AI Sovereignty Makes Everyone Weaker—The US Can Lead Differently
At the 21-nation Asia-Pacific Economic Cooperation meeting in South Korea last month, White House Technology Advisor Michael Kratsios endorsed the idea of AI sovereignty as he advanced the new U.S. strategy for exporting American AI. “With the AI export packages we are developing,” he said, “we want you to have the AI sovereignty, data privacy, and technical customization that you so rightly demand on behalf of your peoples.”
At first glance, that line might seem paradoxical. AI sovereignty is typically used to refer to a nation’s effort to cultivate its own capabilities and avoid dependence on foreign-controlled inputs. Under that definition, the notion of countries gaining AI self-sufficiency by importing systems that American companies produce sounds contradictory. However, Kratsios was not talking about self-sufficiency. He was invoking another meaning of the term: AI sovereignty as the ability of a nation to dictate the terms on which foreign AI operates within national borders. For clarity, let us call the first definition self-sufficient AI sovereignty and the second control-oriented AI sovereignty.
Governments may present control-oriented AI sovereignty to U.S. policymakers as a fair demand, one they argue safeguards their national interests while still leaving American companies a route to market access. However, the effect is far from fair because it allows those governments to cherry-pick, drawing on American innovation when it suits them while shutting out U.S. competitors when it doesn’t. This is still protectionism, a different method from the push for self-sufficiency, but protectionism all the same, and it fractures the collective leadership the West needs to build.
To see why, consider the United Kingdom. It is not trying to build an indigenous alternative to the American AI stack like Europe is. On the contrary, it has signaled that it will partner with U.S. providers such as Anthropic and Nvidia to expand its national computing capabilities. But its recent compute roadmap makes clear its plans to privilege AI infrastructure built on British soil and under domestic oversight, ensuring that participation in critical government-led AI projects flows through a domestically anchored system. In other words, even if two foreign AI services are equally secure and capable, Britain will give better market access to the one “that is physically located in the UK and strategically aligned with our national goals.”
The UK gives two sets of justifications for its approach—one framed in terms of security and resilience policy, the other in terms of industrial policy—but both amount to protectionism. Its strategy declares: “Sovereign infrastructure allows us to protect sensitive data, direct resources toward national priorities, and ensure we remain resilient in the face of global instability and supply chain disruption. Beyond resilience, sovereignty is also about opportunity: it allows us to shape our own future, build competitive advantage, and grow high-value sectors that support jobs and innovation across the UK.”
The first set of animating goals—promoting privacy, resilience, and national security—may be legitimate, but the sovereignty measures imposed in their name do little to advance them and at times are even counterproductive. The physical location of a server does not guarantee security, nor does it change the obligation of firms to comply with national privacy rules. Indeed, at times, data localization rules may even heighten risk, as on-premises infrastructure can be more exposed to supply-chain disruption than foreign cloud services. Insisting on ineffective measures, such as AI compute localization, rather than allowing U.S. firms the flexibility to comply with national standards through other effective means, is not about security or resilience at all. It is about disadvantaging foreign providers and shoring up domestic firms.
The second set of justifications—limiting foreign competition to support and protect domestic AI industries and national economic interests—dispenses with any pretense. It is protectionism, plain and simple.
AI sovereignty, whether focused on self-sufficiency or on control, does more than harm foreign companies; it undermines the project of collective AI leadership that the United States and its partners should be advancing together. In a vision of collective leadership, nations still compete intensely, but they also trade for mutual gain and seize opportunities for scientific, technical, and commercial collaboration that expand the base of AI knowledge. By committing to positive-sum policies—open trade, fair economic competition, and cross-border collaboration—these countries strengthen their collective position far more than any could alone, and push back against the real competitor: China, whose state-backed model relies on unfair trade and market-distorting practices.
The problem is that U.S. policy today does not convey this vision. To many partners, the offer to partner sounds like a binary choice between being a customer of the United States or China. That framing reduces them to clients in someone else’s system rather than partners in building shared leadership. Unsurprisingly, countries boxed into this binary reach for sovereignty as their way of carving out influence, erecting barriers, localizing infrastructure, and duplicating capacity. The result is not a whole greater than the sum of its parts, but a scattering of isolated efforts, each weaker on its own and all weaker together than they would be as a collective.
Washington’s own turn toward protectionism only sharpens this problem. When U.S. trade policy drifts toward tariffs and insularity, it undermines the credibility of any call for openness in AI and makes sovereignty abroad look like a defensive response. The UK’s 2024 AI Sector Study, released this week, illustrates the point: It reports that UK businesses are justifying AI sovereignty as a reaction to U.S. protectionism in trade. According to the study, firms “flagged concerns about the potential risks of relying on US technologies and infrastructure, including, for example, the impact of protectionist policies and trade disputes, access to cloud infrastructure, or pricing rates for services set by the major providers… In light of these concerns raised during the interviews, some businesses advocated for the development of sovereign AI capabilities within the UK.”
Protectionism met with protectionism creates a self-defeating spiral, leaving both sides worse off and eroding the trust and collaboration necessary for a collective AI leadership. To break the cycle, restore its authority, and build the collective strength needed to counter China, the United States needs to break from its own knee-jerk protectionist reactions and revive a tradition of national developmentalism, prioritizing advanced-industry dynamism and selective globalization as the basis for an AI strategy that rewards collaboration instead of fragmentation.
If the United States wants to win the AI race, it will have to change its posture and its pitch to partner countries. The offer must be clear: a partnership with the United States that secures the privacy and resilience countries need while giving them a genuine stake in building influence, capturing value, and exercising leadership as part of a collective. But that opportunity has conditions. It depends on openness: pro-competition, pro-collaboration, and pro-innovation. The idea that nations can invoke “AI sovereignty” to draw on U.S. technology when convenient, while walling off their markets, is not a bargain U.S. policymakers should entertain.
Image Credits: No.10 Flickr page
