Summary of Key Points
This analysis, based on the latest research from the Brookings Institute, elucidates the new changes in the global geo-economy in the era of AI. AI has evolved from a common technical tool, such as chatbots, to a core national infrastructure on par with power grids, water supply systems, and transportation networks. Moderate powers like Europe, Japan, and South Korea are now facing an unprecedented risk of "invisible sovereignty erosion"—if their industries, healthcare, defense, and public services all rely on AI systems developed by Chinese and American companies, and those companies decide to cut off services or change rules, they may retain their territorial and legal sovereignty in name, but they would actually lose control over their own critical systems.
The report does not advocate for moderate powers to attempt to develop 100% domestically produced AI solutions; instead, it proposes a practical approach: to maintain control over key areas and participate normally in the global division of labor in non-core areas. It suggests avoiding the competition in the field of general-purpose large models dominated by China and the US, and focusing on "physical AI" applications in industries and infrastructure. The report also uses the public procurement funds that governments already spend each year as a lever to drive industrial development. It warns China that new access rules based on security and jurisdiction are being established, and proactive planning is far better than trying to fix problems after they occur.
---
Detailed Interpretation
1. Sovereignty is no longer just about territorial integrity
Many still think that sovereignty is intact as long as territory is not occupied and laws are not changed. However, the new risks in the AI era go beyond this framework. There are real-world examples: In 2026, the US required the AI company Anthropic to prohibit non-US users from accessing certain models, and Anthropic simply stopped services for all users globally because it was technically impossible to verify nationalities in real-time. US regulatory rules can even extend to chips manufactured in other countries; if those chips use US technology, they cannot be sold if the US decides so.
The gap in investment is even more significant: In 2025, US private investment in AI was nearly $286 billion, 23 times the amount publicly disclosed by China. The five major US tech companies plan to invest $700 billion in AI infrastructure in 2026, while the total AI investment of all moderate powers combined is far less. A crucial point in the report is that the difference between dependence and subordination is not about how much foreign technology is used, but whether there are alternative options. If there is only one provider of AI, and that provider can cut off services at any time, it means complete subordination and the loss of control over one's own systems.
2. Don't chase 100% domestic production; "modular sovereignty" is a more practical approach
Many countries have called for 100% domestic AI autonomy, aiming to control everything from mining to chip manufacturing. The Brookings Institute dismisses this idea as unachievable. The AI industry chain is a global network with many concentrated bottlenecks, and no country can control all aspects. The report proposes "modular sovereignty," which means focusing on the most critical areas and using the necessary technologies while participating in the global division of labor for the rest.
However, the US's "Silicon-based Peace" initiative threatens this approach by implicitly excluding countries that sign the agreement from joining AI cooperation organizations led by China. Vague exclusivity rules are more intimidating than explicit bans, as countries don't know what actions will trigger issues, leaving them with no choice but to follow the US.
3. Stop focusing on chatbots; find new areas not dominated by China and the US
The field of general-purpose large models is already locked by China and the US. Moderate powers cannot compete on cost-effectiveness and scale. The report suggests focusing on "physical AI" in industries and infrastructure, where the US has less advantage. China has already taken the lead in industrial robotics, with 54% of the global installed capacity in 2024, four and a half times that of Japan. Moderate powers should not compete on model performance but on who can be responsible for system operations within their legal jurisdiction. For example, Japan's industrial AI projects use a distributed training model that keeps data within the country, preventing the theft of business secrets.
4. Don't waste money on subsidies; public procurement is a powerful tool
To build a strong AI capability, governments need to allocate significant funds. Europe and Japan cannot afford to subsidize the industry directly; the EU needs unanimous approval from all members to increase taxes or borrow money, which can take years. The report highlights the power of public procurement: the EU spends 2 trillion euros annually on services, projects, and supplies, accounting for 13.6% of its GDP. This amount alone can significantly influence an industry's fate. By setting clear rules for AI system purchases (e.g., data localization, local responsibility, and open-source models), governments can encourage investment without additional subsidies. This was how Airbus became a global leader; European governments provided firm orders, allowing Airbus to secure funding and grow.
5. Is the window for Chinese AI exports closing? Proactive planning is better than reactive action
This report is a guide for moderate powers to develop AI autonomy. For China, it means that exporting AI will not rely solely on cost-effectiveness and model performance. Future public procurement rules will include requirements such as legal responsibility and data localization. China needs to prepare products and comply with local regulations in target markets. The concept of "modular sovereignty" is also valuable for us, as it focuses on identifying critical areas and ensuring adequate alternatives, which is more important than simply calculating the percentage of domestically produced components.