Stanford flags lock-in risk in Big Tech sovereign AI push

Stanford’s Institute for Human-Centered Artificial Intelligence says the largest U.S. cloud and AI vendors are turning digital sovereignty demand into a product category that may deepen customer dependence. The study reviewed sovereign AI and cloud offerings from Microsoft, Google, AWS, Nvidia and OpenAI, finding that the tools can improve some local control while leaving core infrastructure, models and software stacks tied to the same suppliers.

The report lands as governments and enterprises reconsider where AI data is stored, how models are deployed and which providers control the underlying compute. Sovereign AI is increasingly being sold as a response to geopolitical risk, data residency rules and national industrial policy. Stanford HAI’s point is narrower: buyers may gain more control over location, operations or access, but not necessarily over the technology layer that determines switching costs.

Nvidia is the clearest example in the study. Stanford HAI said the chipmaker has become a dominant vendor in sovereign AI, with the category accounting for about 14% of Nvidia revenue, or roughly $30 billion. The report said Nvidia’s sovereignty offerings could make moves to alternative chip suppliers such as Intel or AMD costly because customers would be adopting a broader Nvidia stack, not just buying GPUs.

Nvidia describes its AI factories as locally owned and operated AI clouds for training and inference, built through public-private partnerships. The company has also said its AI Nations initiative helps countries build AI ecosystems. In one example cited by Nvidia, France’s Ministry of Economy and Finance is using AI agents built on Nvidia’s platform, through infrastructure controlled inside France, to automate workflows and process millions of documents.

Control is being packaged by the existing platform owners

Stanford HAI said OpenAI’s sovereign AI efforts could expand local access to advanced models, but would not give buyers more authority over the underlying systems. That distinction matters for governments and regulated enterprises trying to reduce exposure to foreign platforms: access to a model endpoint or locally hosted deployment is different from control over model development, weights, infrastructure dependencies and upgrade paths.

The study also pointed to sovereign cloud offerings from Google, Microsoft and AWS. Those products address some customer requirements around jurisdiction, operations and compliance, according to the analysis, but can be expensive and hard to run. Stanford HAI described the three cloud providers as having the broadest global set of sovereignty-related offerings among the companies it reviewed.

Demand is not coming from nowhere. Kyndryl’s 2025 Cloud Readiness Report found that three-fourths of business leaders are concerned about geopolitical risks from storing data in global cloud environments. The same report said 65% of business leaders are changing cloud strategies because of digital sovereignty requirements, mainly outside the U.S.

Policy pressure is also rising. The European Union introduced its European Technological Sovereignty Package last month as part of an effort to reduce reliance on U.S. technology companies. Gartner has projected global sovereign cloud spending will rise 35.6% this year.

For vendors, sovereignty is becoming another way to defend the stack. For buyers, Stanford HAI’s study suggests the trade-off is less about sovereignty versus dependence and more about which layer of dependence they are willing to accept. Local operations and data residency may satisfy regulators, but the economics of compute, models and cloud services still favor the incumbent platforms that already own the distribution.

This story draws on original reporting from CIO Dive.