G20 AI Governance: US-China Rivalry and the Global South’s Dual Hedging Strategy—Implications for Middle Powers

Executive Summary

The G20 Finance and Innovation Ministers' meetings have become a stage for a direct clash between the U.S. approach of minimal regulation, embodied in the "Carolina Principles," and China's proposed path of open cooperation. The United States sought to rally support for strengthening trade controls against China, but a majority of participating nations remained passive, citing their own economic interests and concerns over U.S. credibility. Meanwhile, ahead of a presidential summit, China has adopted a dual strategy: advancing cooperative rhetoric while simultaneously invoking its leverage over rare earth elements. Given the G20's structural limitation as a consultative body lacking mechanisms for adopting binding norms, the most probable outcome is a sustained rivalry, with both the U.S. and China using the multilateral stage for narrative competition and as a tool to bolster their bilateral bargaining power. Within this framework, developing countries of the Global South are likely to counter with a dual hedging strategy, selectively leveraging U.S. infrastructure and computing resources alongside China's low-cost open-source models depending on the situation. For South Korea, already deeply embedded in the U.S.-led technology alliance, a full dual hedging strategy is not feasible. Nevertheless, a sophisticated positioning is required, one that combines the selective adoption of U.S. principles with the establishment of independent standards for utilizing open-source models.

I. Analysis of the Current Situation

US-China Rivalry over G20 AI Governance and the Response of the Global South

1. Background and Developments

The G20 Finance Ministers and Central Bank Governors Meeting in Asheville, North Carolina, followed by the G20 Innovation Ministers' Meeting, became a stage for the United States and China to present diametrically opposed approaches to AI governance [4][1]. U.S. technology advisor Michael Kratsios introduced a set of principles at the meeting, dubbed the "Carolina Principles" [4]. Their core tenet is to "reserve regulation only for new considerations" when establishing AI rules, effectively calling on all nations to adopt a policy of minimal regulation [4].

The meeting held particular significance as it took place just before a summit between Presidents Trump and Xi Jinping, making it more than a routine multilateral conference [1][7]. According to Reuters, the United States attempted to rally support for strengthening trade controls on China at the G20 Finance Ministers' Meeting. However, other participating countries, even besides China, adopted a cautious stance, weighing their own economic interests and concerns about U.S. credibility [12]. The Global Times criticized the U.S. for turning the multilateral platform into a "one-man show" [16].

Around the same time, China's Minister of Science and Technology, Yin Hejun, urged cooperation rather than competition over AI at the G20 Innovation Ministers' Meeting [1]. This move from Beijing, coming amidst escalating technological tensions ahead of the summit, is interpreted as serving the dual purpose of securing leverage in negotiations with the U.S. and appealing to the Global South [1].

2. Current Situation

The U.S. approach of minimal regulation is failing to gain the expected traction at the G20. The Global Times assessed the U.S. proposal as having "stalled," attributing this to member states' fatigue with America's practice of using multilateral discussions merely as a vehicle for its own agenda [16]. While this view from Chinese state media should be taken with caution, it aligns with the Reuters report that a majority of G20 participants did not immediately endorse the U.S.-led framework for containing China [12].

Concurrently, the U.S. Congress passed the "Open-Source AI Leadership Act" [15]. The law encourages the adoption of American-made open-source models and mandates the disclosure of risks associated with using foreign models, effectively targeting those from China [15]. The Global Times characterized this as the U.S. codifying its anxiety over the rise of China's open-source AI ecosystem, arguing that geopolitical containment cannot secure leadership in this domain [15].

The Chinese narrative goes a step further. In an interview with the SCMP, Victor Gao, vice president of the Center for China and Globalization, declared that "China will never tolerate any country having AI hegemony," and asserted that China's control over rare earth elements serves as leverage to prevent the "misuse of AI" by the United States [17]. Although this statement comes from a think tank and has a different tone from official state media, it is a noteworthy development, as it shows Beijing explicitly discussing its resource leverage in parallel with its rhetoric of cooperation.

Amidst this situation, UNCTAD is bringing the development discourse to the forefront. It projects that the AI market will grow to $4.8 trillion by 2033—a scale comparable to the German economy—and highlights practical examples, such as AI assisting rural medical staff with symptom analysis, to emphasize that extending AI's benefits to the Global South is a core development challenge [5]. This can be read as an attempt to establish the agenda of AI accessibility for developing countries as a distinct track, separate from taking sides with either the U.S. or China.

Meanwhile, with the summit approaching, the two countries are also operating a separate dialogue channel. The U.S.-China AI Safety Dialogue, scheduled for mid-September, will reportedly be led on the American side by Treasury Secretary Scott Besent [10]. However, as an EAI report points out, this channel exists within a dual structure in which "the White House's cooperation track and the pressure track from regulatory agencies like the FCC and Congress operate separately" [3]. As the narrative of summit-level cooperation unfolds in parallel with pressure on China from working-level regulators, the rivalry at the G20 is likely to persist, irrespective of the summit's outcome.

