Nvidia's supply chain integrates Palantir

Nvidia's supply chain will leverage Palantir as the two companies aim to expand sovereign AI and custom Nvidia Nemotron open models into more enterprises.

The news lands as Palantir kicks off its AIPCon 11 showcase.

The Palantir-Nvidia AI stack is the first being deployed in Nvidia's supply chain. The supply chain deployment combines Nvidia's Nemotron open models with Palantir Foundry and Artificial Intelligence Platform (AIP) as well as Palantir Ontology.

• Palantir, Nebius forge sovereign AI, data pact
• Palantir: Worst nightmare for OpenAI, Anthropic?
• Palantir's Q2 shines, CEO Karp says it's benefiting from LLM economics revolt
• Palantir, Nvidia bring Nemotron open models to sovereign AI

The setup aims to design a supply chain system that can offer visibility, identify constraints and continuously optimize. Nvidia's supply chain, which has to manage the 1.3 million parts in each Nvidia Vera Rubin rack, offers a strong test case for the supply chain.

Both Nvidia and Palantir are touting open models, notably Nemotron, as a way to deliver sovereign AI with enterprise control. Palantir noted that customers can train Nvidia Nemotron with their own data to recommend actions, outline tradeoffs and highlight risks so supply chain leaders can make decisions.

• Why you should arbitrage your AI models like your vendors do

The significance of the Palantir announcement is that Nvidia is confirming the company is a part of its internal supply chain stack and positioned as a data, ontology and decision layer above existing applications.

Keep in mind that Nvidia's job postings and public statements indicate that the AI infrastructure giant is a heavy SAP shop using SAP Integrated Business Planning, Demand Planning, ERP and SAP Business Network. Nvidia is also using its own AI models to plan as well as Anaplan.

Nvidia also has an internal supply chain optimization engine, Nvidia cuOpt, which uses end-to-end models for supply chain use cases. Nvidia cuOpt will be integrated with the Palantir stack.

The Palantir layer within Nvidia's supply chain will include an operational command center, AI reasoning and optimization.

Although SAP and Palantir don't collide too often that may be changing for the two vendors. Speaking at an investment conference, SAP CEO Christian Klein noted:

"Data, ontology, Business Data Cloud is super important. We don't want to give away our crown jewels, our semantics, but we want to make sure that our agents also understand non-SAP data. That the ontology layer, the semantic layer, you can build it with Palantir, but with SAP, you get a lot of that out of the box."

At AIPCon 11, Nvidia's Alex Neefus, Senior Director of Solutions Architecture, outlined how Nvidia, Palantir and Nemotron open models are revamping supply chain operations.

Key points from Neefus:

Nvidia’s biggest supply-chain challenge is optimizing constrained critical components across a massive, complex global network. The company needs end‑to‑end visibility to make better allocation decisions. "We call this the constrained material allocation problem, or CMA, and it’s the hardest problem we face in supply chain," said Neefus. "We want to optimize not just production volume, but we want to minimize the amount of time that those critical components are sitting waiting for other material to show up before we can build those boards. And we have a name for this. We call it time of ownership, or TOO. It’s the most important metric that we have."

Nvidia and Palantir built Command Center, a unified operational view that aggregates data across the supply chain to expose constraints, risk and capacity. "We think of our supply chain as going both upstream and downstream, and we want everybody to see the path of those critical components all the way from the wafer through the system to the first token produced," said Neefus.

Palantir’s ontology acts as the “brain” connecting all entities (materials, factories, sites, commits, capacity) and combining quantitative and qualitative data in a governed data layer. "The ontologies connecting materials, manufacturers, sites, commits, capacity allocations. It’s helping us put expected output versus the actual output. It all lives in the ontology in a governed data layer," said Neefus.

Nvidia started by building workflows for human planners in Foundry so their "tribal knowledge" and instincts could be captured, benchmarked, and then used to train and evaluate AI models.

Given real supply chain data was sparse and sensitive, Nvidia used NeMo Data Designer to generate synthetic and anonymized data to bootstrap model training while preserving signal. "What it lets us do is take one week’s worth of sample information and make that look like months worth of samples," said Neefus.

By post-training on supply-chain data and expert feedback, a small Nemotron 3.5 Lightning model outperformed a much larger Nemotron 3 Ultra model at lower cost and with fast retraining. "A small-purpose-trained model will outperform the predictive correctness of the biggest model, one that's 10, 20 times bigger," said Neefus. "We can retrain this model on just a single GPU in a couple of hours. This lowers the bar, meaning that we can be retraining more and more often, really as often as we like, weekly, nightly."

Nvidia's supply chain AI operating system runs entirely on premises due to security. "The hardware baseline is Nvidia's AI factory reference architecture. We worked with Palantir to port their stack and run on this opinionated version of the reference architecture," said Neefus.

Palantir's AIPCon 11 will highlight industry use cases for Palantir in agriculture, manufacturing, pharmaceutical and retail supply chains.