Nvidia and Palantir are selling a sovereign AI stack, starting with Nvidia’s own supply chain

Nvidia and Palantir have announced today a joint product for running AI over corporate supply chains without handing the underlying data to anyone else, and the first customer is Nvidia, which is using it on its own.

It pairs Palantir’s Foundry, its Artificial Intelligence Platform and its Ontology with Nvidia’s open Nemotron models and cuOpt optimisation software, packaged as something Palantir calls its Sovereign AI Operating System reference architecture.

It can be deployed on-premises, in a cloud, or in a colocation facility, with Dell, Cisco, Rackspace and Nebius named as partners.

The use case is real, and the number that explains it is 1.3 million: the count of parts in a single Nvidia Vera Rubin rack. Tracking that many components across suppliers, and working out which constraint is the one actually holding up a build, is a genuine optimisation problem rather than a slide.

“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.

The word to examine is sovereign. In this product it means the customer keeps control and ownership of its proprietary data, which is a meaningful commitment and is nothing.

It does not mean the customer controls the ontology that structures that data, the models that reason over it, the optimisation software, the accelerators underneath, or the two American companies that supply all four. Sovereignty over the contents, dependency on the machinery.

Europe has spent the week arguing about precisely this distinction, mostly without naming it. Nvidia signed eight Australian data centre operators to build local capacity described as sovereign AI, where the racks are Australian, and the stack is American.

The Commission proposed procurement rules with a European preference framework whose scope for software is undecided, having already produced Made in EU rules that reached concrete and aluminium and left cloud hosting alone.

And Germany answered the question in the sharpest available form. Berlin ruled out Palantir for military decision support in April, and is now evaluating 30 alternatives, mostly from German suppliers, with a prototype due in 2027.

France’s intelligence service is dropping Palantir for a homegrown rival. Paris and Berlin have pledged to build a European competitor, and German officials promptly worried that would be another dependency.

A company launching a product with sovereign in the name, in the same month two European governments are unpicking their dependence on it, has read the market correctly.

Whether the product answers the objection is a different question. European resistance to Palantir has rarely been about where the data sits.

It has been about who holds the analytical layer, what happens to access in a diplomatic dispute, and whether a supplier that is deeply embedded with the US government is the right place to run a national security function. A deployment model that puts the servers in your building does not change any of those.

Alex Karp’s own description is worth quoting for connoisseurs.

“NVIDIA has arguably the most valuable, intricate and complex supply chain in the world. Our sovereign stack, powered by Nemotron models and Ontology, is delivering capabilities that exceed the frontier while providing alpha protection qualities unavailable otherwise,” said Alex Karp, cofounder and CEO of Palantir Technologies.

“We are very proud of our partnership and its cornerstone role in the sovereign AI revolution.”

For a European manufacturer, the practical question is narrower and more useful than the geopolitics. This is a serious tool for a serious problem, offered by the company that has the hardest version of that problem and is using it internally.

That is a stronger reference than most enterprise software ever gets. It is also a deeper commitment than buying a licence, because an ontology is not a thing you swap out later.