How Sovereign Is Sovereign Compute? A Review of 775 Non-U.S. Data Centers
By analyzing a new dataset of 775 non-U.S. data center projects, this paper reveals that U.S. companies control nearly half of the global non-U.S. compute capacity by investment value, suggesting that foreign operators can serve as a regulatory lever for AI governance while highlighting that local infrastructure alone does not ensure digital sovereignty if managed by foreign entities.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
The Big Idea: Who Really Owns the "Digital Real Estate"?
Imagine the world's computer power (used to run AI) is like a massive network of apartment buildings. These buildings are called "data centers."
For a long time, countries have been trying to build their own apartment buildings within their own borders to ensure they have full control over them. They call this "Digital Sovereignty." The idea is: "If the building is on my land, I am the landlord, and I make the rules."
However, this paper asks a tricky question: Just because a building is on your land, does that mean you are the landlord?
The authors found that often, the building is on your soil, but the management company running it is from a different country. If a U.S. company is managing a data center in France, the U.S. government might still have legal power over what happens inside that building, even though it's not on U.S. soil.
The Investigation: Counting the Buildings
The researchers (from RAND and other think tanks) acted like real estate detectives. They didn't just look at the map; they looked at the deeds and management contracts.
- The Dataset: They tracked 775 data center projects in 123 countries (excluding the U.S.).
- The Method: They didn't just count how many buildings there were; they weighed them by how much money was invested. Think of it like this: Counting 100 small sheds is less important than counting one massive skyscraper. They used investment value as a proxy for how much "computer power" (compute) is actually there.
- The Goal: To see how often these foreign buildings are run by U.S. or Chinese companies, which would give those governments a "legal hook" (a way to enforce their laws) inside those countries.
The Findings: The "American Manager" Effect
Here is what they discovered, broken down simply:
1. The U.S. is the Super-Landlord
Even outside the United States, U.S. companies run nearly half (48%) of the total value of these data center projects.
- The Analogy: Imagine you are in Germany. You see a data center. It's on German soil. But the sign says "Managed by Microsoft" or "Managed by Amazon." The authors argue that because a U.S. company is the manager, the U.S. government can likely step in and say, "We have rules for this building, too."
- The AI Factor: This is even more true for AI. For data centers specifically built for Artificial Intelligence, U.S. companies manage 56% of the value.
2. China is Present, But Smaller
Chinese companies (like Huawei or Alibaba) also manage data centers abroad, but they manage a much smaller slice of the pie (about 5% of the total value). They are mostly found in Africa and parts of South Asia.
3. The "Unknown" Landlords
About 36% of the money invested in these projects doesn't have a clear manager listed yet. The researchers couldn't tell who is running them. This is a big gap in the data.
4. The "Sovereignty" Illusion
Many countries (like India, South Korea, or European nations) are building these centers to be independent. But the paper suggests that if they hire a U.S. company to run the facility, they haven't actually achieved full independence. They are still subject to U.S. laws.
- The Metaphor: It's like renting a house in a foreign country. Even if you live there, if the landlord is from another country, they can still evict you or change the rules if they have a contract that says so.
Why Does This Matter?
The paper suggests two main takeaways for the real world:
- For the United States: They have a powerful new tool. It's not just about selling chips (export controls); it's about who manages the buildings. If the U.S. can get its companies to run data centers globally, the U.S. government can potentially regulate AI development happening anywhere in the world, not just inside the U.S.
- For Other Countries: If you want true control over your digital future, you can't just build the building; you have to make sure your own people are the ones running the show. If you hire a foreign manager, you might be borrowing their rules along with their technology.
The Limits of the Study
The authors are careful to say this is an estimate, not a perfect census.
- Some countries (like China) don't always publish clear data, so the numbers might be off.
- They used "money invested" as a stand-in for "computer power," which is a good guess but not a perfect measurement.
- They only looked at projects announced up to late 2024.
Summary
The paper argues that location isn't everything. In the digital world, who runs the show matters more than where the building sits. Currently, U.S. companies are running the show for almost half of the world's non-U.S. data centers, giving the U.S. government a massive, often overlooked, amount of influence over global AI.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.