The US-China Rivalry for AI Supremacy and the Risks of Infrastructure Dependency for Middle Powers: Conditions and Limits of Hedging Strategies

Executive Summary

As the focus of the AI race shifts from model performance to securing infrastructure such as data centers, semiconductors, and power, the technological dependence on the United States by middle powers like the EU, UK, Canada, Japan, and Australia, and developing countries like Brazil and India, is being re-examined as a security vulnerability. Leveraging its overwhelming dominance in the number of data centers (5,427) and cumulative private AI investment ($335.2 billion), the United States is pushing its “Pax Silica” concept, effectively demanding that its allies choose a side. Meanwhile, China is pursuing a dual strategy of building its own ecosystem spearheaded by WAICO and promoting a narrative of cooperation on the G20 stage. The “Compute Coalition” proposed by the Carnegie Endowment for International Peace could offer an alternative for middle powers to achieve economies of scale multilaterally, but it carries the structural risk of ultimately leading to their reabsorption into U.S. technological standards during implementation. Even if the planned U.S.-China AI safety dialogue in September materializes, it is likely to be limited to risk management discussions. The most probable future scenario (50–55% probability) involves middle powers continuing to hedge without fully aligning with either bloc, amidst a selective decoupling limited to security-related items. In light of these trends, South Korea should consider a strategy that combines participation in norm-setting through multilateral channels like the G20 with the diversification of its semiconductor and cloud procurement.

I. Analysis of the Current Situation

Middle Powers Navigating the US-China AI Rivalry: An Analysis of the Current Situation

1. Background and Developments

The center of gravity in the AI race has shifted from competition over model performance to a race for infrastructure. Data centers, semiconductors, and power supply capacity have emerged as key variables determining the gap in AI capabilities between nations [11]. The United States operates 5,427 data centers, more than ten times the number of any other country [11]. From 2013 to 2023, private AI investment in the U.S. reached $335.2 billion, a figure significantly higher than China's [11]. According to an analysis by the Carnegie Endowment for International Peace, this investment gap has led to U.S. dominance in data center infrastructure.

Amid this structural gap, the position of middle powers has become a pressing issue. EU member states, the UK, Canada, Japan, and Australia rely on U.S. Big Tech for most of their semiconductors, cloud services, and advanced AI models. Major developing countries like Brazil and India also have a high degree of external dependency in key sectors. As U.S.-China technological tensions intensify, these nations have begun to seek policies to mitigate their vulnerabilities [12]. The Cypriot newspaper *Cyprus Mail* reported on this within the broader trend of “the global economy being increasingly shaped by geopolitics,” noting that middle powers and developing countries are crafting policies to address vulnerabilities stemming from their dependency in key sectors [12]. Similarly, a geopolitical analysis by EY assesses that the combination of supply chain vulnerabilities exposed by the COVID-19 pandemic and U.S.-China geopolitical tensions has moved industrial policy from the periphery to the center of national economic strategies [8].

2. Current Situation

In a June 2026 report, the Carnegie Endowment for International Peace proposed the concept of an AI infrastructure alliance for the “free world,” dubbed the “Compute Coalition.” The report warns that AI infrastructure will determine the future global balance of power and that time is running out for the democratic bloc [2]. An analysis of the data center industry by Germany's Roland Berger shares a similar concern. It argues that the AI revolution is reshaping global data center infrastructure and that Europe is falling behind in this competition [4]. From Europe's perspective, it faces the dual pressure of reducing its dependence on U.S. cloud and semiconductor supply chains while avoiding absorption into Chinese-style alternatives.

