Recently, I have heard quite a few friends in the US begin talking about AI doomsday, an idea that AI will soon go out of control and replace humanity. It’s particularly popular in the circle of AI developers in Silicon Valley. While probably because of the quiet nature of STEM people, I have yet to see Chinese AI developers express similar views. I am not ready to pass judgment on the views of those working at the frontiers of AI research. Instead, I have chosen to share some discussions from China that I have recently read about AI’s impact to give readers a sense of the topics Chinese society discusses around AI.
This is the first installment in the series. (hope I can have enough authorizations) The article is written by Jing Jun, a professor at the Institute for Advanced Studies in Humanities and Social Sciences at Sun Yat-sen University, and Tang Yongyan of the Institute for International and Area Studies at Tsinghua University.
The article’s argument is that sovereign states cannot currently achieve complete autonomy across every layer of AI—from chips, computing power, training data, and large models to talent and even regulatory rules. Even China and the United States cannot achieve complete independence and remain reliant on others. Full autonomy is therefore not a realistic policy goal for most countries. Yet the reality is that the high concentration of AI technology and infrastructure could further entrench unequal technological dependencies in the Global South. For most countries, a more realistic option is to embrace “layered sovereignty” by developing the capacity to make independent choices about critical dependencies and govern them effectively. For countries in the Global South, the key to achieving AI sovereignty is to prevent dependence on any single partner from turning into domination.
The paper was first published on Beijing Cultural Review. Thanks for their kind authorization I can put the English version of the article here.
Comprehensive AI Sovereignty: An Illusion
Jing Jun and Tang Yongyan
Introduction
By 2025, a total of 172 countries worldwide had enacted data security laws, also known as “data and privacy protection laws.”[1] As countries pay increasing attention to data security, regulatory measures intended to protect sensitive data have gradually evolved into demands for AI sovereignty. At the same time, AI research, development, and deployment are moving along increasingly divergent paths, with three major models emerging.
The first is the American model. Owing to the United States’ first-mover advantage in artificial intelligence, most countries depend on large language models, silicon-based computing chips, and cloud services developed in the United States. American technology giants have also aggressively promoted regionally tailored strategies, deepening this dependence in the name of regulatory flexibility.[2]
The second is the European model. EU countries likewise depend on American models, chips, and cloud services, but have increasingly turned to regulation to address the problems and potential risks arising from this dependence.[3]
The third is the Chinese model. In light of the Chinese government’s AI Capacity-Building Action Plan for Good and for All and China’s infrastructure-building capabilities, a growing number of developing countries are aligning their AI governance frameworks with China’s, engaging in consultation and cooperation on key issues such as risk management, cybersecurity, and standards for data flows. Technology sharing is another important component of this cooperation. Jointly building computing infrastructure, developing open-source platforms, and sharing tools such as anti-fraud applications all help countries strengthen their capacity to deploy AI technologies.[4]
What is AI sovereignty? First and foremost, it is an extension of the concepts of data sovereignty, digital sovereignty, and cyber sovereignty. This conceptual extension is evident in publication trends in international scholarship, and the same is true in China. Searches of the China National Knowledge Infrastructure database using the keywords “data sovereignty,” “digital sovereignty,” “cyber sovereignty,” and “artificial intelligence sovereignty” or “sovereign artificial intelligence” offer a partial picture of the distribution of Chinese scholarship in this field (see Table 1).
Judging by the number of publications, “AI sovereignty” remains a relatively new concept in the politics of technology. One major challenge in discussing it is the absence of an agreed definition: its meanings vary considerably across contexts. At the national level, the term generally refers, in broad terms, to a government’s aspiration to exercise national sovereignty in the field of AI. This generality and ambiguity have both disadvantages and advantages. They may entangle stakeholders in serious conflicts, but they may also give stakeholders room to discuss AI sovereignty with greater flexibility.
