Treating AI as an Arms Race frames the technology as a zero-sum geopolitical contest rather than recognising its potential as a broad-based driver of economic growth. To realise the benefits of AI, nations should focus on securing their place within complex, shifting global production networks and addressing vulnerabilities in the infrastructure of the AI stack. True economic security requires moving past industry hype and military narratives to ensure AI delivers tangible benefits to society.
Beyond a military metaphor
Why is the AI arms race metaphor so stubborn when – although it certainly has military applications – AI, like many earlier innovations, is so obviously a wide-ranging technology that can drive economic growth? Economists refer to these as ‘general purpose technologies’ (confusingly abbreviated to GPTs) because they have such broad potential application. Past examples include electricity and telecommunications, and AI clearly joins these ranks. And it is technological innovation that drives economic growth over time, as recognised in the 2025 Economics Nobel Prize.[1]
One explanation for the stickiness of the idea that nations are competing with each other in a zero-sum game when it comes to frontier AI is the escalating rivalry, both economic and political, between the United States (US) and China. Another may be that the leading companies, which are American and Chinese, are in something close to a zero-sum game, fighting for market share in a classic example of oligopolistic rivalry. It is tempting, although misleading, to project the corporate battle to the nation state stage.
But for all other so-called ‘middle powers’ (all the OECD and middle-income nations except the US and China), the arms race metaphor is not just misleading; it is actively harmful to their interests. It makes sovereign interest in AI seem a matter of choosing a side in this Great Power contest, when what national economic security actually requires is an understanding of the changed character of the economy in the 21st century, as compared to the pre-digital era.
Navigating global supply chains
Since the 1990s, global production networks have come to dominate manufacturing and, increasingly, services. About two-thirds of global trade in manufactured items consists of components rather than finished goods, with Apple’s products – including the iPhone – a symbolic example of a product designed in California but made in many countries.[2] So while frontier AI models emerge from labs, built from intangible ideas and data, the whole AI stack rests on heavy material infrastructure with multiple interconnections across borders.
National vulnerabilities around access to rare earth materials and advanced chips are well known, but all of the kit needed to build a national digital infrastructure – including the linked energy and communications networks – requires traded components. And there are multiple points of exposure to companies based overseas in data, models and applications. Despite increasingly protectionist rhetoric, the world economy has barely deglobalised at all. Production ecosystems are tightly linked; many types of components are so specialised, and markets so concentrated, that they have few manufacturers.[3]
In this context, it is difficult for governments to make key strategic choices. The Trump Administration has opted for boosting Silicon Valley’s leading companies, although it has repeatedly about-faced on restricting trade in chips. The US is also blatantly using ‘weaponized interdependence’[4], its control of key global infrastructures, to punish those it sees as not falling in line with American interests, including an outrageous visa ban on former European Commissioner Thierry Breton. China has chosen a different game play, emphasising open models and applications. Its prior Belt and Road infrastructure investments give it leverage over many Sub-Saharan African and central Asian economies.
For other governments, though, it is unclear where national economic and strategic advantage will lie in AI, although choices made now will have long-term implications for growth and living standards.
Middle powers must make strategic choices
These stakes are well understood, with policy conversations across Europe and Asia focusing on AI sovereignty – the fuzzy political term of art for not becoming an economic hostage in the uncertain global context. In economic policy circles, the discussion focuses on ‘modern’ industrial policies, with different countries aiming to identify their particular advantages and vulnerabilities.
This needs to be a forensic exercise with a strategic framework. The details of the multiple points of connection between the domestic AI economy and the global will differ depending on the country, and the information needed for effective policies is either sparse, diffused, or both. Most governments will have a better idea of their strongest assets; for example, for the UK they are a strong research base; sectoral expertise in domains like finance, professional services and creative industries; regulatory experience; and rich international networks.
The strategic framework is important for policy consistency. As AI affects the whole economy, it is all too easy for different parts of government or the public sector to take contradictory actions. Yet, to give the UK example again, there is a lack of policy coordination. As our ai@cam colleagues point out in their recent report on UK AI sovereignty:
“Innovation policy invests in compute and research capabilities without connecting them to institutional adoption commitments that would create scaling pathways for successful innovations. Industrial strategy identifies sector and technological strengths but does not link those insights to procurement policies that could provide anchor customers for UK capabilities.”[5]
The specifics will differ between countries. Some nations have already made sophisticated strategic choices. For instance, Vietnam has committed to shaping its AI stack in a distinctively national way while encouraging cross-border flows of talent, data, and computation.[6] Its emerging AI stack is modular, combining a strong commitment to FPT, its flagship domestic firm, domestic research on frontier local language models, and key international partnerships. India has invested significantly in its own Digital Public Infrastructure and is selling the ‘India stack’ in other markets.