Simultaneously, there are parallel developments at the bilateral level between the U.S. and China. At the G20 Innovation Ministers' Meeting, China's Minister of Science and Technology, Yin Hejun (尹晗俊), called for international cooperation rather than competition over AI. His remarks came ahead of the planned summit between Xi Jinping and Donald Trump [1]. The United States, in contrast, is pushing for the principle of “minimal AI regulation,” the so-called “Carolina Principles,” at the G20. This principle states that new regulations should only be introduced for “unprecedented matters” [14]. The *Nikkei* reported that despite these conflicts, both countries are weighing discussions on AI guardrails for the Trump-Xi summit. Although Silicon Valley figures are proposing ideas for cooperation, skepticism remains prevalent [9]. Singapore's *Business Times* reported that the U.S. and China are preparing for an AI safety dialogue in mid-September. Led on the U.S. side by Secretary of the Treasury Scott Bessent, this dialogue would be the first official bilateral consultation dedicated solely to AI since the start of a potential second Trump term [17].

3. Key Actors and Positions

The United Statesis leveraging its superiority in AI infrastructure to maintain a policy of effectively demanding that its allies choose a side. At the same time, on the G20 stage, it advocates for minimal international regulation to protect the autonomy of its domestic AI industry [14]. This constitutes a dual strategy, combining demands for “minimal regulation” in multilateral forums with calls for infrastructure alignment among its allies.

ChinaIn response to the U.S. offensive on regulatory controls, China is building an alternative narrative centered on “openness and cooperation.” The remarks by Yin Hejun at the G20 and commentary from Chinese experts cited by the *Global Times* are in the same vein. They argue that U.S. tariff strategies have reached their limits and that the AI boom demands global cooperation, not confrontation [18]. This can be read as an external narrative strategy aimed at diluting the U.S. framework of excluding China from technology supply chains.

The EU, UK, Canada, Japan, and AustraliaThese middle powers are seeking an independent path, caught between excessive dependence on U.S. infrastructure and absorption into the Chinese bloc. The Carnegie Endowment's “Compute Coalition” concept is an attempt to consolidate the U.S.-led “free world” bloc by uniting these countries into a single infrastructure alliance [2]. However, whether this concept moves to the implementation stage will depend on each country's financial capacity and its bargaining power with the United States. In Europe's case, as the Roland Berger analysis points out, the data center gap is already structural, so participation in the coalition is more likely to secure conditional access to U.S. infrastructure than to achieve genuine autonomy [4].

Brazil and IndiaDeveloping countries like these have a different calculus than advanced middle powers. India is simultaneously pursuing its own AI application models using its domestic digital infrastructure, such as preparing a UPI-based AI agent payment framework [16]. This can be seen as a pragmatic approach to enhance its own platform competitiveness without fully aligning with either the U.S. or Chinese bloc.

4. Key Issues

The first issue is the feasibility of the “Compute Coalition” concept. If the infrastructure alliance results in a U.S.-led realignment of blocs, participating countries will bear the burden of de facto joining the effort to contain China. EAI's own analysis has previously assessed that, given Europe's skepticism and the incomplete nature of China's WAICO framework, the probability of continued selective decoupling limited to security-related items is highest, at 55–60%, rather than a full hardening of blocs [11].

The second issue is the competition over norms. As the U.S. policy of “minimal regulation” and China's narrative of “cooperation and openness” compete on the international stage, middle powers are in a position where they must decide which normative framework to align with. It is also noteworthy that the G20, a multilateral channel, has become the main arena for this competition [1][14].

The third issue is whether the infrastructure gap will become entrenched. The question is being raised as to what will be the decisive variable in the AI race: nanometers (semiconductor fabrication processes) or gigawatts (power and data center capacity) [5]. If the latter becomes more important, the bargaining power of middle powers and developing countries that lack power infrastructure and capital is likely to weaken further.

II. In-Depth Analysis

Middle Powers Navigating the US-China AI Rivalry: An In-Depth Analysis

1. Analysis of Root Causes

The fundamental reason middle powers face the problem of AI infrastructure dependency lies in economies of scale. The computing resources required for training and inference of AI models operate on a structure where profitability is only achieved after initial investment surpasses a critical threshold. From 2013 to 2023, cumulative private sector AI investment in the U.S. reached $335.2 billion [11]. China's investment was significantly smaller [11]. This gap is directly reflected in the number of data centers. The United States operates 5,427 data centers, a figure that no other country comes close to even one-tenth of [11]. Middle powers such as the EU, UK, Canada, Japan, and Australia cannot compete at this scale on an individual national basis. This is because both their national financial capacity and their ability to mobilize private capital fall far short of that of U.S. Big Tech.