At its core, the demand for AI sovereignty is a demand to resist techno-political hegemony. Increasingly concerned about their heavy dependence on American digital technologies, many countries are seeking to develop independent AI industries and ecosystems. They hope to build homegrown AI models and thereby strengthen their resilience against potential risks.[5]
For example, with the introduction of the AI Continent Action Plan and the AI Act, the European Union has filled gaps in its regulatory sovereignty over AI technologies and applications, yet it still lacks technological sovereignty in AI. Conceptually, AI regulatory sovereignty refers to the independent, exclusive authority of a sovereign state, free from external interference, to legislate, enforce laws, adjudicate, and establish rules governing AI research, development, deployment, use, operation, and cross-border activities within its jurisdiction, in accordance with its own laws, values, public interests, and security needs. Its core concerns are autonomy in governance and authority over rule-setting.
AI technological sovereignty, by contrast, refers to a sovereign state’s supreme authority and capacity to independently control and develop core AI technologies, infrastructure, data, models, talent, and industrial ecosystems without external coercion. Its core concerns are technological autonomy, security, and control.
To address the lack of technological sovereignty in AI, France’s Mistral AI and Germany’s SAP have jointly launched a sovereign AI model, seeking to protect EU member states’ data through the principles and practice of AI sovereignty.[6] The United Kingdom has established an AI sovereignty research institution to promote investment in domestic AI companies and make the country more attractive to foreign AI firms.[7] India’s use of datasets in local languages to train large language models represents a pursuit of cultural diversity in generative AI.[8] The Gulf states, meanwhile, have spent heavily on large quantities of American AI chips, treating AI sovereignty as a matter of national security.[9] Chile has invested substantially in independently developed open-source models, seeking both autonomy in AI technology and the preservation of Latin American cultural independence through generative models.[10] Brazil likewise invokes AI sovereignty in regulating AI technologies and strengthening their oversight.[11] Australia, Canada, and Thailand are also discussing how AI sovereignty can safeguard national sovereignty and cultural representation.[12]
The Debate over “Layered Sovereignty”
If these forms of sovereignty are ranked by importance, technological sovereignty comes first and regulatory sovereignty second. Without technological sovereignty in practice, subsequent regulation—or discussion of any critical issue—will take place from a position of dependence. Yet AI sovereignty is by no means a static, unitary concept. It is better understood as a set of capability relationships layered within a technology stack. Chips, computing capacity, data, models, applications, supporting industries, technical standards, ethical rules, and regulatory mechanisms each have distinct structures of interdependence and allow different degrees of control.
“Comprehensive sovereignty” is more akin to a technological imaginary in the strong sense: it assumes that an actor can achieve domestic provision, autonomous operation, and effective governance at every critical layer of an AI system. It seeks a self-contained chain extending from foundational infrastructure to higher-level institutional arrangements, and may therefore be described as “full-stack autonomy” or “autonomy at every layer.”
Unlike comprehensive sovereignty, “layered sovereignty” does not take complete autonomy as its goal. It recognizes the layered structure of AI technology. Data, computing capacity, models, applications, talent, rules, and governance capabilities each constitute a layer. Countries possess uneven capabilities across these layers, and their dependencies are not concentrated in the same places.[13] The essence of layered sovereignty is therefore to identify critical dependencies, retain alternatives, and command the levers of governance. Where external dependence cannot be eliminated entirely, AI sovereignty ceases to mean a self-contained value chain. Instead, it entails identifying, controlling, and restructuring dependencies layer by layer.
In short, comprehensive AI sovereignty seeks independence across the entire chain; layered AI sovereignty emphasizes strategic choices at different layers. The former resembles an enclosed tower, striving for completeness. The latter is more like adjustable scaffolding, allowing cooperation, repair, and reconstruction.
Comprehensive AI sovereignty remains an illusion, even for the United States. It requires at least the following six conditions.
First, a complete domestic hardware production chain and infrastructure, including the capacity to produce chips across the full stack and factories for manufacturing advanced semiconductor equipment. The United States still depends on Taiwan, China, to manufacture the AI chips it needs, outsources large amounts of data annotation overseas, and must establish AI computing facilities and data centers abroad because of insufficient domestic electricity supplies. It therefore does not currently possess a complete ecosystem of AI sovereignty.
