The Geopolitics of the AI Arms Race

In 2022, the deployment of ChatGPT and other large language models (LLMs) onto the consumer market has drawn increased attention to advancements in generative artificial intelligence. Public debate has focused on the short- to medium-term economic impacts of this technology, primarily with respect to the anticipated dislocation in the labor market. Some have ventured further, outlining visions of AI as a gateway to a post-scarcity utopia, or warning of the ascent of autonomous, super-human intelligences, seeking to wrestle control over human societies in pursuit of their own aims. Parallel to these public debates, AI has long occupied the minds of policymakers behind closed doors; not just as a catalyst of socio-economic transformation, but as a tool of geostrategy. The leaders of the world’s great powers — primarily China, and the United States, but also Russia, and the EU — increasingly see AI as the deciding factor in a new Cold War. The battle lines of the struggle for global preeminence run not just across the plains of Eastern Ukraine, or the waters of the Pacific. The AI arms race is being decided in laboratories, factories, and in the digital space.


Whether AI development can be viewed as a ‘race’ is in itself contentious. Technological progress in most spheres does not function as a zero-sum game — indeed, the fruits of progress are eventually distributed to the sectors and areas, which are not located on the cutting-edge of innovation. Europe being the first to utilize the steam engine did not prevent other regions of the world from adopting the technology, and reaping its economic benefits, at a later date. One could argue that the space of time between Europe’s industrial revolution, and the industrialization of the remainder of the world, was crucial in enabling the continent’s imperial pre-eminence in the 19th and 20th centuries. However, contemporary dynamics of technological progress, particularly in the digital sphere, reduce the gaps between technological leaders and stragglers. The proliferation of open source software, the widespread availability of communications infrastructure, and the accessibility of research make it easier for states not located on the cutting-edge of AI development to make up for their position. If one company or nation attains victory in the global AI race, it will be short-lived.

Nevertheless, even a narrow lead can translate into a massive, potentially decisive geopolitical advantage. The nature of this advantage differs based on the scenarios of AI development. Projections for the trajectory of AI can be split into two broad categories: slow-takeoff and hard-takeoff. Proponents of slow-takeoff theory predict that improvement in AI will be linear and continuous. The capabilities of AI will steadily increase, but there will be no decisive threshold, which alters the balance of power between technological leaders and stragglers, similar for instance to the acquisition of nuclear arms. Hard-takeoff scenarios envisage this threshold as the creation of AI general intelligence (AGI) — a fully autonomous AI entity, capable of rapid, recursive self-improvement. Once the cycle of self-improvement sets off, AGI could speed-run decades and centuries of technological progress in the space of mere years or months, transforming its makers into undisputed global hegemons.

The prospective emergence of AGI simultaneously raises issues of ‘alignment’ — the term which refers to the harmonizing of the values of AGI with those of its makers. Gaining a decisive lead in the AI ‘race’ would allow the nations located at the cutting-edge of progress to align AI in accordance with their priorities. The hopes of keeping under human control an artificial super-intelligence with cognitive power infinitely larger than that of the whole of humanity combined are flimsy at best. This has led some of the proponents of the hard-takeoff scenario to call for a halt to AI development until the issue of alignment could be solved — not in accordance with the interest of a particular nation or company, but of humanity. However, the potentially decisive implications of AI for the balance of geopolitical power make this vision far-fetched at best. No state is willing to prioritize safety over innovation in circumstances where it could be overtaken by its rivals. Regardless of ideological principles, all participants in the AI race are ultimately guided by the same logic. The best way to tackle the prospect of AGI is to race to win, and hope to either solve the alignment problem along the way, or at least, implement the measures in place at the point of takeoff and hope for the best — AGI which stays faithful to the principles of its makers.

The emergence of AGI currently largely remains the preoccupation of a minority of alarmist AI researchers; however, even in the absence of a hard threshold, temporary preeminence in AI technology could prove decisive. Ultimately, debates on the topic center on categorization. The broad applications of AI mean that it is usually not thought of as a specifically military technology. This has implications for the wider framing of the AI race. The narrow gaps between the leaders and stragglers in the sphere of digital technology mean that for civilian technologies, establishing leadership confers negligible benefits. When viewed through this prism, victory in the AI race appears to be of little significance. For this reason, AI is indeed perhaps best seen as a military technology — not necessarily due to its uses, but due to the fact that a lead in the sector — albeit short-lived — could massively alter the strategic balance of power. Even in a slow-takeoff scenario, AI could enable hard-takeoffs in other sectors — from manufacturing synthetic pathogens, to producing massive volumes of adaptive disinformation –, which could prove decisive in the event of a military confrontation. In a sense, a slow-takeoff scenario amplifies, rather than negates, the race dynamics inherent to AI development, perpetuating continuous, open-ended competition.