A second cause is layered on top of this: the geographical concentration of semiconductor manufacturing processes. The production of cutting-edge logic chips is effectively dependent on a single company, Taiwan's TSMC. The design of AI accelerators is dominated by an oligopoly of a few U.S. companies, including Nvidia. An analysis by *The Diplomat* points out that the decisive battleground of the AI race is shifting from nanometers (semiconductor fabrication processes) to gigawatts (power supply capacity) [5]. This implies that middle powers are at a structural disadvantage not only in semiconductors but also in the power infrastructure needed to operate data centers. In the cloud services layer, three companies—Amazon, Microsoft, and Google—have effectively divided the global market among themselves. The governments and companies of middle powers have little choice but to entrust their national data to the servers of these three firms.

The third cause is that both the U.S. and China have begun to use this dependency structure as geopolitical leverage. An EAI report analyzes that “the United States, through its Pax Silica concept, is effectively demanding that its allies make a binary choice, while China is moving to build its own bloc spearheaded by WAICO” [11]. While technological dependence was a matter of cost in peacetime, it transforms into a security vulnerability in an era of strategic competition. The fact that middle powers are belatedly seeking infrastructure autonomy is a result of their recognition of this transformation.

2. Structural Context

Economic Structure

As diagnosed by the *Cyprus Mail*, the global economy is being reshaped in a way that is increasingly influenced by geopolitics [12]. An EY analysis of industrial policy traces the origin of this trend to the supply chain shocks of the COVID-19 pandemic. It argues that the pandemic exposed the vulnerabilities of global supply chains, and when combined with U.S.-China geopolitical tensions, this moved industrial policy from the periphery to the center of the policy agenda [8]. This diagnosis applies directly to AI infrastructure as well. Semiconductors, cloud services, and data centers are now being reclassified from purely commercial assets to national strategic assets.

An analysis by the Council on Foreign Relations (CFR) explains this structure within the framework of economic security competition. It posits that countries with greater economic and financial resources also have a greater ability to achieve their national security and geopolitical objectives [15]. A country that possesses AI infrastructure can use it as diplomatic leverage, whereas a country that rents infrastructure is exposed to that leverage. The analysis by Germany's Roland Berger points out that this asymmetry is particularly pronounced in Europe. It warns that Europe is lagging behind the U.S. in the race to build new data centers, and failure to close this gap could lead to a decline in its overall industrial competitiveness in the AI era [4].

Political and Diplomatic Structure

U.S. foreign policy follows a dual strategy of protecting its infrastructure dominance while minimizing intervention in norm-setting. At the G20 Innovation Ministers' Meeting, U.S. technology advisor Michael Kratsios proposed the “Carolina Principles” to various countries, centered on “minimal AI regulation” [14]. This principle states that new regulations should only be introduced for unprecedented matters. This is seen as an attempt to create a favorable environment for the overseas expansion of U.S. Big Tech. At the same time, the U.S. is applying parallel pressure on its allies, effectively demanding they choose a side in terms of infrastructure supply [11].

China employs the opposite rhetoric. Minister of Science and Technology Yin Hejun called for cooperation rather than competition over AI on the G20 stage [1]. Coming ahead of the planned Xi-Trump summit, this can be seen as Chinese diplomatic rhetoric promoting an open cooperation framework to counter the U.S.-led strategy of exclusion. The *Global Times* (環球時報) points out that the widening U.S. trade deficit stems from a surge in computer and semiconductor imports driven by the AI boom, arguing that U.S. attempts to reshore manufacturing are hitting their limits given the AI industry's reliance on global supply chains and cross-border cooperation [18]. The U.S. narrative of containing China and China's narrative of cooperation are thus competing simultaneously on the same stage: the G20.

