|
|
Middle powers
|
|
Canada
|
8th
|
|
•
|
•
|
|
|
|
|
France
|
5th
|
|
•
|
•
|
|
|
|
|
Germany
|
7th
|
|
•
|
•
|
|
|
|
|
Japan
|
11th
|
|
•
|
•
|
|
•
|
|
|
India
|
10th
|
|
•
|
•
|
|
|
|
|
Indonesia
|
49th
|
|
•
|
•
|
|
•
|
|
|
Israel
|
9th
|
•
|
•
|
•
|
•
|
|
|
|
Saudi Arabia
|
14th
|
|
•
|
•
|
•
|
•
|
|
|
Singapore
|
3rd
|
•
|
•
|
•
|
|
|
|
|
South Korea
|
6th
|
|
•
|
|
|
|
|
|
Spain
|
18th
|
|
•
|
|
|
•
|
|
|
Sweden
|
25th
|
|
•
|
•
|
|
•
|
|
|
Thailand
|
43rd
|
|
•
|
•
|
|
•
|
|
|
United Arab Emirates
|
20th
|
|
•
|
•
|
•
|
•
|
|
|
United Kingdom
|
4th
|
•
|
•
|
•
|
|
|
|
|
Vietnam
|
58th
|
•
|
•
|
•
|
|
|
|
|
European Union
|
–
|
|
•
|
•
|
•
|
•
|
|
|
AI superpowers
|
|
United States
|
1st
|
•
|
•
|
•
|
•
|
•
|
•
|
|
China
|
2nd
|
•
|
•
|
•
|
•
|
•
|
•
|
Source: Compiled by the authors, based on Tortoise (2024), ‘The Global AI Index’, https://www.tortoisemedia.com/data/global-ai#rankings.
Note: The Global AI Index ranks 83 countries based on their international AI capacity. They are scored based on 83 indicators, falling within three umbrella areas: implementation, innovation and investment.
National security
For most countries, national security is a pressing reason for building and investing in domestic AI capabilities. The US frames technological leadership as essential to its national security – seeking to onshore critical infrastructure, export AI only to trusted allies and restrict the supply of advanced chips to ‘countries of concern’. The Stargate Project, which was announced in 2025, embodies this strategy, combining massive domestic investment ($500 billion) in AI infrastructure with efforts to curb China’s competitiveness. Likewise, China’s approach is security-driven, focused on regime stability, public safety and control over data and AI model use. China’s interim measures for governing AI acknowledge the importance of ‘placing equal emphasis on development and security’.
Focusing on middle powers, the UK’s approach to sovereign AI is supported by a different national security rationale. The country’s Strategic Defence Review 2025 commits to ‘innovat[ing] at a wartime pace’, denoting advanced AI systems as essential for next-generation defence. Meanwhile, the AI action plan identifies three compute (computational power) tiers: publicly owned or allocated, privately owned but UK-based, and international via foreign partners. While fully homegrown AI capabilities might be preferential from a national security perspective, for the UK, such developments are not realistic in economic or practical terms.
Israel also harbours ‘AI superpower’ ambitions, notably captured by its AI National Program. Its sovereign AI strategy is highly securitized. The country has long prioritized building technology and cybersecurity hubs for international trade and to bolster its national security. Its investments in sovereign AI capabilities – for example, in military robotics and autonomous warfare – fulfil both aims.
For other middle powers, a growing national security driver of sovereign AI is the legal and jurisdictional risk of relying on foreign cloud and AI providers, especially those in the US. Mistrust of US data governance and technology infrastructure shapes middle powers’ national AI and cloud procurement strategies. This anxiety is pushing countries towards sovereign cloud solutions, domestic compute capacity and open-source models that reduce exposure to foreign jurisdictions and regulations.
Concerns about the US CLOUD Act 2018 – which allows American authorities to compel US-headquartered companies to provide access to data stored both in the US and abroad – are growing, even among allies.
Concerns about the US CLOUD (Clarifying Lawful Overseas Use of Data) Act 2018 – which allows American authorities to compel US-headquartered companies to provide access to data stored both in the US and abroad – are growing, even among allies. In November 2025, a Canadian government white paper warned that the country’s data sovereignty is at risk if legal control over data stored in Canada is claimed and acted upon by another jurisdiction. In the same month, the European Commission launched market investigations into cloud computing services from Amazon and Microsoft to determine if their practices limit competitiveness in the EU cloud computing sector, in accordance with the Digital Markets Act’s provisions.