Second, domestically developed AI models. This requires training AI models on domestic data and infrastructure to achieve autonomy in model research and development. China’s development of large language models draws on American experience. The Transformer paradigm, foundational architectures, principles of data training, alignment procedures, and underlying algorithmic logic of mainstream Chinese models originated in the United States. China also continues to lack advanced, large-format AI chips. It therefore does not currently possess a complete sovereign AI ecosystem either.
Third, cloud technologies and computing capacity. Most countries currently rely on Amazon, Microsoft, and Google cloud services for related activities. This arrangement runs counter to the principle of data localization and creates numerous problems involving data security, network latency, and compliance.
Fourth, dedicated AI data centers. The United States has the largest number of conventional data centers, followed by Europe and then China. In terms of dedicated AI data centers, however, the United States and China are now almost evenly matched, with other countries trailing far behind.
Fifth, ethical principles. Ethical considerations form the foundation for regulating AI research, development, and deployment. The American government has consistently pursued a strategy of devolving ethical responsibility to companies, placing a high premium on innovation, deliberately downplaying regulation, and putting excessive faith in the market. The European Union, by contrast, has adopted stringent regulatory measures and made an exclusionary conception of European values central to its AI ethics. China’s approach to AI ethics regulation rests on the integrated pursuit of economic development, social stability, and national security. Many countries in the Global South, however, have yet to formulate clear, systematic AI ethical principles that reflect both their cultural and ethical traditions and their national circumstances. How to incorporate national sovereignty into ethical regulatory frameworks therefore remains unresolved.
Sixth, talent development. The United States, China, the United Kingdom, the core EU member states, and India are major centers of AI talent and knowledge production. Most other developing countries face substantial shortages of AI talent, compounded by brain drain, and have little choice but to depend on external chains of knowledge production.
Since no country can fully command every critical component of the AI technology stack at once, AI sovereignty must be discussed from the perspective of layered sovereignty.
The first component of AI sovereignty is the output of knowledge production and the size of the scientific and technological talent pool, reflected above all in publications, patents, frontier models, and the number of people using those models. Globally, Chinese scholars lead in the total number of AI papers, citations, and granted patents. The United States has an advantage in highly cited papers, while its technology industry produces the world’s largest number of frontier models and serves the largest global user base.
Frontier models here refer specifically to large language models with substantial industry influence, stronger performance, and a representative, leading role in the field. In 2025, 89 frontier models were released worldwide: 50 from the United States, 31 from China, five from South Korea, and one each from Canada, France, and the United Kingdom.[14] At present, only China and the United States enjoy an overwhelming advantage in their capacity to develop frontier models independently.
The second component of AI sovereignty is computing capacity and infrastructure. The infrastructure supporting computing capacity includes silicon-based chip accelerators, graphics processing units, computing clusters, dedicated AI data centers, and cloud providers’ supercomputing platforms. Where this infrastructure and computing capacity are located, who owns them, and whose jurisdiction they fall under determine the degree of autonomy in AI sovereignty.
According to publicly available data, by 2025 only 32 countries worldwide had deployed AI computing systems. These systems can be understood as regional hubs of AI computing data centers. The United States had 26 such hubs, China 24, and Germany seven; no other country had more than five.[15] Apart from China, countries that have established AI computing data centers rely primarily on silicon-based chips supplied by the American company NVIDIA. China is expected to achieve a domestic chip supply rate of 70 percent in 2026.
The third component of AI sovereignty is the regulatory model and its guiding principles. The three principal forces driving advances in large language models and AI tool models today are the United States, China, and the European Union. As noted above, the American federal government takes a market-led approach to regulation, the Chinese government a state-led approach, and the European Parliament, as an EU legislative institution, a human-rights-led approach.[16]
The American model is characterized by technological optimism and the pursuit of innovation. It maintains an uncompromising faith in the market, advocates a limited role for government, and regards subnational legislation that impedes AI research and development as unconstitutional.
The Chinese model follows a state-led approach. It seeks to maximize the country’s nationwide technological strengths while maintaining social stability and social equity, and upholds the position that AI should benefit all countries, particularly those in the Global South.