3. Key Actors and Positions

The United States, through Michael Kratsios, is attempting to establish minimal regulation as an international norm, but its persuasive power in the G20 multilateral forum is limited [4][16]. Simultaneously, at the congressional level, it is pursuing a de facto containment policy of excluding Chinese models through its open-source legislation [15]. This reveals a dual-track approach on the U.S. side as well, where moderate rhetoric in multilateral settings coexists with hardline legislative action.

China projects a moderate image to the Global South and G20 participating countries through the cooperative rhetoric of Yin Hejun [1]. However, given that this is largely pre-summit negotiating posture, whether Beijing is genuinely willing to yield its ambitions for leadership in norm-setting remains a separate question. As Victor Gao's statement reveals, a hardline stance also coexists, one that explicitly invokes resource leverage, including rare earths [17].

UNCTAD represents a third axis, distancing itself from the U.S.-China competitive dynamic by presenting a development-focused framework [5]. However, UNCTAD's involvement is more about shaping discourse and setting agendas than about creating binding norms.

Developing countries, including Libya, also show notable engagement. Libya's Minister for Digital Economy and AI, participating in the International Dialogue on AI Governance in Geneva, expressed a commitment to help build an AI governance framework consistent with the Sustainable Development Goals [14]. This case illustrates that developing countries are seeking alternative multilateral governance channels, moving beyond a binary choice between the United States and China.

Other G20 member states did not immediately embrace the call to strengthen trade controls on China [12]. This outcome stems from a combination of their own economic interests and skepticism regarding the credibility of U.S. policy, suggesting that middle powers are maneuvering flexibly based on pragmatic interests rather than fully aligning with either the U.S. or China.

4. Key Issues

The first key issue is the competition over regulatory frameworks. While the U.S. principle of minimal regulation and China's rhetoric of open cooperation appear to be in opposition, both sides share the objective of creating a favorable environment for the international expansion of their own companies and models [4][1][15].

The second issue concerns the effectiveness of multilateral forums. As the G20 becomes a proxy battleground for U.S.-China competition, there is a risk that the development agenda demanded by developing countries will be sidelined [5][16].

The third issue is the disconnect between the summit-level and working-level tracks. Separate from the cooperative rhetoric preceding the Trump-Xi meeting, legislative action by the U.S. Congress and measures by regulatory agencies continue to accumulate pressure on China [15][3]. As long as this dual-track structure persists, signals of cooperation in multilateral forums like the G20 should be interpreted with caution and not necessarily as evidence of a substantive improvement in relations.

II. In-Depth Analysis

US-China Rivalry over G20 AI Governance and the Response of the Global South: An In-Depth Analysis

1. Analysis of Root Causes

The root cause of this dynamic lies in the fact that AI governance standards themselves possess market-defining power. The "Carolina Principles," which the U.S. presented at the Asheville meeting, ostensibly advocate for minimal regulation. In reality, however, they represent an attempt to cement the frontier model ecosystem, already dominated by American companies, as the international standard [4]. The principle of confining regulation only to "new considerations" paradoxically benefits existing market leaders, as maintaining a regulatory vacuum allows first-movers to solidify their advantage.

China's rhetoric of cooperation can hardly be seen as pure altruism. The very timing of Minister Yin Hejun's call for cooperation, just before the summit, carries strategic implications [1]. As the United States leans toward an exclusive, bloc-based logic, China aims to preemptively secure an inclusive image. This strategy is designed to expand the market for Chinese models in the Global South—precisely those markets that American firms find difficult to access due to regulatory and security concerns. Victor Gao's declaration that "China will never tolerate any country having AI hegemony" [17] reveals that beneath this cooperative rhetoric lies a coexisting intent to use leverage, particularly its dominance in rare earths, to check its rivals. For Beijing, cooperation and containment are not contradictory; they are two instruments aimed at the same objective.

Another root cause is that for both great powers, the G20 is being instrumentalized as a means to enhance their bilateral bargaining power. According to Reuters, the U.S. attempted to rally support for strengthening trade controls on China at the G20 Finance Ministers' Meeting, but a majority of participants remained passive, citing their own economic interests and concerns about U.S. credibility [12]. This serves as evidence that neither the U.S. nor China is treating the G20 as a genuine forum for shaping multilateral norms.

2. Structural Context

Political Structure: The G20 is a consultative body that lacks mechanisms for reaching unanimity or adopting binding norms. This structural limitation allows both the U.S. and China to unilaterally declare their respective principles. While the Global Times criticizes the U.S. for turning the G20 into a "one-man show" [16], the same critique could be applied to China's rhetoric of cooperation. As a non-binding forum, it creates greater incentives for both countries to invest resources in narrative competition rather than in substantive compromise. As previous EAI analysis has noted, the U.S.-China relationship is characterized by a dual-track structure in which "the White House's cooperation track and the pressure track from regulatory agencies... and Congress operate separately" [3]. This duality is replicated on the G20 stage, exemplified by the simultaneous occurrence of Yin Hejun's cooperative remarks and the passage of the Open-Source AI Leadership Act by the U.S. Congress [15].