China’s data governance regime – including the mandatory requirement for companies storing data within its jurisdiction to enable state access – has long sparked global concern. Anxieties about over-dependence on China-supplied technology – and its implications, such as surveillance and strategic risks – have shaped technology policies in London, Brussels and Washington.
Economic growth
Building domestic capabilities for national economic growth is another prevalent rationale in the AI sovereignty strategies reviewed for this paper. Vietnam’s 2021 strategy on AI outlines the country’s aspirations to be a ‘center for [AI] innovation’, and to become one of ASEAN’s top four nations on AI R&D and applications by 2030. The country’s strategy commits to developing a competitive AI ecosystem – including attracting investment and mobilizing capital – and promoting the application of AI in businesses. In December 2024, the Vietnamese government and major semiconductor designer and producer NVIDIA agreed to open a new AI R&D centre, focusing on software development to accelerate AI adoption. AI automation in manufacturing and agriculture, two of Vietnam’s major industries, could be highly beneficial for the country.
Japan’s approach to building sovereign AI capabilities also centres on economic advantages and productivity. Recent deals with NVIDIA commit to public, private and academic partnerships geared towards accelerating adoption and ‘homegrown’ innovation. A common theme of these new initiatives is a commitment to protect Japan’s data sovereignty with tailored solutions. The country’s approach to AI benefits from the population’s ‘high degree of affinity with generative AI’ due to existing digitalization, the robotics industry, technology R&D and is partly motivated by ‘a sharp decline in the working population’. This state of play highlights twin drivers of sovereign AI investments: a sober assessment of future productivity challenges and existing economic advantages.
Further afield, Canada’s Sovereign AI Compute Strategy directly responds to the constraints faced by middle powers: high costs and limited domestic capacity in compute. The country’s strategy outlines major commitments to public supercomputing infrastructure and an AI Compute Access fund as drivers of economic growth.
Moving to Europe, Germany was one of the first countries to publish an AI strategy, launched in 2018. The strategy calls for Made in Germany AI (or Made in Europe AI) to leverage existing economic strengths to create both value and societal benefit. This strategy links with efforts in other areas designed to enhance German and European AI leadership, such as upskilling and training strategies, as part of a €5 billion federal package on AI. As the country seeks to build domestic AI capabilities and improve its competitiveness, Germany’s sovereign AI rationale depends on coordination with other middle powers in the EU bloc.
Public service capacity-building
Many countries view the development of sovereign capabilities in AI as an enabler of other public policy goals. These efforts are often consistent with whole-of-government digitalization strategies. This is certainly the case in Singapore, India, Indonesia and Thailand, four South and Southeast Asian countries with markedly different political systems and levels of digitalization, but similar public capacity ambitions for AI.
Singapore’s Smart Nation initiative aims to advance the digitalization and therefore efficiency of public services with an ‘AI for public good’ approach, from healthcare and housing to security and immigration control. India is also a hub of growing digital public infrastructure (DPI) solutions – secure, open and interoperable digital systems – across the public sector, such as the country’s digital ID system, Aadhaar. India’s strategic approach to building sovereign AI capabilities is similarly rooted in public capacity ambitions, particularly in relation to AI for social welfare in healthcare, agriculture and linguistic diversity. Despite their economic and infrastructural differences, Singapore and India’s strategic approaches clearly promote the use of AI to strengthen governing capacity.
For Indonesia, Southeast Asia’s largest digital market, investments in sovereign AI are viewed as essential for catering to the country’s different cultures and languages, building public capacity and opening the door to AI applications. The multilingual Sahabat-AI models are a case in point: a collection of open-source large language models (LLMs) ‘built by Indonesians for Indonesians’, which are the result of a public–private partnership involving NVIDIA. The country’s latest strategy emphasizes the need to avoid foreign dependencies and identifies five priority areas – all in the public sector – including bureaucratic reform and smart cities (using digital technology to improve urban living). Indonesia’s forthcoming AI roadmap is expected to provide further clarity.
