Deepening U.S.-China AI Rivalry: The Computing Infrastructure Gap, Alliance Realignment Pressure, and South Korea's Dual-Track Approach

Executive Summary

The focus of the AI competition is shifting from model performance to securing infrastructure, including data centers, semiconductors, and critical minerals. Through its Pax Silica initiative, the United States is effectively demanding that its allies choose a side, while China is attempting to build its own bloc spearheaded by the World Artificial Intelligence Cooperation Organization (WAICO). However, considering Europe’s skepticism and the incomplete nature of Beijing’s framework, the most likely scenario (55–60% probability) is a continuation of selective decoupling limited to security-related items, rather than a full hardening of blocs. For South Korea, a dual-track response is the most realistic option: strengthening proactive risk management in strategic technology areas such as advanced semiconductors and AI models, while maintaining its existing business approach toward China in general commercial sectors. Given the structural U.S. advantage in infrastructure and the limited appeal of China’s alternative framework, adhering to a principle of sector-specific responses, rather than prematurely committing to one side, better serves South Korea’s national interest.

I. Analysis of the Current Situation

Deepening U.S.-China AI Hegemonic Competition: Computing Infrastructure and Alliance Realignment

1. Background and Developments

The center of gravity in the AI competition is shifting from model performance to securing infrastructure. Full-stack AI capabilities are overwhelmingly concentrated in the United States and China. Last year, the two countries produced 80 notable AI models[1]. The U.S. operates 5,427 data centers, more than ten times the number of any other country[1].

Underlying this gap is a difference in the scale of private investment. From 2013 to 2023, private AI investment in the U.S. reached $335.2 billion, while China recorded a significantly smaller amount[10]. Previous EAI analysis has concluded that this investment gap has led to the disparity in data center infrastructure[10].

A turning point came early last year with the release of DeepSeek-R1 by a Chinese startup. Touting low cost and high efficiency, the model gained attention as an alternative to the U.S.-led narrative of AI development[6]. It is assessed that the Chinese government gained confidence after DeepSeek's debut[6]. The subsequent V4 Pro model adopted an open-weight approach. Nikkei Asia reported that this has sparked an unprecedentedly intense debate in Washington and Silicon Valley over how to respond to China's AI technology[13].

In a July 7 interview with Punchbowl News, President Trump stated, "I don't want to see China win in artificial intelligence"[5]. He also confirmed his intention to raise the AI issue directly at the U.S.-China summit scheduled for late next month[5]. The fact that four of the top ten AI models are Chinese—and that they were developed at one-tenth the cost—is fueling a sense of crisis in Washington[5].

2. Current Situation

The United States launched its Pax Silica initiative in 2025, aiming to restructure the supply chains for AI models, semiconductors, and critical minerals under its leadership[9][15]. According to an internal document obtained by Reuters, the U.S. State Department is preparing to demand that dozens of countries choose a side, warning that any nation signing on to Beijing's competing framework will be excluded from the U.S.-led AI coalition[9][11]. Tanzania's The Citizen reported this as a "notification with no choice"[15]. Kazakhstan is cited as a prime example of a country caught between the U.S. and Chinese frameworks[11].

Beijing's response has been a direct rebuttal. The Global Times claimed that Washington's strategy of dividing the world into blocs was "dead on arrival"[18]. As its own alternative, China established the World Artificial Intelligence Cooperation Organization (WAICO), creating an independent framework that excludes the Western actors who have traditionally led technology policy[16]. The Astana Times interpreted this as Beijing's countermove to Trump's exclusion of Europe from U.S. AI models[16]. Europe, however, remains skeptical of this proposal[16].

Within China, the theory of an independent technological path is gaining traction. Zheng Yongnian, a scholar at The Chinese University of Hong Kong, Shenzhen, cited Huawei's Atlas 950 SuperPod computing system as an example, arguing that China has the ability to adapt to external pressure[17]. The South China Morning Post noted that this argument is linked to the need for China to build its capacity for setting international standards[17].

Meanwhile, a process of selective decoupling is underway. Following Beijing's hardline response to Meta's attempt to acquire the Chinese AI startup Manus, Tencent acquired a stake in Meta, becoming its largest shareholder[12]. Nikkei Asia analyzed this incident as a turning point that led Chinese authorities to strengthen controls on the cross-border flow of data and AI technology[12]. Nokia's withdrawal from its China operations and the early phase-out of Windows by Chinese government agencies are part of the same trend[12].