Economic Structure: UNCTAD's projection that the AI market will grow to $4.8 trillion by 2033 [5] signifies that for countries in the Global South, this competition is not an abstract debate over norms but a tangible issue of development opportunities. However, the U.S. is still assessed as having an overwhelming lead in infrastructure and investment [3]. Amid this asymmetry, developing nations are incentivized to preserve their policy autonomy and pursue practical gains rather than becoming subordinate to either side's standards. Libya's participation in the Geneva AI governance dialogue, where it expressed its "commitment to join efforts to build effective AI governance consistent with the Sustainable Development Goals" [14], illustrates this trend. It is a strategy of securing a stake through participation in multilateral channels rather than by joining a specific bloc.

Security Structure: The competition over AI hardware and models is expanding into a complex security issue entangled with semiconductor export controls, rare earth supply chains, and control over data infrastructure. Victor Gao's mention of rare earths [17] suggests that China frames the AI competition not merely as an issue of advanced technology, but as a comprehensive contest for leverage that encompasses mineral supply chains. This is directly linked to pressures for supply chain realignment in the domain of trade and economic security.

3. Historical Precedents and Comparative Cases

The current situation is structurally similar to the technology standards competition of the Cold War, particularly the U.S.-Japan semiconductor friction in the 1980s and the rivalry over telecommunication standards (GSM vs. CDMA). Then, as now, the standard-setting process itself was used as a geopolitical bargaining tool, based on the understanding that preempting technology standards would directly lead to market dominance. However, the nature of the standards in the current case is different. While semiconductor and telecom standards were technical specifications for interoperability, AI governance principles are distinct in that they represent an ideological contest over regulatory philosophy—specifically, the degree to which the state should intervene in the market.

Furthermore, the competitive dynamic between open-source and closed ecosystems can be compared to the past rivalry between Linux and Windows, or to the tension between U.S. leadership in internet standards and China's attempts to build its own independent intranet. The dual strategy of the U.S. Congress's Open-Source AI Leadership Act [15]—which encourages the proliferation of domestic open-source models while simultaneously publicizing the risks of foreign ones—is analogous to the historical pattern of the U.S. championing free trade rhetoric while concurrently implementing measures to protect its own industries.

The approach of the Global South is less a revival of the Cold War's Non-Aligned Movement and more akin to the "issue-based coalition" strategy employed by developing countries in the multilateral trading system since the 2000s. Rather than fully aligning with a particular bloc, this strategy involves pursuing practical interests by participating in multiple multilateral channels—such as UNCTAD and various G20-affiliated bodies—on an issue-by-issue basis. This differs from the Non-Aligned Movement under the U.S.-Soviet bipolar system, which attempted to form an ideological third axis. Today's developing countries are focused more on securing a voice at the norm-setting table than on establishing an independent ideological platform.

4. Key Variables Shaping Future Developments

The first variable is the outcome of the Trump-Xi Jinping summit. The degree to which the AI safety dialogue channel is institutionalized at the summit will determine whether the rivalry at the G20 eases or continues via the pressure track involving regulatory agencies and Congress [3][10]. A previous EAI analysis assessed that "the most likely path is the permanent establishment of a dual track, where the dialogue channel is limited to confidence-building information exchanges" [3]. If this scenario materializes, the competition at the G20 will likely also take a form that combines superficial de-escalation with sustained underlying rivalry.

The second variable is a potential shift in the attitudes of G20 member states, particularly middle powers and developing countries. The fact that participants prioritized their own economic interests and adopted a cautious stance toward the U.S. call for stronger trade controls on China [12] could constrain the reach of future U.S.-led initiatives. If this trend solidifies, the United States might shift the center of gravity of its strategy toward minilateral forums or bilateral relationships.

The third variable is the actual trajectory of the open-source ecosystem's proliferation. The key question is the extent to which the U.S. Congress's Open-Source AI Leadership Act [15] can genuinely facilitate the adoption of American-made models in the Global South. Legislation alone cannot guarantee actual market adoption rates. If Chinese open-source models, leveraging their superior cost-performance ratio, gain traction in developing country markets, U.S. regulatory pressure could prove counterproductive.

The fourth variable is whether control over critical mineral supply chains, including rare earths, is actually exercised. The landscape of the AI hardware competition will shift depending on whether Victor Gao's remarks [17] remain mere rhetoric or translate into the actual use of export controls. This issue is directly linked to ongoing discussions about supply chain realignment in the domain of trade and economic security, and it would have direct repercussions for countries dependent on semiconductor and AI infrastructure, including South Korea.

The fifth variable is the extent to which international organizations like UNCTAD can translate the development framework into tangible resource allocation. Whether the projection of the AI market growing to $4.8 trillion by 2033 [5] remains mere political rhetoric for developing countries or leads to actual infrastructure investment and technology transfer will determine the substantive bargaining power of the Global South.