States are taking steps to improve national AI capabilities and technological control. But with the US and China dominating AI, complete independence remains unreachable. Middle powers need pragmatic pathways to seek meaningful AI sovereignty.
The drive for sovereignty
Strengthening the capacity of states to exercise power over digital technology is not a new challenge. For decades, governments have grappled with the ways that digital technology has reshaped their economies, societies and cultures, and the limits of their own national capacity to influence that transformation.
Artificial intelligence (AI) has reinvigorated debates over what constitutes sovereignty in relation to technology. This is especially true among middle powers, which, in the context of AI, are understood as countries that cannot compete with the US and China in the AI race, but nonetheless aim to influence AI’s development, deployment and governance – both on a national and international level.
At the AI Action Summit in Paris in February 2025, French president Emmanuel Macron described the future of AI as ‘presenting a political challenge, a question of sovereignty and strategic self-determination’, later remarking that ‘there is no such thing as happy vassalage’. Various EU policy initiatives, such as the International Digital Strategy, have called for the bloc to commit to sovereign digital infrastructure as a strategic imperative for becoming more technologically independent and to strengthen security across the region. The International Digital Strategy was published soon after the new EU AI Continent Action Plan, which outlined steps for the EU to become a global leader in AI.
Dozens of other countries have now set out their own national strategies for ‘sovereign AI’ – the ability of a country to influence, develop and deploy AI in line with national interests. The UK’s AI Opportunities Action plan promises major sovereign AI infrastructure investment, international partnerships and incentives to bring AI companies onshore; prompting commitments for more than £14 billion in fresh inward investment across the country. Further afield, Saudi Arabia has set out its ambition to become an AI superpower by 2030 through capital investments, headlined by the launch of a state-owned company and major US deals on AI and defence. No two countries’ sovereign AI strategies are completely alike.
But strategic ambitions aside, no middle power can feasibly compete with the scale and expenditure of US and Chinese sovereign AI efforts.
How states – regardless of size and influence – might fit into an AI world order is still unclear. Some voices – notably major US AI companies – argue that the world is on the cusp of an AI revolution that will result in the greatest transformation of political and economic power in human history. Sceptics are less sure, both on the timeline and the scale of the projected change. For governments, their sovereign AI strategies represent attempts to gain leverage over a technology expected to shape economic competitiveness, national security and societal resilience.
This paper’s core objective is a pragmatic assessment of the gap between the current AI superpowers – the US and China – and middle powers, and to highlight realistic strategies for middle powers to build and exercise AI sovereignty in that context.
Full AI self-sufficiency – even for AI superpowers – is not possible. AI’s global supply chains are too interconnected, and interdependencies run deep. Middle powers are severely constrained on nearly every critical input into building sovereign AI, from talent and data to raw materials and semiconductor chips. We conclude that complete technological independence is unachievable for most countries, but certain levels of control and leverage are possible.
Chapter 2 maps the rationales for middle power investments in sovereign AI, ranging from national security to economic competitiveness. Chapter 3 proposes that the abilities of middle powers to exercise sovereignty over primary (the technical foundations) and secondary (the enabling environment) building blocks for AI are constrained by entrenched dependencies on the US and China. Chapter 4 argues for an urgent reset in strategic thinking, sketching out pragmatic, informed pathways for different forms of sovereign AI. The recommendations have relevance to a diversity of national contexts excluding the US and China.
Methodology
For this paper, the authors conducted research on policy and strategic documentation associated with national approaches to AI, contextualizing this information with secondary sources (news releases, scientific reports, company updates, analysis and commentary from media, research institutes and universities). Countries were selected on the basis of diversity in their AI implementation, innovation and investment (gauged by the Global AI Index 2024) and the prospects for regional cooperation (particularly in Europe and Southeast Asia). An important part of these criteria is the presence of a national strategic plan on, or related to, AI. EU-wide AI initiatives are included alongside national plans.
The authors supplemented and validated this research with anonymized, informal, semi-structured interviews with middle power sovereign AI builders from several of the case study countries and workshops, including representatives from the EU (France and Belgium), ASEAN (Thailand, Singapore and Vietnam), and the UK. The paper’s approach also benefits from insights gathered during the Paris AI Action Summit and the International Association for Safe and Ethical AI Conference, both held in February 2025.
All countries have a stake in the development of AI. This paper’s approach and recommendations are designed to encourage strategic thinking among a diverse array of policymakers. The work should be of particular interest to strategic decision-makers in middle powers. Exercising power over AI is a new and unfamiliar undertaking, with few options and many constraints. But there are also emerging pathways for states to carve out autonomy in the sector, including new multilateral forums for decision-making, the proliferation of big data and increasingly globalized supply chains. The nature of AI as a networked technology with multiple inputs – not to mention its entanglement in geopolitical dynamics – means that any ambition to deploy AI in the domestic interest within national borders has an inescapably international strategic dimension.
