventual operation of the AI infrastructure.
The deployment does not automate the final decision-making authority. Post-trained Nemotron models can recommend actions, explain trade-offs and flag emerging risks, while NVIDIA’s supply chain specialists retain control of the decisions.
Models customised with operational data
Palantir customers will also be able to build supply chain AI systems by post-training Nemotron models with their own operational data through Foundry and AIP.
NVIDIA NeMo Data Libraries can be used to prepare and augment the data employed in this process. The aim is to create specialised models that reflect an organisation’s own value chain, supplier network, operating constraints and decision criteria.
NVIDIA cuOpt is integrated into Palantir AIP to support optimisation and scenario planning. Teams can use the software to model supply constraints, assess trade-offs and examine the operational effects of different allocation decisions.
The system also incorporates Palantir Autopilot with NVIDIA NeMo AutoModel and NeMo RL libraries. According to the companies, this creates a governed feedback loop in which recommendations and production outcomes can be evaluated and used to support further model improvement.
“Supply chains are the operating system of the physical economy, and AI factories are among the most complex systems ever built,” said Jensen Huang, founder and CEO of NVIDIA.
Huang said the collaboration combines Nemotron models with Palantir Ontology to support reasoning, planning and orchestration across the process extending from wafers and components to manufacturing, system assembly and customer delivery.
On-premises, colocation and cloud deployment
The NVIDIA implementation runs on NVIDIA reference architectures and the jointly developed Palantir Sovereign AI Operating System Reference Architecture, or SAIOS. The architecture is also supported by Dell Technologies and Cisco.
The term sovereign AI in this context refers to an organisation’s ability to retain control over its models, proprietary data and deployment environment. The stack can be deployed on premises using systems from Cisco and Dell, or through colocation and cloud infrastructure provided by Rackspace and Nebius.
This deployment flexibility may be particularly relevant to manufacturers and operators that cannot move sensitive operational information into a general-purpose public cloud environment.
NVIDIA and Palantir plan to apply the experience gained from the NVIDIA deployment to supply chains in manufacturing, energy, healthcare, automotive and aerospace. However, the companies have not yet provided quantitative results showing the system’s effect on production times, inventory levels, operating costs or delivery performance.
Credit: NVIDIA and Palantir Technologies
Sources
- NVIDIA technical announcement
- NVIDIA Nemotron models
- Palantir Sovereign AI Operating System
- NVIDIA Vera Rubin platform