The EU model bases regulation on fundamental human rights and adheres to the principle of fair market competition. It imposes stringent controls on high-risk uses affecting human health, personal safety, and citizens’ fundamental rights. Specifically, these restrictions encompass the use of AI in critical infrastructure, the collection and use of biometric data, assessments of reoffending risk, election campaigning, credit scoring, résumé screening, student admissions, and determinations of eligibility for welfare benefits.
The three regulatory models represented by the United States, China, and the European Union exert varying degrees of influence across different countries. The American model faces resistance from the European Union and many developing countries. The EU model provides a reference point for AI ethical principles in most Western countries. The Chinese model is gaining ground in the Global South. Yet a stark reality remains: the world’s internet information flows primarily through platforms controlled by American technology companies. This presents a formidable obstacle to developing countries’ aspirations and efforts to build sovereign AI models and systems.
The fourth component of AI sovereignty is well-developed internet infrastructure and a high rate of internet penetration. In 2025, AI adoption exceeded one-quarter of the population in 30 countries, most of which also had high internet penetration rates. Judged solely by its AI adoption rate in 2025, China’s performance was remarkable. Its rate of 42.8 percent, although below those of the United Arab Emirates (59.4 percent), Singapore (58.6 percent), and Norway (45.3 percent), was clearly above those of France (40.9 percent), the United Kingdom (36.4 percent), and the Netherlands (36.3 percent), and far above those of Germany (26.5 percent) and the technologically powerful United States (26.3 percent).
It must nevertheless be emphasized that, although domestic AI adoption in the United States was far lower than in China and ranked behind 22 other countries, American technology giants remained dominant in the circulation, collection, storage, mining, and use of information across the global internet and social media. Two billion users worldwide send 100 billion messages each day through Meta’s WhatsApp.[17] More than 70 percent of active social media users worldwide use Meta platforms to post, comment, and converse.[18] Google holds more than 80 percent of the global search engine market, and internet users worldwide conduct more than five billion Google searches daily.[19]
American technology companies not only exert overwhelming influence over internet and social media users worldwide—with exceptions in a small number of countries, most notably China—but also continually export, through their products and services, the market-oriented values and rules embodied in the American regulatory stance. For political reasons, the United States tends to export AI chips to selected allies, while the European Union chooses its partners according to human-rights-oriented values. Despite the emergence of cooperation on an equal footing among a number of developing countries, international AI cooperation remains constrained by bloc formation and fragmentation under the influence of geopolitics and great-power rivalry.
The importance of these four components depends on specific contexts and changing circumstances; it cannot be determined simply by assigning them fixed rankings or weights. Together, they are shaping four forms of AI sovereignty: independent, cooperative, dependent, and externally administered.
These four forms depend to a considerable extent on infrastructure development, chip research and manufacturing, the development of leading models, the diffusion of applications, the orientation of government policy, data protection rules, and the degree of autonomy in technological innovation and the formulation of laws and policies.
Independent AI sovereignty requires full national sovereignty. The state must exercise effective control over domestic data flows, possess adequate domestic infrastructure, have a legal and policy framework supported by comprehensive institutional mechanisms, maintain a reasonably robust domestic innovation ecosystem, and be able to independently establish technical standards and principles for international cooperation. Relatively speaking, the United States, China, and EU member states belong to this category.
Cooperative AI sovereignty is characterized primarily by reciprocal interdependence and technological partnerships. Through bilateral or multilateral cooperation, countries strengthen their autonomy in regulation, infrastructure development, and technological innovation. Singapore, South Korea, the wealthy Gulf states of the Middle East, and South Africa are relatively clear examples of this form.
Dependent AI sovereignty is characterized by a governance orientation shaped by external forces, limited control over data, underdeveloped regulatory systems, reliance on external actors to build infrastructure, weak technological innovation capabilities, heavy dependence on imported technologies, and unequal partnerships. Many low-income countries are at this stage.
Externally administered AI sovereignty is typified by extremely impoverished countries and regions that depend heavily on foreign assistance projects for regulatory rules, infrastructure development, and the use of technology.
