Nvidia is the first customer of a joint software stack that moves its own supply chain onto Palantir's Foundry platform. The stack pairs Nvidia's open-weight Nemotron models and the cuOpt optimisation library with Palantir Foundry. A Vera Rubin accelerator cabinet carries a bill of materials of 1.3 million pieces, drawn from thousands of vendors. The setup can run on a company's own machines or in a cloud.
Nvidia is applying the stack it announced with Palantir to its own production first. 24/7 Wall St. records the chipmaker as that stack's first customer; THE DECODER writes that an operation spanning millions of pieces and thousands of vendors is moving onto Foundry. The figure that marks the scale sits on a Vera Rubin accelerator cabinet: the bill of materials reaches 1.3 million pieces.[1], [2]
THE DECODER writes that Nemotron weights are adapted on a firm's own operating records, and that cuOpt is used when scenarios are chosen. Palantir Foundry is the layer that work sits on. Manufacturing, pharmaceuticals and energy sit among the sectors named. Nemotron is an open-weight model family; cuOpt is Nvidia's optimisation library. The two sit side by side on Foundry. That account shows how Foundry is used as the data layer.[1]
The reference architecture is billed as a sovereign AI operating system: it can stand on a company's own machines or in a cloud. Nebius, Rackspace, Cisco and Dell appear as infrastructure partners. 24/7 Wall St. adds that Palantir's AIP layer and Ontology see the bottleneck and steer material from wafer through to the first token. Palantir put extra detail on the project at AIPCon 11.[1], [2]
