Nvidia has built a Digital Supply Chain Intelligence command center on Palantir Foundry to pull material, capacity, and qualitative signals into one place for weekly factory allocation, the company said in a Sept. 10, 2026 technical blog as Palantir’s AIPCon 11 conference opened.
The setup sits beside Nvidia’s existing SAP and Anaplan tools rather than replacing them, according to Constellation Research’s account of remarks by Nvidia’s Alex Neefus. Allocation is framed as a mixed-integer problem that Nvidia’s cuOpt library solves to minimize “Time of Ownership” — how long scarce parts wait before boards can be built.
Nvidia said back-tests showed human planners still beat the pure math by using emails, weather, geopolitics, and supplier debriefs the solver could not see. It did not publish a numeric gap. To capture that judgment, Nvidia said it post-trained an open-weight Nemotron model on recorded allocation decisions inside a Palantir Autopilot workflow, keeping the work inside Nvidia’s own compute boundary.
Nvidia reported that the specialized model scored 86.7% accuracy on an internal development benchmark — figures StorageReview noted are Nvidia’s numbers on Nvidia’s data and are not independently audited.
Open weights matter for the pitch: Nvidia argues operators can customize models on premises without shipping operational data to a closed cloud. That factory stack is separate from, but related to, a June 29 Palantir announcement of an engine to deploy Nemotron models in sovereign and air-gapped environments for U.S. agencies and critical infrastructure. Neither company has named a new agency customer or contract value for that government package in the coverage reviewed here.
Nvidia is acting as both vendor partner and first customer — turning its constrained GPU supply chain into a reference for Palantir Foundry and domain-tuned open models. The buyer for the government sovereign pitch remains unspecified; the clearest live case so far is Nvidia itself.