Finally, Thailand’s strategic approach to developing sovereign AI links the building of public capacity with overall regional competitiveness. Thailand’s ambition is, like Indonesia and Singapore, to become a leading AI hub. Strengthening governing capacity is critical to achieving this goal. Annual reporting by the Thai government recognizes the importance of ‘AI readiness’ in policymaking, while the country’s strategy develops in line with the number of users (in the public and private sectors) adopting AI for innovation. Front and centre of the national strategy is the potential for AI-powered public services to improve well-being, quality of life and governing efficiency. LLMs developed for the Thai language may accelerate these efforts.
Leverage geopolitical competition
Building domestic AI capabilities can also further geopolitical and strategic aspirations. While the US–China AI race is a major geopolitical consideration in developing these technologies, it is by no means the only factor. Seeking to maximize their AI sovereignty – without necessarily striving for global AI dominance – middle powers must navigate competing offerings from leading AI providers. Some middle powers, such as the United Arab Emirates (UAE), have picked a side. Others are more agile in navigating the geopolitical crosswinds of US–China AI competition, aiming to build a third way.
The US and China have long appreciated the geopolitical value of the UAE, due to its access to cheap electricity and significant public wealth – including a commitment to invest $200 billion in AI infrastructure. In the last year, however, the UAE has moved closer to the US, leveraging its position for an unprecedented deal: the country will host OpenAI’s first international deployment of Stargate. This deal is part of a series of major investments into data centres – with the 1 gigawatt (GW) Stargate UAE cluster in Abu Dhabi – and strategic partnerships with US and UAE AI companies, including the Acceleration Partnership deal with the US government. For the UAE, the reasoning is clear: geopolitical competition can be leveraged to secure investments and partnerships for the country’s sovereign AI.
Saudi Arabia’s newly launched sovereign AI initiative – HUMAIN, a Public Investment Fund-owned company – faces a similar geopolitical landscape but adopts a slightly different rationale. The country’s national strategy outlines Saudi Arabia’s aspirations to be a global leader in AI by 2030, with the aim of attracting at least $20 billion in investment. HUMAIN emphasizes Saudi independence (from US and Chinese AI offerings) with major investments in local, Arabic LLMs. At the same time, Chinese technology companies – particularly Huawei Cloud – are proactive in Saudi Arabia, supporting government cloud usage and brokering strategic partnerships across the telecommunications sector. Analysis suggests this two-pronged approach means Chinese AI has a central role in public–private partnerships while strictly adhering to local data laws. This presence is a sovereign AI puzzle for Saudi Arabia, building sovereign capabilities but reckoning with intense dependencies on China-supplied infrastructure.
South Korea’s approach focuses on more investment. Gaining geopolitical autonomy is a priority issue for the country, having announced a $75 billion investment in sovereign AI in June 2025, with a new AI policy unit and a dedicated presidential secretary on AI. Crucially, the lead strategist has called for these investments in an effort to achieve ‘full-stack’ sovereignty – that is, end-to-end control of the AI system so no critical part depends on a foreign company or government – and freedom from the ‘neo-imperialism’ of the US–China technology race.
Alignment with national values
National strategies on technology and sovereignty are heavily influenced by societal, ethical and political values, alongside discrete notions of public good. These can significantly contrast between nations, which has implications for cooperation.
EU efforts to develop sovereign AI illustrate the core considerations of values alignment for democracies. The EU’s 2025 AI Continent Action Plan champions democratic values, cultural diversity and trustworthy and human-centric AI. Across the EU, efforts have focused on the development of open, multilingual models that cater to European languages to enable broader societal input, which can better reflect local values. For example, the Open Euro LLM, which is a series of models that promote transparency in AI, seeks to build and improve access to multilingual foundation models that can then be fine-tuned for local applications. Similarly, AI Singapore’s Southeast Asian Languages in One Network (SEA-LION) is a multilingual AI model that aims to provide LLMs with a better understanding of ‘diverse contexts, languages and cultures’. Such cases underline how regional or international blocs can collectively build sovereign AI.