3. Key Actors and Positions

United StatesThe U.S. defines its superiority in AI infrastructure as a geopolitical asset. The Carnegie Endowment for International Peace has urged the formation of a "free world computing coalition," warning that "time is short for the democratic bloc"[4]. The Pax Silica initiative is the institutional tool for building this coalition[9]. However, concerns have been raised within the U.S. that forcing countries to choose sides could backfire.

ChinaChina is bypassing U.S. containment efforts with low-cost, high-efficiency models and an open-weight strategy. WAICO is the diplomatic extension of this bypass strategy[16]. At the same time, it is pursuing a dual strategy domestically: strengthening state control over data and AI technology while managing security and openness separately[12].

Southeast Asian and Gulf CountriesThese nations have emerged as a new variable in the computing infrastructure competition. The Diplomat assessed that this region has the potential to exert its own influence rather than being absorbed into the U.S.-China bipolar structure[1]. Some countries, like Kazakhstan, are already being forced to choose between the two sides[11].

EuropeEurope is in an awkward position, being excluded from U.S. AI models while also remaining skeptical of China's WAICO proposal[16]. Roland Berger has diagnosed that Europe's data center infrastructure gap is widening[3].

South KoreaSouth Korea is adjusting its diplomatic course in line with its industrial networks and the landscape of the computing economy. A report by the Carnegie Endowment for International Peace assessed that "President Lee Jae-myung's travel schedule shows that Seoul is reorienting its diplomacy away from a Pyongyang-centric focus and toward the geography of industrial networks and the computing economy"[4]. EAI has previously recommended that the South Korean government and companies adopt a dual-track approach, combining proactive risk management in the domain of national strategic technologies with the maintenance of the status quo in general commercial areas, rather than making a binary choice between complete withdrawal and business as usual[12].

4. Key Issues

First is the effectiveness of the pressure to choose sides. While the U.S. Pax Silica initiative aims to isolate China from resources, it remains unclear whether middle powers will actually comply, as suggested by the Global Times' rebuttal[18][9].

Second is the trade-off between the infrastructure gap and model efficiency. The U.S. maintains a dominant advantage in data center scale, but China is attempting to offset this gap with models developed at one-tenth the cost[5][1]. The key question is which side this dynamic will favor in the long run.

Third is the room for hedging by middle powers. Countries like Kazakhstan, as well as those in Southeast Asia and the Gulf, are hesitant to be drawn into a framework of binary choice[11][1]. South Korea is also pursuing a sector-specific response, separating national strategic technologies from general commercial areas[12]. A key point to watch going forward is how long this hedging space can be maintained under pressure from both the U.S. and China.

II. In-Depth Analysis

Deepening U.S.-China AI Hegemonic Competition: Computing Infrastructure and Alliance Realignment

In-Depth Issue Analysis

1. Root Cause: Why the Infrastructure Competition Now?

There are structural reasons for the shift in the AI competition's axis from algorithms to infrastructure. The "DeepSeek shock" changed perceptions in U.S. policy circles, leading to the judgment that model performance alone cannot maintain the U.S. lead. DeepSeek-R1 achieved performance close to U.S. models at one-tenth the cost[5], and the V4 Pro was released with open weights, accelerating its proliferation[13]. Algorithmic innovations are quickly imitated and disseminated. In contrast, data centers, power grids, and semiconductor supply chains require accumulated capital and time to build. The U.S. possession of 5,427 data centers is the result of decades of accumulated private investment[1][10]. Washington has shifted the focus of competition to computing infrastructure based on the calculation that this is an area where China cannot catch up in the short term.

The assessment by the Carnegie Endowment for International Peace reflects this perception precisely. Based on the premise that "AI infrastructure will determine the global balance of power," it asserted that "there is a narrow window of time for the democratic bloc to get ahead"[4]. This phrasing also conveys a sense of crisis—that the U.S. advantage is not permanent. President Trump's direct statement that he "does not want to see China win in AI" and his preview of making it a summit agenda item are in the same context[5]. Ultimately, there is one root cause: the judgment that while catch-up is possible in model competition, preemption in infrastructure competition leads to a permanent gap. This has led the United States to play the card of alliance realignment.

