AI Business

Enterprise AI Sovereignty Faces a Five-Layer Control Problem, Study Finds

A new assessment framework reveals that no major cloud provider has adequately addressed how U.S. legal authority could override data protection commitments, leaving enterprises vulnerable to forced data disclosure regardless of where their infrastructure sits.

·4 min read
Enterprises fear AI success more than failure, because success means lock-in they can’t escape
Enterprises fear AI success more than failure, because success means lock-in they can’t escape

Data centers that once housed scientific computing workloads now run artificial intelligence systems that transform information into business value. This shift has surfaced a question that corporate boards rarely address openly: who truly owns the data flowing through these systems and the intelligence they generate? A fresh buyer's guide attempts to quantify the degree to which enterprises retain command over their AI infrastructure and the outputs it produces.

Sovereignty in the AI context is not a simple yes-or-no proposition but rather a layered structure requiring scrutiny at each level, according to Amit Eyal Govrin, lead analyst for sovereign AI at theCUBE Research and chief executive officer of Agentcy Labs. Govrin presented his findings during an exclusive broadcast at SC26, speaking with theCUBE Research's John Furrier.

When it comes to sovereignty, it's all about control. And when we have to look at control, there's a lot of layers to go into it. And the first one is, are you in fact comfortable … ceding that control to somebody else? Are you comfortable being in somebody else's jurisdiction where they can potentially compel your data if ever required?

Amit Eyal Govrin

Five pillars structure the sovereignty framework

Govrin's assessment rests on five distinct pillars: territorial, operational, technological, legal and financial. The territorial dimension—essentially where physical data centers operate and where information resides—represents the easiest checkpoint for providers to satisfy. The operational layer proves considerably more complex.

Amazon Web Services Inc. illustrates this complexity. While AWS maintains a sovereign region in Brandenburg, Germany, serving EU customers, physical location alone does not guarantee sovereignty. Govrin offered an example: a German customer using the AWS Frankfurt region might appear protected, but if that region experiences an outage, backup operations could route to facilities in Seattle or Hyderabad, potentially circumventing intended jurisdictional boundaries.

You're in the AWS Frankfurt region as a German customer. Perfectly fine. Are you sovereign because you're in that region? If that region goes down, that PagerDuty may not necessarily go to somebody out of Germany. It may go somewhere else. It may go into Seattle or to Hyderabad for all you know.

Amit Eyal Govrin

Methodology and findings across hyperscalers

The research team evaluated seven hyperscalers across 40 distinct offerings, producing 117 individual grades across five geographic regions. The scoring methodology employed 17 binary tests, with legal considerations assessed separately due to the variability of local regulations. The framework drew from more than 300 sources and hundreds of interviews with enterprise buyers. Google LLC's air-gapped offerings performed relatively strongly, yet one discovery applied uniformly across every major provider.

No hyperscaler across the board had an answer to the U.S. CLOUD Act. Now imagine you're a European entity or anywhere else outside the world, and now there's a production order from a U.S. court system that states that this hyperscaler needs to turn over your data. At that point, there is no contest. Your data becomes Uncle Sam's data.

Amit Eyal Govrin

This vulnerability represents a critical gap. The U.S. CLOUD Act grants American courts authority to compel data disclosure from U.S.-based technology companies, regardless of where data physically resides or what contractual protections enterprises believe they have negotiated.

Why AI sovereignty matters more than general data privacy

Furrier drew a distinction between sovereignty concerns and privacy frameworks like the General Data Protection Regulation, characterizing the latter as less operationally consequential. The stakes differ fundamentally because AI models generate revenue-producing intelligence from data at rest. When financial outcomes depend on data control, the competitive and legal dynamics shift dramatically.

https://www.youtube.com/embed/xN9GxNIddxo?feature=oembed

When money's on the table, you know what happens. Everyone fights for it or argues for it. So I think this is a tinderbox moment where one match could blow everything up.

John Furrier