The AI Sovereignty Paradox: Should Countries Buy, Build, or Lease to Maintain Strategic Control of Their AI?

“AI sovereignty” has become an important but poorly defined concept in global AI governance. Governments across the world are trying to limit their dependency on other countries for AI development, operations, and oversight to protect their data, their national security, and their economic prospects.

As more governments invest heavily to address these concerns, an emerging commercial market has developed to meet demand. Technology companies — from major cloud providers to local startups — now offer “sovereign AI solutions” that promise greater control over AI infrastructure and data.

But true sovereignty isn’t straightforward. While these commercial offerings provide benefits like local data hosting and regulatory compliance, they also reconfigure rather than eliminate dependencies on foreign providers. In many ways, the question isn’t whether to pursue complete AI sovereignty, says HAI Denning Director James Landay, but how to navigate the complex spectrum of interdependent arrangements in practice.

“The core policy challenge of AI sovereignty is not eliminating dependence but calibrating interdependence,” Landay says. “Decision-makers should avoid treating sovereignty as an end goal and prioritize solutions that retain and expand strategic choice without losing access to frontier capacity.”

In a new issue brief called “The Commercial Landscape of AI Sovereignty Offerings,” Landay and scholars from Stanford HAI explore different approaches to this calibration — and what tradeoffs they require. Here Landay details some of their findings. 

You find that technology companies are selling sovereign AI tools. What does that mean?

Landay: Right now, we see Big Tech adapting products and services for government procurement, and local startups pushing out home-grown solutions for their countries. 

For example, Nvidia offers “AI Factories” across more than 25 countries — essentially data centers optimized for AI that promise to replace the need for countries to independently build their own data centers and source all the hardware. Microsoft, Google, and AWS compete in the sovereign cloud market, offering disconnected clouds or jurisdictional shields to improve autonomy, while OpenAI launched an OpenAI for Countries initiative that allows localized deployment and customization of its large models. 

These initiatives are meeting a real and urgent need — they offer local data hosting, regulatory compliance, and local infrastructure. But they also deepen dependencies on U.S. providers. Think about how hard it is to switch your cable company. Now imagine switching cloud providers that run your country’s entire AI infrastructure. You get locked in to U.S. companies very easily and it can be very expensive to change. 

You mentioned other, often smaller companies outside of the U.S. that are also selling sovereign AI solutions — would their products be more beneficial to countries?

Landay: We see plenty of smaller companies marketing sovereignty solutions, but what we’ve found is few are fully domestic. Most still depend on some type of U.S. technology — Nvidia chips, U.S. cloud partnerships, foreign-built models. That gives governments some ability to control their own AI, but not complete self-sufficiency. We have seen that governments that financially back startups can reach greater independence. 

How should countries navigate this?

Landay: It’s important to understand that sovereignty isn’t binary — it exists on a spectrum of interdependent arrangements. Most countries will have to accept some dependencies, but they should also consider pathways for reducing these dependencies — funding home-grown companies and enterprises, selecting open-source alternatives, considering where in the AI ecosystem they should limit dependencies versus where it’s less important for their own unique national objectives. 

For example, we see lots of countries making decisions aimed at reducing their US CLOUD Act legal exposure and improving domestic data privacy by either building their own domestic clouds or purchasing fully locally controlled cloud products. The former is a more cost-efficient option, but the latter may open more opportunities for domestic research and development.

In general, though, sovereignty should be understood as shaping and negotiating dependencies, not eliminating them totally. 

“The Commercial Landscape of AI Sovereignty Offerings” was written by Stanford HAI researchers Caroline Meinhardt, Juan N. Pava, and Caroline Yee, and Stanford HAI Denning Director James Landay.