2. Structural Context

Political Structure.The U.S. Pax Silica initiative is designed to require allies to choose a side. An internal State Department document obtained by Reuters specifies that countries signing on to Beijing's competing framework will be excluded from the U.S.-led coalition[9][11]. This is similar to Cold War-era bloc logic but with a crucial difference. Cold War alliances were primarily based on military security treaties. This time, economic incentives like supply chains and standards take precedence. The fact that Tanzania's The Citizen perceived this as a "notification with no choice"[15] and that Kazakhstan is in a difficult position between the U.S. and Chinese frameworks[11] shows that this pressure is creating a real dilemma for middle powers and developing countries.

China's response, WAICO, is a direct counter to this dichotomy. Through its independent framework that excludes the West, it puts forward a message of "inviting everyone"[16]. The Astana Times interpreted this as Beijing's countermove to Trump's exclusion of Europe from U.S. AI models[16]. However, Europe's reaction is skeptical[16]. The gap between the Global Times' claim that "Washington's bloc strategy is already dead"[18] and Europe's actual wait-and-see approach[16] suggests that China's alternative framework has yet to gain substantial traction.

Economic Structure.The gap in data center infrastructure between countries is extreme. The U.S. is the overwhelming leader, and a Roland Berger report points out that Europe is falling behind in this competition[3]. The report highlighted both the risk of a correction in the data center boom and the need for a European response[3]. The gap is also widening in the industrial AI sector. According to a survey by the Oliver Wyman Forum, a few companies in the transportation, logistics, and defense industries are far ahead in AI adoption[7]. This means the AI competition is creating a winner-take-all structure not only at the national level but also at the corporate level.

Security Structure.Previous EAI analysis has pointed out that U.S.-China technology decoupling is proceeding selectively, not comprehensively. The assessment is that it is "most likely to proceed not as a complete severance but as a selective separation limited to areas directly related to national security, such as AI, semiconductors, and telecommunications equipment"[12]. This is supported by Beijing's hardline response to Meta's attempt to acquire the Chinese AI startup Manus and the subsequent trend of the Chinese government strengthening controls on the cross-border flow of data and AI technology[12]. It is highly likely that a dual structure will persist for the time being, with separation accelerating in the security domain while interdependence is maintained in the commercial sphere.

3. Comparison with Historical Precedents

The current situation has similarities to the technological bloc formation of the Cold War era, but there are crucial differences. The Cold War's Coordinating Committee for Multilateral Export Controls (COCOM) regime was centered on controlling military technology exports and was a competition between countries with clear bloc affiliations. The current U.S.-China AI competition is far more fluid, as it unfolds on a structure of economic incentives involving supply chains, standards, and platforms. The very fact that non-aligned countries like Kazakhstan and Tanzania are being explicitly pressured[11][15] shows that this competition cannot be simply reduced to Cold War-style bloc logic.

A comparison with the phase of semiconductor export controls is also valid. The trends pointed out by EAI, such as Nokia's withdrawal from China and the accelerated phase-out of Windows in government agencies[12], are in line with the supply chain separation patterns seen during the 2019–2020 sanctions against Huawei. However, the current situation represents a step further, as China is now responding by creating its own alternative multilateral organization, WAICO[16]. This shows a difference from the Huawei sanctions period, when China's response was largely defensive; this time, it is proactively proposing its own order.

4. Key Variables Shaping Future Developments

The first variable is the U.S.-China summit scheduled for late next month. As President Trump has announced his intention to raise the AI agenda directly[5], the outcome of this meeting—whether it results in an agreement or a breakdown—will determine the direction of the next phase.

The second variable is the actual choices made by middle powers. The key question is which side countries will actually align with in response to U.S. pressure to choose, or how sustainable attempts to maintain relations with both sides, like Kazakhstan's, will be[11]. In South Korea's case, as EAI has already pointed out, the diagnosis is that a "dual-track response is necessary, combining proactive risk management in the domain of national strategic technologies with maintaining the existing approach and strengthening monitoring in general commercial areas"[12]. Whether this approach is actually sustainable depends on the intensity of U.S. pressure.

The third variable is a potential shift in Europe's stance. Currently, Europe is skeptical of China's WAICO while also being partially excluded from U.S. AI models[16]. Whether Europe strengthens its independent path or eventually converges with the U.S. bloc will serve as a reference case for the choices of other middle powers.

The fourth variable is the tangible outcome of infrastructure investment. If the data center boom enters a correction phase[3], the entire U.S. strategy based on computing superiority could be reconsidered. Conversely, if China's independent computing systems, like Huawei's Atlas, deliver substantial results[17], the persuasive power of Washington's infrastructure gap strategy could be weakened.