AI Sovereignty Is Geopolitical Leverage

Earlier this year, the U.S. government imposed export controls on Anthropic’s Claude Fable 5 and Mythos 5, restricting access by foreign nationals inside and outside the United States. With the order taking effect immediately and no reliable way to verify users’ nationality, Anthropic suspended both models for everyone. The controls were lifted on June 30, and Fable returned globally on July 1. Some U.S. organizations regained access to Mythos earlier. Whatever you think of the decision, every government outside the United States should have taken note. Organizations relying on those models lost access because of a decision made in another country’s capital. That’s what dependency looks like when the technology in question is starting to sit underneath everything else.

Much of the world has little infrastructure with which to reduce that dependence. A 2025 Oxford-led study of nine major public-cloud providers found AI compute concentrated in a few dozen countries. It was a survey of those providers, not a census of every AI data center, but the imbalance is clear. The World Bank reports that low-income countries held less than a tenth of one percent of global co-location data-center capacity as of June 2025. Those countries will still use AI. They’ll increasingly rent it, running public services, health systems, and analysis on models trained elsewhere, priced elsewhere, and governed by rules written elsewhere. English also dominates much of the training data, leaving many local languages and contexts poorly represented.

A sovereign tech stack doesn’t mean building everything yourself. The chip supply chain alone makes full independence unrealistic for most countries. It means controlling enough of the critical layers, including power, compute, data, and capable models, to keep essential services running when a foreign supplier or government changes the terms. Hosting a data center isn’t enough on its own. A country can house a facility without controlling its hardware, cloud platform, or the data passing through it.

That’s why Canada and Britain are funding public AI computing capacity, while Saudi Arabia is building infrastructure through its state-owned HUMAIN company. India is expanding shared compute access and supporting models built around its own languages and needs. These efforts still depend on international suppliers, but they can give governments more options. For smaller countries, open-weight models, regional partnerships, and local hosting offer practical ways to reduce dependence without trying to reproduce the entire frontier AI industry.

The countries that fail to develop those options risk falling behind in ways that compound. Their scientists will have less access to the compute increasingly needed in fields such as drug discovery, and their talent will have more reasons to leave. Their intelligence services will face AI-enabled adversaries while depending on analytical tools they cannot fully control. Their economies risk paying foreign providers for a growing share of the value AI delivers, and their governments will negotiate for access from a position of weakness. And let's not even mention the potential weapons that they will face.

In the industrial era, countries without energy security learned how much leverage others held over them. Compute and models are becoming another source of that leverage. Sovereignty has long been measured in borders and currencies. Increasingly, it will also be measured by who controls the machines a country thinks with.