Security Structure

The issue of AI infrastructure is also merging with traditional security concepts. The U.S. and China are preparing for a dialogue on AI safety in mid-September. It would be the first official bilateral dialogue dedicated solely to AI since the start of a potential second Trump administration, with the U.S. side led by Secretary of the Treasury Scott Bessent [17]. This shows that the AI infrastructure competition has been elevated from the realm of pure industrial policy to an agenda item for national security dialogues. An analysis by the Brookings Institution assesses that the gap between the U.S. and China is actually widening as the AI agenda is successively addressed at the G7 Summit (in Évian), UN-related meetings, and other forums over the course of summer 2026 [6]. This creates a paradoxical structure where dialogue channels to discuss cooperation are increasing, yet the substantive gap is not narrowing.

3. Comparison with Historical Precedents and Similar Cases

The current “Compute Coalition” concept is structurally similar to the Cold War-era Coordinating Committee for Multilateral Export Controls (COCOM), the Western bloc's body for coordinating export controls to the Communist Bloc. However, there is a crucial difference. COCOM was a system for coordinating “what not to give” to protect the West's technological superiority. In contrast, the “Compute Coalition” is a framework for exploring “how to build together” in a situation where an infrastructure gap exists even within the West. The fact that the EU, Japan, Canada, and Australia are not agents of control but rather subjects dependent on U.S. infrastructure means there is an inherent tension different from the internal hierarchy of the Western bloc during the COCOM era.

In terms of semiconductor industrial policy, the U.S. CHIPS and Science Act and the EU's European Chips Act already provide precedents for a similar response. These laws were attempts to expand domestic production capacity in the wake of the semiconductor shortages during the pandemic. However, as EY points out, such industrial policies are not free from criticism regarding increased government intervention and market distortion [8]. The AI infrastructure alliance concept is also likely to repeat the dilemma of the chips acts, as it will be difficult to implement effectively without accompanying government subsidies and tax incentives from each country.

The precedent in the field of energy security is also worth noting. The case of Western countries establishing the International Energy Agency (IEA) to create a system for oil stockpiling and joint response after the 1970s oil shocks shows a typical pattern of multilateral cooperative bodies forming when external dependence on a specific resource turns into a security crisis. This is also the context for the analogy of AI computing resources as the “oil of the 21st century.” However, there is a difference: whereas diversifying alternative oil suppliers was relatively easy, cutting-edge semiconductor manufacturing capacity is tied to a single bottleneck—Taiwan's TSMC—making substitution much more difficult.

4. Key Variables Shaping Future Developments

The first variable is the outcome of the Trump-Xi summit. The *Nikkei* reports that the two countries are weighing discussions on AI guardrails [9]. If even a minimal safety cooperation channel is agreed upon at the summit, the pressure on middle powers to make a definitive choice of bloc could be somewhat alleviated. Conversely, if negotiations break down or export controls are further tightened, middle powers will face pressure to accelerate their efforts toward infrastructure autonomy.

The second variable is Europe's stance. An EAI analysis projects that “given Europe's skepticism and the incomplete nature of the Beijing framework, the probability of continued selective decoupling limited to security-related items is highest, at 55–60%, rather than a full hardening of blocs” [11]. Whether Europe fully joins the U.S.-led “Compute Coalition” or pursues a parallel path of expanding its own data centers could become a benchmark that shapes the options for other middle powers.

The third variable is the course of action taken by developing countries like India and Brazil. India already possesses significant capabilities in building its own digital infrastructure, such as introducing AI agent payment functions to its UPI payment network [16]. This suggests that even among developing countries, there is considerable variation in the capacity for infrastructure autonomy. Countries like India that have their own digital public infrastructure, and those that do not, will likely differ in their ability to pursue a third path independent of both the U.S. and Chinese blocs.

The fourth variable is power supply capacity. If, as *The Diplomat* points out, the decisive battleground of the AI race is shifting from semiconductor fabrication processes to power supply capacity [5], then even middle powers without semiconductor manufacturing capabilities may have an opportunity to gain a relatively advantageous position in the competition for computing resources by expanding their power infrastructure. This could be an opportunity for countries with strong energy infrastructure, including South Korea, and an additional constraint for developing countries with weak power grids.