There are nevertheless some general principles for ensuring domestic economic security. There are few components of the AI stack where there are fundamentally no options, although de-risking points of vulnerability may require more planning and preparation than policymakers are currently putting into this. Identifying them, with the involvement of detailed industry know-how, is the first step. Supporting open source model development will help reduce dependence on the foreign companies currently leading in frontier AI.
The other principles involve the coordinated and active use of classic industrial policy tools. Technological transformations can’t be left to ‘the market’; or rather, if they are, this is just as much a policy choice as strategic direction by the government. For AI use to benefit the domestic economy, the government needs to be actively involved. Public sector AI procurement is one tool – as Vietnam’s strategy recognises. Procurement decisions by national and local governments and public bodies should consider whether the choice supports domestic AI capabilities and applications, or increases dependencies.
Competition policy enforcement is also essential, as concentrated markets are by definition sources of vulnerability to supply shocks. Given the existing dominance of a small number of companies in digital markets, competition policy is also needed to ensure the possibility of entry by growing domestic companies. The UK is a good example of the dangers of a lack of policy coordination in this respect, with an Industrial Strategy selecting AI as a focus area for the growth of UK companies, yet simultaneously signalling a retreat from vigorous enforcement against the dominant tech companies. The political dilemma is obvious, given the US’s political support for its tech giants, but the outcome is policy incoherence. Strategic choices involve choices.
Delivering on AI’s promise
The focus of this essay has been the directly geopolitical aspects of AI economic security; but there is a broader question of ensuring that the development and use of AI delivers a strong economy. This requires growth, which has been slower globally since the 2008 financial crisis than previously, and faces demographic headwinds in much of the world. It also involves ensuring AI benefits the population broadly, rather than increasing the inequalities of wealth, income, and health that have already stretched the social and political fabric to breaking point.
On both economic growth and its distribution, AI has yet to prove itself. There is public caution about the technology.[7] The ‘job-pocalypse’ has not occurred but is the stuff of everyday headlines and may now be starting to affect young new entrants to the labour market.[8] In this context, hype from the AI industry itself – including fanciful notions of superintelligence next year – is actively counter-productive.
Many governments are considering how to increase the use of AI systems in public services and by the private sector, but not yet with enough focus on the many complementary investments and useful applications that will be needed. One well-known historical study[9] points out that it took decades for electricity to deliver measurable productivity advances – the ‘killer app’ was arguably the washing machine.[10] Digital technologies needed advances in telecommunications, such as 3G networks and compression technologies, to bring us platform business models and the free online services that have transformed how most people work, shop, and play.
This challenge is not just for governments, though. The AI industry needs to prove its value to society through useful, affordable applications that can turn the technology into the driver of economic progress they believe it to be. The ultimate economic security is a prospering middle class with high-quality jobs and excellent public services. So far, this is not a story that can be told about AI – nor will it become one if the arms race metaphor continues to distort government AI strategies.
Read the anthology: Reimagining the Arms Race
[1]‘Prize in Economic Sciences 2025: Popular Information’, NobelPrize.org, 2025
[2]C. Miller and V. Venugopalan, ‘Apple’s Supply Chain: Economic and Geopolitical Implications’, American Enterprise Institute, 2025
[3]D. Coyle, ‘Economic Progress and Adam Smith’s Dilemma’, National Institute Economic Review, Volume 265, Autumn 2023, pp. 5-11
[4]H. Farrell and A. Newman, ‘The Weaponized World Economy: Surviving the New Age of Economic Coercion’, Foreign Affairs, 19 August 2025
[5]J. Montgomery, N. Lawrence, D. Coyle & G. Neff, Navigating AI Sovereignty: Strategic Choices for the UK, ai@cam, University of Cambridge, 2025
[6]D. Nguyen, ‘A Third Path for AI Beyond the U.S.-China Binary’, Noema Magazine, 2025
[7]‘Public Opinion: The 2025 AI Index Report’, Stanford HAI, 2025
[8]R. M. del Rio-Chanona, E. Ernst, R. Merola, D. Samaan, & O. Teutloff, ‘AI and jobs: A review of theory, estimates, and evidence’, arXiv, 2025
[9]‘Why didn’t electricity immediately change manufacturing?’, BBC News, 20 August 2017
[10]H. Rosling, ‘The Magic Washing Machine’, TED, 2011
The views and opinions expressed in this post are those of the author(s) and not necessarily those of the Bennett School of Public Policy.