The nature of AI in particular, and the digital economy more broadly, is characterized by a combination of two contravening trends: vertical consolidation, and horizontal fragmentation. The same network effects, which allow for rapid transmission of technological innovations across borders, also entrench the power of a few national and corporate hegemons. The population of sub-Saharan Africa has access to products and services provided by Google, Apple, or Huawei, but has little say over what transpires in the halls of power in Silicon Valley, Washington, or Beijing. Despite its globalized nature, the shape of the world’s AI ecosystem is ultimately determined by the relationship between America and China, and by the relative vitality of their economic and political models.

This reality has been laid bare by the ongoing collapse of Sino-American ties. In October 2022, the Biden administration imposed a ban on the exports of advanced chips to China to derail its progress in AI development. The outbreak of the Sino-American ‘chip war’ has confronted other participants in various segments of AI-related supply chains with a choice between the two powers. The South Korean tech giants, Samsung and SK Hynix, presently rely on China for up to 40-50% in multiple key segments of chip manufacturing. In wake of the US embargo, both companies will have to redirect their future capital investments towards other markets. Meanwhile, Japan and the Netherlands — both major players in chip production — are considering imposing export bans of their own, modeled on the American example. All this heralds a reconfiguration of the AI ecosystem around the rivalry between Washington and Beijing. Similarly to rocket technology or space travel during the Cold War, AI reflects the technological aspects of superpower conflict. What separates AI from the aforementioned technologies is that competition in this sphere is not just a symptom of polarization, but one of its main catalysts.

The structure of the AI sector is in itself conducive to mercantilism. Given that AI functions as a general purpose technology, catalyzing development across a range of sectors, regulations cannot merely target a particular category of AI-assisted technology. The chips which power search engines can just as easily be used in drones or missiles. In China’s case, the character of AI as a technology is accentuated by the nature of the political system, defined by extensive interdependence between the Communist Party and Chinese tech firms. The only way for the West to prevent China from weaponizing AI is to impose sweeping, indiscriminate export bans. These then lead to retaliation and devastating diplomatic fallout, exacerbating the mutual rivalry. To a large extent, the AI race lies at the root of worsening Sino-American polarization.

This dynamic is exacerbated by the nature of AI-related supply chains. Due to a combination of economies of scale and network effects, the sector is prone to oligopolization, creating bottlenecks in key segments of the supply chain. Perhaps the most prominent example is the Taiwanese firm TSMC, which is the only one in the world capable of manufacturing the latest generation of AI chips. This supply chain structure incentivizes conflict, as competing powers seek to wrestle control over bottlenecks and cripple their rivals. The physical constraints on the digital economy means that the AI race is as much a race for technological innovation as it is a race for resources necessary to enable this innovation.

Chips are merely the most visible aspect of this battle for resources. Innovation also depends on the supply of data to train AI systems, and the human capital capable of developing them. China holds an advantage in the former sphere. Its population vastly exceeds that of its principal rival, and the CCP’s efforts to advance digitalization and develop new instruments of population monitoring provide Chinese firms with a wealth of data. In contrast, America benefits from its position as a magnet for high quality human capital. More than 60% of the world’s leading AI researchers are employed in US universities and as of 2019, almost 25% of the STEM workers in the country were foreign-born. Its relative openness to immigration, and the economic opportunities offered by its market system, allow the US to offset the Chinese edge in state capacity and population size. This goes to show that resource imbalances reflect a more fundamental dimension of the global AI race — a battle of ideologies and institutions.


At its most basic level, the AI race is a confrontation between Chinese technocratic authoritarianism, and America’s free market liberalism, manifested in two diverging regulatory frameworks. China’s technology sector is organized around multiple hand-picked, state-backed ‘national champions’, such as Alibaba, or Baidu. In contrast, the relationship between the US government and its AI companies is much looser, and the state’s involvement in the sector is limited. The American approach has its benefits — allowing new firms to rise up and challenge industry incumbents, and easing the commercialization of AI technology; however, the US also lacks the long-term vision and commitment characteristic of its trans-Pacific rival. Already in 2014, China established the National Integrated Circuit Industry Investment Fund, with the aim of becoming a global leader in all segments of chip manufacturing. American legislators only moved in a similar direction much later. In August 2022, the Congress passed the CHIPS and Science Act, seeking to promote the growth of the semiconductor industry. Nevertheless, the $53 billion dollars allocated are a mere drop in the bucket compared to China’s investments into AI development over the past decade, which exceed US spending on the sector more than tenfold. In terms of facilitating top-down support for AI research, the Chinese system possesses a clear edge.