The arrival of new AI technologies is widening the digital divide between the Global North and the Global South. In 2025, global AI adoption reached 16.3 percent, compared with 24.7 percent in the Global North and 14.1 percent in the Global South. Of the 30 countries where AI adoption exceeded one-quarter of the population, only five were in the Global South. Within just a few years, the gap in AI adoption between developed and developing countries has widened: among ASEAN countries, Singapore’s adoption rate was 12.74 times that of Cambodia.
Technological precedence does not necessarily translate into the widest use of a technology, as the difference in AI adoption between China and the United States illustrates. It must nevertheless be pointed out that the present American government and major American technology companies are attempting to turn AI technologies and digital platforms into new historical and geographical centers of global power.
As early as the colonial era, large corporations served as the advance guard of European powers, both driving imperial expansion and profiting from it. These companies generally operated with the support of their governments, trading in new territories under royal charters and, in some cases, even possessing the power to wage war. In the British Empire alone, the British East India Company controlled India’s economic and political lifelines and played a crucial role in the Opium Wars between Britain and China. The Royal African Company established trading posts along the West African coast and became a major force in the slave trade. The British North Borneo Company controlled extensive tracts of land and traded in timber, rubber, and tobacco, serving as a mainstay of British expansion into Southeast Asia.
The colonial powers of the past controlled tangible centers of global historical and geographical power, competing in succession for land, gold, spices, furs, minerals, oil, sugar, coffee, opium, and other forms of material wealth and resources. Today, new forces of digital colonialism are competing for electronic information flows on the internet.
According to the United Nations Conference on Trade and Development’s Digital Economy Report 2019, Google already held 70 percent of the global search engine market at that time, Facebook 60 percent of the global social media market, and Amazon 40 percent of the global online retail market.[20] The profits these three companies generated in North America and worldwide depended to a considerable extent on market monopolies.
In recent years, the American government has repeatedly expressed its intention to use antitrust law to break up Google, Facebook, Amazon, Microsoft, and Apple, yet no such action has materialized. One central reason is that these technology giants have displaced the financial conglomerates of the past as leading forces in American lobbying. These companies are also important instruments through which the United States maintains its global dominance in the digital economy.
Through proprietary software, corporate cloud servers, and centralized internet services, large technology companies can track users worldwide around the clock. Search engines, maps, advertising, location services, email, and AI recommendation systems become points of data entry, channeling social, cultural, political, economic, biometric, environmental, health, and consumer data from around the world into corporate clouds under their control. Many countries in the Global South are unable to develop domestic information technology industries capable of competing with American companies, or to protect their populations from the domination and exploitation of American digital colonialism.[21]
Conclusion
China’s international standing is closely bound up with the Global South. So, too, are the greatest opportunities and strategic depth for the international expansion of Chinese AI technologies. In its AI Capacity-Building Action Plan for Good and for All, the Chinese government declares that China will actively assist countries, particularly those in the Global South, in developing AI technologies and services, helping the Global South gain genuine access to artificial intelligence.[22]
China’s proposed action plan comprises the following ten points:
1. China is willing to work with all countries on North–South, South–South, and triangular cooperation in AI to jointly implement the outcomes of the United Nations Summit of the Future. It will actively cooperate with countries, particularly developing countries, on AI infrastructure and the establishment of joint laboratories.
2. China is willing to cooperate on AI model research and development and on AI-enabled development, particularly the application of AI to poverty reduction, healthcare, agriculture, education, and industrial manufacturing. It seeks to deepen international AI cooperation and unlock the benefits of AI as a new quality productive force.
3. China is willing to work with countries, particularly developing countries, to explore AI’s potential to advance green development, respond to climate change, and protect biodiversity, contributing to global climate governance and sustainable development.
4. China is willing to establish an international cooperation platform for AI capacity-building. Chinese AI industries and industry alliances are willing to engage in exchanges of various kinds with countries, particularly developing countries, share best practices, responsibly build open-source and open AI communities together, and foster a cooperative ecosystem spanning multiple levels and forms of activity.
5. The Chinese government will organize short- and medium-term AI capacity-building education and training programs for developing countries, share AI educational resources, and undertake joint educational programs, visits, and exchanges, helping developing countries cultivate high-level talent in AI science, technology, and applications.
6. The Chinese government is willing to strengthen cooperation with developing countries on human resource assistance. By the end of 2025, it will hold ten additional AI training and seminar programs primarily for developing countries.
