AI sovereignty is about choice, not ownership, says Ren Ito

This is the latest confirmation that AI has become a geopolitical issue of the utmost importance. Until recently, countries were competing to develop services, infrastructure, and applications based on cutting-edge AI systems. But now, access to these systems has become a matter of strategic concern.

It was previously believed that AI would follow the logic of globalization. Relying on a handful of cutting-edge AI models (developed primarily in the U.S.), countries would compete in services for end users, semiconductor manufacturing, and the data and applications markets.

Access to the most advanced AI systems was taken for granted.

But that is no longer the case, and the central question is no longer which model is better, but rather which one can be accessed.

Since cutting-edge capabilities have become a matter of national security and diplomacy, governments may be tempted to pursue “AI sovereignty” by creating national champions or domestic alternatives to leading U.S. products (ChatGPT and Claude). But while this instinct is understandable, it won’t solve the problem.

The fact is that AI is evolving too quickly for such a strategy to pay off. Technical advantages that seem decisive today may disappear in just a few months. Today’s breakthrough becomes tomorrow’s baseline standard. Models may dominate the news for a couple of months, but competitors quickly catch up or surpass them.

Even countries willing to invest tens of billions of dollars in model development have little chance of success in direct competition with the world’s largest tech firms.

And so the question is not whether a particular country can create a cutting-edge model, but whether it can ensure reliable access to cutting-edge intelligence, wherever it may emerge.

The case of Anthropic illustrates this point. If access to a leading AI model can be cut off instantly, dependence on a single provider becomes a strategic risk.

This does not mean that every country must develop its own cutting-edge model. But it does mean that no country can count on uninterrupted access to other countries’ systems.

Cooperation and coordination are essential

For the United States’ closest allies, maintaining access must be the top priority.

Japan and other G7 countries are united with the United States not only by democratic values but also by significant security interests. And supporting the technical resilience of its allies ultimately strengthens America’s own strategic position.

Furthermore, AI is still an immature, rapidly evolving technology whose future trajectory is not yet determined. Continued cooperation between American model developers and allied nations—which contribute technology, talent, infrastructure, and markets—will be essential for expanding the AI ecosystem.

AI should not be a technology kept to oneself; it should be a technology developed collectively.

In the emerging AI economy, competitive advantages will increasingly stem not from owning a particular model, but from the ability to evaluate, select, and coordinate the work of multiple models.

Organizations that can seamlessly switch between competing systems will be more resilient than those that have become dependent on a single provider and are exposed to too many risks (ranging from technical failures and commercial disputes to geopolitical pressure). Much like countries seeking to diversify their energy and semiconductor supplies, they will diversify their AI service providers.

But coordination alone is not enough. The government must be able to independently evaluate AI systems. Governments need the ability to determine for themselves which models are truly useful (compared to others) and what risks may arise from specific systems. Such an assessment cannot be fully delegated to foreign firms or governments.

The New Role of Cybersecurity Institutions

This is precisely why the AI safety institutions and national cybersecurity agencies currently being established are of particular importance. Their role is not limited to attempts to regulate AI. They provide the independent technical expertise needed to make informed government decisions.

Countries that cannot secure access to AI systems will inevitably find that their decisions are determined by someone else.

Relying on a variety of AI models rather than a single provider is already a form of deterrence, as it reduces the effectiveness of coercive measures.

When governments and organizations can switch freely between cutting-edge models, restricting access to any one of them largely loses its strategic significance. Multimodality becomes not only an economic factor but also a geopolitical one.

However, diversification alone is not enough. To achieve true AI sovereignty, states must learn to combine national data with a variety of cutting-edge AI systems; accumulate knowledge about which measures work; and apply that knowledge in decision-making.

This final step—sovereign decision-making—is difficult to achieve when a country depends on others to interpret critical information. If the analysis of military intelligence, economic statistics, medical threats, or critical infrastructure begins to depend on external actors, political autonomy is compromised.

The implications extend beyond the realm of AI. As soon as access to intelligence becomes a matter of geopolitics, a clear challenge arises for middle powers. Instead of attempting to replicate the capabilities of the United States or China, middle powers must secure room for strategic maneuver while remaining deeply integrated into the global economy.

This is a familiar challenge: throughout modern history, successful middle powers have achieved prosperity not by isolating themselves from major powers, but by maintaining strong alliances and the ability to act independently.

Trust Becomes a Strategic Choice

The new geopolitical rivalry will center on who develops, finances, operates, and optimizes the critical infrastructure for the AI era: data centers, power grids, communications networks, logistics systems, ports, and digital public infrastructure. AI capabilities will be embedded in the systems on which modern societies depend.

In this context, the meaning of the concept of “safe and trustworthy AI” must also evolve. Discussions about AI safety today often focus on technical issues: model requirements, transparency, bias, misinformation, and harmful outputs. These are important issues, but for governments, businesses, and the public at large, building trust requires something more.

A reliable AI ecosystem is one in which users can count on continuous access, retain meaningful control over their data, and remain independent of political decisions made in other countries.

It is an ecosystem in which access is not contingent on political alignment, and participation does not require relinquishing digital sovereignty.
Trust turns out to be linked not to any single model, but to the institutions, governance mechanisms, and international relations surrounding it.

As AI becomes critical infrastructure, its reliability, resilience, and political neutrality may become just as important as the performance metrics of the underlying model.

The most powerful model will not necessarily be the most valuable if access to it can be suddenly cut off, or if dependence on it limits strategic options.

AI sovereignty is not about creating a national version of ChatGPT, but about preserving freedom of action in a world where access to intelligence is beginning to be contested.

The point is to have the ability to choose, not the right of ownership.