Nvidia picks Palantir Foundry to run the 1.3 million-part supply chain behind its Vera Rubin AI racks

NVDA has chosen PLTR Foundry as the system of record for the supply chain behind its Vera Rubin AI racks, a build that Nvidia says draws on roughly 1.3 million individual parts per rack. The companies announced the arrangement on September 10, 2026 and showcased it at Palantir's AIPCon 11 event, packaging it as a Sovereign AI Operating System reference architecture.

The technical shape matters more than the label. Palantir contributes Foundry, AIP and its Ontology layer, the data model that maps physical parts and suppliers to software objects; Nvidia contributes its Nemotron open models and cuOpt, its optimization library for routing and scheduling problems. Jensen Huang framed the combination as an operating system for the physical economy, which is Nvidia positioning the deal as a template it can resell rather than an internal IT project.

For Palantir, the reference-architecture framing is the commercially interesting part. The company reported second-quarter revenue of about $1.94 billion, up roughly 93% year over year, and trades at a price-to-earnings multiple near 240 against a market capitalization of about $390 billion. A named deployment inside Nvidia's own manufacturing chain is the kind of proof point that supports that multiple, though it is a design win rather than a disclosed contract value: neither company put a dollar figure on the work.

The scale on Nvidia's side explains why the problem is worth solving. Nvidia's data center segment produced $89.02 billion in revenue, up 117% year over year, and the company has guided to $108.0 billion plus or minus 2% for the current quarter. Coordinating million-part assemblies at that run rate is a logistics constraint on revenue, not a back-office concern. Shares in both names were soft around the announcement, with Palantir trading near $167.15 and down about 1.4% midday, so the market treated this as strategic rather than immediately material. What to watch is whether Nvidia or Palantir converts the reference architecture into named third-party customers, which is the step that would turn a single deployment into a product line.