7. China is willing to work with countries, particularly developing countries, to improve public AI literacy. By combining online and offline approaches, it will disseminate accessible and specialized AI knowledge across multiple dimensions, levels, and media, striving to improve people’s AI literacy and skills, with particular attention to safeguarding and strengthening the digital rights and interests of women and children.
8. China is willing to work with countries, particularly developing countries, to build AI language corpora, take active steps to eliminate racial, algorithmic, and cultural discrimination, and preserve and promote linguistic and civilizational diversity.
9. China is willing to work with countries, particularly developing countries, to develop and improve data infrastructure and jointly promote the equitable and inclusive use of data worldwide.
10. China’s AI capacity-building action plan seeks to actively help countries in the Global South benefit equally from AI development. It aims to advance AI-enabled poverty reduction, healthcare, agriculture, education, and industrial manufacturing; help developing countries cultivate high-level AI scientific and technological talent; promote the equitable and inclusive use of data; and bridge the global digital divide.
The layered approach to AI development embodied in these ten commitments to cooperation is unmistakable, and it already has a practical foundation. AI infrastructure and application projects supported by Chinese companies through South–South technological cooperation primarily encompass open-source models, medical imaging, digital government, and smart cities. Cooperation in some areas has already entered practical use.
The first example is earthquake monitoring. The Institute of Geophysics and the Development Research Center of the China Earthquake Administration, together with Nepal’s Department of Mines and Geology, jointly undertook the construction of the National Earthquake Data Center in Kathmandu. Ten seismic observation stations were installed in northern Nepal to monitor and provide warnings for earthquakes of magnitude three and above.
The second is satellite launches. China’s Ministry of Commerce signed an agreement with Egypt’s National Authority for Remote Sensing and Space Sciences to establish a satellite assembly, integration, and testing center. Chinese technicians were responsible for installing and commissioning all new equipment supplied to Egypt and assisted the Egyptian authority in launching a satellite.
The third is high-performance computing. The Chinese government provided an interest-free loan for Zimbabwe’s High-Performance Computing Center project. Through collaboration between technicians from both countries, a supercomputer was installed at the University of Zimbabwe for use in agriculture, mining, genetic technologies, and weather forecasting simulations.
The fourth is national data center construction. Bangladesh’s National Data Center, built with assistance from ZTE, received international Tier IV certification, attaining the highest level of data center performance, with annual downtime not exceeding 30 minutes.
The fifth is education. Xidian University worked with Equatorial Guinea to jointly establish the Oyala Institute of Information Technology and Cybersecurity. It is especially important to emphasize the value of AI education for low- and middle-income countries in the Global South. Without education as a foundation, the prospects for AI development in the Global South will be substantially diminished.
Across the world today, the power of artificial intelligence is opening new possibilities for development. Yet most countries in the Global South face three interlocking predicaments: first, inadequate infrastructure, insufficient funding, and a scarcity of talent; second, a lack of voice in the formulation of governance rules; and third, insufficient preparation on ethical principles and the absence of established frameworks. Breaking this interconnected impasse requires a firm commitment to ensuring that AI benefits the peoples of all countries and to resisting exclusionary practices in the politics of technology.
At present, almost no country has secured sovereignty across every layer of AI. All must determine which links in their industrial and application chains are most critical. In the foreseeable future, China and the United States may be the only countries with a prospect of achieving AI sovereignty at every level of both chains.
It must be emphasized here that South–South cooperation in AI necessarily involves questions of national sovereignty. Layered sovereignty represents both a realistic choice and a strategic approach: South–South cooperation requires carefully designed strategies differentiated by sector, stage of development, national circumstances, culture, and political system, together with sustained efforts in talent development, infrastructure, appropriate regulation, and cultural sensitivity.
Notes
* This article presents interim findings from the National Social Science Fund of China’s Key Project, “Research on the Construction of a Southern School of Social Theory” (22ASH001).
[1] Graham Greenleaf, “172 Countries with Data Privacy Laws—Year by Year 1973–2025,” Privacy Laws & Business: International Report, No. 195, June 2025, pp. 16–17.
