The Sovereignty Pitch Two Governments Already Rejected

What Nvidia and Palantir Announced

Nvidia and Palantir announced a joint sovereign intelligence stack for critical supply chains on September 10, built to run on a customer's own infrastructure while the customer keeps full ownership of the underlying data.

The stack combines Palantir's Foundry and Artificial Intelligence Platform, structured around its Ontology framework, with Nvidia's Nemotron open models, its cuOpt optimization software, and NeMo data tools. The two companies packaged it as the Palantir Sovereign AI Operating System Reference Architecture, deployable on premise, at the edge, or through partners including Dell, Cisco, Rackspace and Nebius. The first deployment runs on Nvidia's own supply chain, which spans 1.3 million parts and thousands of suppliers.

Nvidia CEO Jensen Huang called supply chains "the operating system of the physical economy." Palantir CEO Alex Karp said the sovereign stack delivers "capabilities that exceed the frontier while providing alpha protection qualities unavailable otherwise." The companies named agriculture, manufacturing, pharmaceuticals, retail, technology and government as target industries.

What Sovereign Means Here, and What It Does Not

In this announcement, sovereign means data control, not legal independence, and the difference is the whole story.

Nvidia and Palantir define sovereign intelligence as keeping ownership of proprietary data while running it on premise or in a controlled environment, instead of sending it to an external cloud. That is a real property, and it answers a real question many buyers ask first: where does my data physically sit.

It does not answer a different question: which country's laws govern the company operating the system. Under the US CLOUD Act, a US-based provider can be compelled to disclose data it controls to US authorities regardless of where that data is hosted. A server in Frankfurt or Paris does not change which government can reach it if the company running it is American, and France's own government has previously described exactly this exposure as a threat to its digital sovereignty.

Two Governments Tested This Claim Already

Germany and France did not have to guess how this trade-off plays out, because both examined Palantir specifically in 2026 and walked away.

In April, Germany's Bundeswehr ruled out a Palantir system built on the Maven framework already used by the Pentagon and several NATO members for battlefield decision support. A defense ministry official said giving industry personnel access to national datasets was "unthinkable" for now. Rather than pick one European alternative, the Bundeswehr is testing 30 systems, mostly German, including Almato and Orcrist, alongside France's ChapsVision, with a prototype due in 2027 and deployment targeted for 2028.

In June, French prime minister Sebastien Lecornu announced that the DGSI, France's domestic intelligence service, would replace Palantir with ChapsVision, ending a relationship struck in urgency after the November 2015 Paris attacks. Matignon said Palantir will keep running the system during a multi-year transition while staff train on the replacement, and Lecornu framed the move around avoiding what he called strategic dependencies.

The UK moved the opposite direction. On September 8, Britain's National Energy System Operator extended its Palantir Foundry contract with a direct award worth 21.2 million pounds, without a competitive tender, running through July 2027 with an option to extend to July 2028.

Three governments, three answers, in the same handful of months. Two walked away specifically over sovereignty. The one that stayed skipped the tender process that would have tested the question at all.

The Same Exposure Already Reached a Different Vendor

The jurisdiction problem is not theoretical, because it already happened to a different American AI company months before Nvidia and Palantir made their pitch.

On June 12, a US Commerce Department export control order suspended foreign national access to two of Anthropic's most advanced models, Fable 5 and Mythos 5, everywhere in the world, including for foreign employees inside the United States. Anthropic disabled both models for all customers to comply, and Washington partially restored access on June 27. No customer contract, hosting location or on-premise deployment could have prevented that suspension, because the control point was the vendor's home jurisdiction, not the customer's infrastructure.

Palantir CEO Alex Karp addressed the sovereignty rejections directly at a G20 gathering on September 3, calling them "backfiring" and arguing the regulatory environment is not helping Europe. He said sovereignty decisions were being made by people who lack technical expertise. Yes, but Germany's own reasoning was not about capability: its defense ministry called industry access to national datasets unthinkable, a governance objection that a faster or more capable model does not resolve.

What This Actually Means for a European Buyer

The diligence question a sovereign AI pitch invites is not where the servers sit, it is who can be legally compelled to act against you, and under what law.

For a European operator shortlisting sovereign AI vendors this year, two governments already ran that exact test on this exact vendor and answered no. A buyer does not have to repeat Germany's and France's diligence from scratch. It can ask what has changed since April and June, and whether an on-premise deployment actually closes the jurisdictional gap both governments identified, or only the physical-location question neither of them was asking about.

Read next: Japan Bought an AI Factory to Own Its Robots  |  Mistral's Own Chip Suppliers Just Became Its Owners