“AI sovereignty” has become important, poorly defined concepts In global AI governance. Governments around the world are seeking to limit their reliance on other countries to develop, operate, and monitor AI to protect their data, national security, and economic prospects.
As governments invest heavily to address these concerns, emerging commercial markets have developed to meet demand. From large cloud providers to local startups, technology companies are now offering “sovereign AI solutions” that promise greater control over AI infrastructure and data.
But true sovereignty is not easy. These commercial products offer benefits such as local data hosting and regulatory compliance; reconfigure Rather than eliminating dependence on foreign providers. In many ways, the question is not whether to pursue full AI sovereignty, says HAI Director Denning. james landdayBut how do you actually navigate the complex spectrum of interdependent arrangements?
“The central policy challenge in AI sovereignty is not to eliminate dependence, but to reconcile interdependence,” Landay says. “Decision makers should avoid treating sovereignty as an end goal and prioritize solutions that preserve and expand strategic options without losing access to frontier capabilities.”
In a new issue brief called “The Commercial Landscape of AI Sovereignty Offerings,” Landay and Stanford HAI researchers explore different approaches to this alignment and the tradeoffs they require. Here, Landay details some of their findings.
I noticed that technology companies were selling sovereign AI tools. What does that mean?
Monday: We are now seeing big tech companies adapting their products and services for government procurement, and local startups promoting homegrown solutions for their countries.
For example, Nvidia offers “AI factories” in more than 25 countries. This is essentially an AI-optimized data center that promises to replace the need for countries to build their own data centers and procure all the hardware. While Microsoft, Google, and AWS compete in the sovereign cloud market, offering disconnected clouds and jurisdictional shields to improve autonomy, OpenAI launched the OpenAI for Countries initiative to enable localized deployment and customization of large-scale models.
These initiatives provide local data hosting, regulatory compliance, and local infrastructure to meet real and urgent needs. But it also deepens our dependence on U.S. providers. Consider how difficult it is to switch cable companies. Now imagine switching cloud providers to run your country’s entire AI infrastructure. It’s all too easy to become locked into a U.S. company, and changing can be very expensive.
You mentioned other, often smaller, companies outside the US that are also selling sovereign AI solutions. — Will their products be more beneficial to the country?
Monday: There are many small and medium-sized companies marketing sovereign solutions, but very few are entirely domestic. Most still rely on some form of US technology, such as NVIDIA chips, US cloud partnerships, or foreign models. This gives governments some control over their own AI, but does not allow them to be completely self-sufficient. We have seen that governments that financially support startups can achieve greater independence.
How should countries overcome this?
Monday: It is important to understand that sovereignty is not dualistic, but exists on a variety of interdependent arrangements. Most countries will need to accept some degree of dependence, but they will also need to consider paths to mitigating these dependencies. This means funding your own companies and companies, choosing open source alternatives, and considering where in your AI ecosystem you should limit dependencies and where dependencies are less important to your own national goals.
For example, we see many countries making decisions aimed at mitigating legal risks under the US Cloud Act and improving domestic data privacy by building their own national clouds or purchasing fully locally managed cloud products. The former is a more cost-effective option, while the latter could provide more opportunities for domestic research and development.
In general, however, sovereignty should be understood as creating and negotiating dependencies, not eliminating them altogether.
“The Commercial Landscape of AI Sovereign Services” The book was written by Stanford HAI researchers Caroline Meinhart, Juan N. Pava, and Caroline Yee, and Stanford HAI Denning Director James Landay.