[2] Filippo Gualtiero Blancato and Madeline Carr, “The Trust Deficit: EU Bargaining for Access and Control over Cloud Infrastructures,” Journal of European Public Policy, Vol. 33, No. 3, 2026, pp. 936–967.
[3] Daniel Mügge, “EU AI Sovereignty: For Whom, to What End, and to Whose Benefit?” Journal of European Public Policy, Vol. 31, No. 8, 2024, pp. 2200–2225.
[4] Cai Jiajun, “The Foundations, Characteristics, and Recommendations for Improving International Cooperation on Artificial Intelligence between China and Other Global South Countries,” World Science and Technology Research and Development, No. 4, 2025. [In Chinese.]
[5] Wang Tianchan, “‘Sovereign AI’ Strategies of Middle Powers: A Comparative Analysis of the Netherlands, Singapore, and South Korea,” Global Review, No. 1, 2026. [In Chinese.]
[6] SAP News Center, “SAP and Mistral AI: A New Alliance for European Sovereignty,” SAP, November 18, 2025.
[7] Department for Science, Innovation and Technology, “AI Firms Pioneering Drug Discovery, Cheaper Supercomputing and More Get First Backing through UK’s Sovereign AI,” GOV.UK, April 16, 2026.
[8] Sarvam AI, “Sarvam to Build India’s Sovereign Large Language Model,” Sarvam AI, April 26, 2025.
[9] U.S. Department of Commerce, “Statement on UAE and Saudi Chip Exports,” November 19, 2025.
[10] Associated Press, “Chile Launches Open-source AI Model Designed for Latin America,” AP News, 2026.
[11] “Bill No. 2338 of 2023,” OECD.AI, April 27, 2026.
[12] Australian Government Department of Industry, Science and Resources, “Capture the Opportunities: National AI Plan,” 2025; Innovation, Science and Economic Development Canada, “Canadian Sovereign AI Compute Strategy,” April 15, 2026; National Science and Technology Development Agency, “Thailand Launches ‘ThaiLLM’: A Sovereign AI Foundation Model for Thai Organizations,” April 1, 2026.
[13] J. N. Pava, C. Meinhardt, E. Cryst, and J. A. Landay, “AI Sovereignty’s Definitional Dilemma,” Stanford Institute for Human-Centered Artificial Intelligence, February 17, 2026.
[14] Stanford HAI, AI Index Report 2025, 2025.
[15] Oxford University researchers, cited in “Global Divide: Only 32 Countries Host AI Data Centers,” TechRepublic, June 23, 2025.
[16] Shangrui Wang et al., “Artificial Intelligence Policy Frameworks in China, the European Union and the United States: An Analysis Based on Structure Topic Model,” Technological Forecasting and Social Change, Vol. 212, 2025, article 123971.
[17] Statista, “Most Popular Global Mobile Messenger Apps as of January 2025,” 2025.
[18] iiMedia Research, “2025 Global Social Media User Rankings,” November 25, 2025; Sensor Tower data, cited in “The State of Social Media: Instagram and Facebook Still on Top,” The Wrap, June 11, 2025.
[19] StatCounter, “Search Engine Market Share Worldwide 2025”; Statista, “Global Search Engine Market Share 2025,” January 5, 2026. The estimate of Google’s daily search volume is based on standard estimates from industry data aggregators such as TechJury and HubSpot. See also Microsoft AI Economy Institute, Global AI Adoption 2025, January 2026.
[20] United Nations Conference on Trade and Development, Digital Economy Report 2019, UNCTAD, 2019.
[21] Marcus Vinicius de Freitas, Digital Sovereignty and Data Colonialism: Shaping a Just Digital Order for the Global South, Policy Paper No. 2519, Policy Center for the New South, 2025.
[22] Zu Wentian, “Cooperation in Global Artificial Intelligence Governance: Issues, Pathways, and China’s Participation,” International Economic Review, No. 4, 2025. [In Chinese.]
Authors’ affiliations: Institute for Advanced Studies in Humanities and Social Sciences, Sun Yat-sen University; Institute for International and Area Studies, Tsinghua University.