At the country level, there are also efforts to uphold cultural and linguistic diversity in sovereign AI initiatives, from Sweden’s GPT-SW3 to Spain’s Alia (available in Catalan, Valencian, Galician and Basque, in addition to other European languages). Switzerland’s Apertus initiative provides a fully open, public model designed to maximize auditability and multilingual support while avoiding reliance on foreign commercial providers. These efforts correlate with rising concerns about the deployment of models from the US and China with ‘non-European’ values. China’s DeepSeek, for example, is facing bans in German app stores due to data protection concerns.
Country-level efforts correlate with rising concerns about the deployment of models from the US and China with ‘non-European’ values.
Both NVIDIA and OpenAI promote international partnerships based on values alignment. NVIDIA touted localization and respect for national values as the basis of its latest series of agreements and partnerships in Asia. Likewise, as part of Stargate, the OpenAI for Countries initiative promotes ‘build[ing] on democratic AI rails’. That said, its first official partner, the UAE, is a federation of monarchies. Non-democracies are motivated by value alignment, too, with prevailing concerns about content control over the outputs of foreign-developed AI models and the subsequent implications for regime stability and public order. Consequently, ChatGPT does not operate in China. Sovereign AI capabilities can be repurposed for digital authoritarian ends, giving governments the tools to automate repression and surveillance.
Indonesia’s sovereign AI model (Sahabat-AI) caters to local languages in an effort to align AI with national values and facilitate local applications.
The strategies reviewed for this paper capture attempts to align the development of sovereign AI capabilities with national values. These values range from interpretations of public good and societal benefit to ideas about political organization, rights and freedoms.
Domination of the frontier
Only the US and China seek to dominate the frontier, or most advanced, AI tools. For these two countries, building domestic AI capabilities is about readiness for – and protection from – future technological disruption, triggered by exceptionally advanced AI capabilities or artificial general intelligence (AGI). The term AGI, regardless of its likelihood or achievability, refers to machine cognition that matches human-level reasoning. It is argued that such a development would fundamentally restructure global power dynamics, including economies and military capabilities. Research and public documentation suggest China has been working towards AGI since 2017, although the country’s policy priority appears to be accelerating the diffusion of less advanced AI for everyday uses. Leading US AI companies also profess their dreams to attain AGI capabilities and, in their words, ‘transform the world for the better’.
For leading AI powers, the pursuit of frontier capabilities and AGI functions as a tool for shaping global norms, securing first-mover advantages and influencing international standards. Investments in frontier AI are not just about technological advancements, these projects carry strategic intent, underpin economic competitiveness and create leverage in scientific collaboration and talent attraction.
For leading AI powers, the pursuit of frontier capabilities and AGI functions as a tool for shaping global norms, securing first-mover advantages and influencing international standards.
Are middle powers motivated to attain frontier or AGI capabilities? It is certainly in their strategic interest to build readiness for – or even stewardship over – future technological disruptions. However, joining the global race towards AGI might not be. French AI company Mistral, for example, criticises the ‘religious’ obsession with AGI, while simultaneously promoting itself as a frontier AI company. France’s strategic approach to sovereign AI capabilities prioritizes making the country an AI ‘powerhouse’ and R&D for pioneering innovation in a limited number of scientific and technological fields with direct societal benefits, like improved healthcare.
Other middle powers engage with frontier AI selectively, often concentrating on specific subcategories with clear societal, economic or industrial value. This sectoral approach allows middle powers to maintain awareness of cutting-edge developments, contribute to scientific knowledge and support domestic innovation ecosystems, without the resource burden of competing directly with superpowers in the AGI race.
Engagement with frontier AI also provides middle powers with strategic flexibility. By cultivating technical expertise and research capacity in select areas, these states can better understand the capabilities and risks of emerging AI technologies, anticipate potential disruptions and align investments with national priorities. Such selective engagement reflects a broader pattern in the global AI landscape, where countries balance ambition, capacity and practical benefits.
While superpowers pursue comprehensive AGI and frontier strategies for global influence, middle powers – given their constraints – are more likely to focus on readiness, selective leadership in niche domains and ensuring that emerging capabilities can be harnessed for societal, economic or industrial benefit. This distinction highlights the diversity of motivations driving sovereign AI development and the different pathways that nations take to participate in the global AI ecosystem.