Assessing the relative effectiveness of competing regulatory frameworks without a wider context promotes a reductive view of the AI race. All policy — not only AI policy — is not a set of levers, pulled by decision-makers at their leisure, but rather an organic product of ideological and institutional structures. America can never emulate China’s approach to AI regulation, and fixating on its failure to do so detracts from its systemic advantages. Attempts to build a Chinese-style system of public-private partnerships in the West could in fact serve to hinder innovation. In Western societies, regulations tend to be reactive, not proactive — aimed at reducing harm, rather than promoting progress. The EU and its member states embody this tendency, as seen in Italy’s recent push to ban ChatGPT in an attempt to safeguard user privacy. In the West, a regime built to establish leadership in AI could be easily perverted to prevent it, paving the way for Chinese hegemony. 


The path to victory in the AI race lies in building a regulatory framework, which strengthens the systemic advantages of the West. This does not mean abandoning state involvement altogether, but ensuring that it promotes market-driven innovation, rather than clientelist corporatism. The US CHIPS and Science Act functions as a potential model in this regard, with a focus on building the manufacturing capabilities and physical infrastructure to accelerate AI development. The US CHIPS Act, or its European analogue, passed in 2023, should be situated in a collective strategy of onshoring and ‘friend shoring’ to build supply chain autonomy. This would make the US and its alliance partners more resilient both to China’s efforts to seize control of the bottlenecks in chip production, and to the potential fallout of armed conflict in Asia, which would threaten many of the world’s major chip producers — not just Taiwan, but potentially also Japan or South Korea. State support could allow private companies — the West’s champions in the global AI race — to emerge on top in the event of geopolitical dislocation.

Ultimately, chips are merely the physical prerequisite for developing AI; shortcomings in this area are the most visible, and on some level, the most easily solved. The West should not resign on promoting innovation, but it should do so in a way that harnesses the advantages of its market system. This includes tailoring the immigration regime to attract high skill labor from across the world, or consolidating privacy and data protection regulations to eliminate barriers to entry between national markets. Such measures would go a long way towards mitigating China’s advantages in population size and state capacity, creating a sort of collective Western digital sovereignty. This framework would allow Western states to safeguard their citizens’ data, while also establishing a basis for further integration. Governments could also play a role in mitigating the oligopolistic tendencies of the AI sector, and facilitating competition. Despite its shortcomings in the regulatory spree, the EU’s recent effort to legislate the mandatory interoperability of digital products and services constitutes an example of this type of state-led integration, and promises to create a sectoral network effect, easing emergence of new market competitors. When it comes to AI regulation, states and private firms need not be at odds — as in the EU currently –, nor fused — as in the Chinese case. There is space for AI governance, which positions the state as a conduit for market-driven innovation. A similar system could then form the foundations for a Western victory in the AI race.



Alexander, Scott (2022). Why Not Slow AI Progress? Astral Codex Ten, accessed 14. 6. 2023,

Alexander, Scott (2023). Most Technologies Aren’t Races. Astral Codex Ten, accessed 11. 6. 2023,

Fontes, Robin; Kamminga, Jorrit (2023). Ukraine: A Living Lab for AI Warfare. National Defense Magazine, accessed 13. 6. 2023,

Ghitis, Frida (2023). AI Is Already Transforming the Geopolitical Landscape. World Politics Review, accessed 13. 6. 2023,

Hirsh, Michael (2023). How AI Will Revolutionize Warfare. Foreign Policy, accessed 13. 6. 2023,

Judah, Ben (2023). The Impact of the AI Revolution on Geopolitics Is Going To Be Terrifying. Evening Standard, accessed 13. 6. 2023,

King, Anthony (2023). AI At War. War on the Rocks, accessed 12. 6. 2023,

Krushel, Kenneth (2022). Geopolitics in the Era of AI: Upending Assumptions. ESCP, accessed 13. 6. 2023,

Larsen, Benjamin C. (2022). The Geopolitics of AI and the Rise of Digital Sovereignty. Brookings, accessed 11. 6. 2023,

Naughton, John (2023). As AI Weaponry Enters the Arms Race, America is Feeling Very, Very Afraid. The Guardian, accessed 13. 6. 2023,

Sanger, David E. (2023). The Next Fear on A.I.: Hollywood’s Killer Robots Become the Military’s Tools. The New York Times, accessed 12. 6. 2023,

Schmidt, Eric (2023). Why Technology Will Define the Future of Geopolitics. Foreign Affairs, accessed 13. 6. 2023,

Sisson, Melanie W. (2023). Artificial Intelligence, Geopolitics, and the US-China Relationship. Konrad Adenauer Stiftung, accessed 13. 6. 2023,

Toews, Robert (2023). The Geopolitics of AI Chips Will Define the Future of AI. Forbes, accessed 11. 6. 2023,

Image Source:

Xiang, Nina (2021). Biden’s Chip ‘Arms Race’ with China Will Add to Global Uncertainty, Nikkei Asia, accessed 14. 6. 2023,



Written by Matyáš Knol