The phrase has become a fixture of government briefings and corporate slide decks, but it has been easier to invoke than to deliver. Sovereign AI, the idea that nations and enterprises should run artificial intelligence under their own control, has been mostly a policy ambition. On Tuesday two companies announced an arrangement intended to make it a product.
Palantir Technologies said it has designated Nebius as its preferred partner for sovereign AI infrastructure, and the two companies are integrating their systems so that Nebius computing and inference endpoints operate inside Palantir’s enterprise security boundary. The architecture means that for eligible customers, models can be brought to the data rather than the other way around, with sensitive information staying within the controlled environment where it already lives.
The alliance pairs two companies with unusual histories in the AI economy. Palantir built its business on government and enterprise software for analyzing sensitive data, and its tools have become a standard for agencies and companies that cannot send their information to public clouds. Nebius emerged from the founders of Yandex, the Russian internet company, and has been rebuilt as an AI cloud operator with data centers in Europe and North America, selling computing power to customers who want the capability of the big clouds without the strings.
Alex Karp, Palantir’s chief executive, framed the deal in terms of control. Sovereign AI, he said in the announcement, requires computing power that runs under conditions an organization can itself govern, rather than capacity rented from a provider whose policies are set elsewhere. The statement gives voice to a concern that has been spreading through governments: as AI becomes central to defense, intelligence and critical infrastructure, dependence on foreign computing platforms has come to look like a strategic vulnerability.
Arkady Volozh, Nebius’s founder, described the joint ambition in commercial terms. The two companies, he said, intend to sell hyperscale-grade computing to government and enterprise customers that want both performance and data sovereignty, a combination the market has rarely offered. The partnership’s framework also includes modular data centers and sites with existing power connections, the physical assets that have become the scarce currency of the AI buildout.
The deal reflects a shift in where the AI industry’s value is being created. Palantir has traditionally sold software that analyzes data wherever it resides; its revenue has grown as governments adopted its platforms for everything from military operations to fraud detection. The partnership with Nebius extends that franchise into the computing layer itself, letting Palantir offer customers a controlled path to run models without ceding their data to a third party’s cloud.
The market for sovereign AI has been growing on both sides of the Atlantic. European governments have funded national AI computing initiatives, wary of depending entirely on American providers, while Gulf states and Asian governments have invested in their own infrastructure. Palantir’s move positions it to serve that demand with its software layer, and Nebius provides the hardware story, a combination each company would have struggled to assemble alone.
There is a competitive edge to the timing as well. The largest cloud providers have all announced sovereign offerings, versions of their services that run in specific countries under specific rules, but those products remain tethered to their parent companies. A partnership between a software firm with deep government relationships and an independent cloud operator offers an alternative for buyers who want the tools of the AI age without entangling their most sensitive operations with a hyperscaler’s global architecture.
The integration work will take time. Making Nebius endpoints operate seamlessly inside Palantir’s security boundary requires engineering on both sides, from identity management to data governance, and the companies did not say when the combined offering would be generally available. For existing Palantir customers, the promise is incremental: the software they already use gains a new way to access computing power without changing where their data sits.
The two companies bring complementary weaknesses as well as strengths. Palantir has been criticized for the concentration of its revenue in a small number of large customers, and Nebius is still building the scale that hyperscale competitors take for granted. The partnership does not erase those constraints, but it gives each company a story that extends beyond its own footprint: Palantir can point to owned computing capacity rather than rented cloud access, and Nebius gains a distribution channel into the government market that is hardest for new entrants to reach.
The announcement’s significance may be most visible in what it says about the direction of the industry. For years, the AI conversation revolved around models, the algorithms that captured attention and funding. The infrastructure underneath has now become the conversation, and companies are being valued on their ability to control where computing happens. Palantir, which built its name on helping organizations control their information, is now selling control of the machines that process it, a natural extension of its franchise and a measure of how far the industry’s center of gravity has moved.
