[Path Selection in the AI Era and a New Korean Peninsula Initiative] ② General Trends Surrounding AI

I. Global Trends and Structural Changes in AI Development

1. The Advent of the AGI Era and the Shift in the Technological Paradigm

In the 21st century, artificial intelligence (AI) has grown into a general-purpose technology poised to reshape the very structure of human civilization. The pace of AI's development has followed a steep upward curve, particularly since the commercialization of deep learning in the mid-2010s. Subsequently, the emergence of the Transformer architecture and the proliferation of Large Language Models (LLMs) have transformed AI from a tool exclusive to technical experts into an everyday tool for the general public. The launch of OpenAI's ChatGPT in November 2022 led this transition, surpassing 100 million users faster than any other application in history.

<Figure 1> Number of models released by Claude in 52 days

Source: JoongAng Ilbo (June 9, 2017)

This explosive development of AI, driven by LLMs, is naturally leading to expectations of a transition to Artificial General Intelligence (AGI), an issue that has risen to the forefront of global policy discourse. With numerous tech gurus, research institutions, and Big Tech companies predicting the advent of AGI within the next decade, major governments have begun to recognize it not merely as a technological innovation but as a strategic asset that could determine national survival. This perception is clearly demonstrated by the United States' announcement of the $500 billion Stargate project the day after President Trump's inauguration in January 2025, and its elevation of AI to a "Manhattan Project" for scientific research through the Genesis Mission in November of the same year. Furthermore, the recently released Claude's Mythos sent shockwaves through global security and defense systems immediately upon its launch, radically shaking up the global AI security landscape.

Another trend to note from a technological paradigm perspective is the rapid rise of open-weight models. The emergence of China's DeepSeek in January 2025 sent a severe shock through the Western AI industry. DeepSeek achieved performance on par with OpenAI while circumventing US export controls on advanced GPUs, and by releasing its model as open source, it fundamentally changed the geographical and economic accessibility of AI technology. As President Trump described it, this event was a "wake-up call for our industry," exposing the vulnerabilities of the US AI strategy to the world (NBC News 2025). China's open-source strategy is effectively proliferating its AI technology in the global model market, and a similar structure is likely to be replicated in the humanoid robotics sector. This trend is accelerating as models from companies other than DeepSeek, such as Alibaba's Qwen, Zhipu, and MiniMax, competitively offer world-class open-weight models for free.

Another important shift is underway in the direction of technological development. Until now, the primary methodology for AI development has been a scaling approach, which involves increasing model size by inputting more data and computing resources. However, as signs emerge that this methodology is approaching the limits of diminishing returns, the industry's focus is gradually shifting toward field performance and efficiency. The "frugal innovation" demonstrated by China's DeepSeek—an approach that achieves more with fewer resources—is the most dramatic example of this shift. The next-generation AI paradigm is expected to move beyond simple scale competition, with the axis of competition shifting toward architectural innovation, energy efficiency, and qualitative leaps in reasoning capabilities. Indeed, deep learning guru Yann LeCun has stated that AGI is more likely to be achieved through multi-modal forms rather than LLMs, and GPT co-founder Ilya Sutskever has also emphasized that the AI race is now moving from an era of scaling to an era of research.

2. The Socioeconomic Ripple Effects of AI

The socioeconomic ripple effects of AI are already becoming visible, but considerable uncertainty remains regarding their scale and direction. In the labor market, 92,000 jobs were reported to have disappeared in the United States in February 2026 alone. The proliferation of AI agents is making the "loop without a human"—where complex tasks are handled without direct human intervention—a reality. According to a study by Cazzaniga et al., while high-income individuals, advanced countries, and men are more exposed to AI, the study also found that they can seize more new opportunities to create greater added value using AI.

<Figure 2> Degree of AI Exposure by Country, Gender, Education Level, and Age

Source: Cazzaniga et al. (2024)

However, viewing the economic impact of AI solely through the lens of job loss is overly simplistic. According to MIT NANDA (2025), although the AI adoption rate within companies is high, only 5% of them have achieved genuine business transformation, and no structural changes were observed in seven out of nine industrial sectors. Some argue that the claim that AI will replace most jobs in the short term is merely an idealistic projection that fails to adequately consider factors constraining actual proliferation, such as the incomplete workflow integration of AI tools, unlearning issues, and organizational cultural resistance.

More importantly, AI dramatically reshapes the distribution of human capabilities. It tends to exponentially amplify the abilities of the top 0.01% of talent while replacing many mid-level tasks. This suggests that organizational structures in the AI era may converge toward ultra-small and hyper-efficient models of "one person + one agent + one humanoid," highlighting the ensuing policy dilemma of deepening inequality versus inclusive growth. Furthermore, the concentration of AI's benefits in English-speaking and high-income countries can deepen the global digital divide, potentially causing severe structural polarization, especially for developing nations in the Global South.

3. Structural Characteristics of the AI Strategic Competition

As the strategic importance of AI technology has grown, the core competitive dynamic in international politics is rapidly converging around strategic competition over AI. In particular, as the technology gap between the two countries repeatedly widens and narrows, attention is increasingly focused on the future development of the competitive landscape.

<Figure 3> Recent Performance Gap Between US and Chinese AI Models

Source: NIST (2026)

To understand the structural characteristics of this competition, it is necessary to examine how the five key strategic resources of AI capability—compute, data, talent, algorithms, and standards—have become core elements of national competitiveness.

First, compute power is the most critical physical infrastructure supporting AI model training and service delivery. NVIDIA's GPUs, TSMC's semiconductor foundries, ASML's EUV equipment, and SK Hynix's HBM have become key bottleneck resources in the AI era, and the competition to control them is forming a new front in geopolitical conflict. Second, data, often compared to the new oil, can be seen as the essential raw material for AI model development. Data localization regulations, cross-border data transfer agreements, and personal data protection regimes determine the legal and institutional foundation of AI competitiveness. Third, talent represents the key players who can create tangible value from the aforementioned infrastructure and resources; it is the crystallization of all knowledge. Because talent with experience and capabilities in frontier research has become more important than ever, major countries worldwide are offering astronomical salaries and benefits to attract top global AI engineers. For instance, Meta reportedly offered some key talents a salary of 100 billion won when launching its new AI lab. Fourth, algorithms, as new systems, have become core assets for the economy, society, and national defense, and are directly used in autonomous systems, cyber operations, and battlefield decision support. Finally, standards function as the new rules of the game and a powerful means of checks and balances. Competing normative frameworks such as ISO/IEC 42001, the NIST AI RMF, the EU AI Act, and China's generative AI regulations are creating a strong competitive structure for leadership in global AI governance.

The competition over these five resources is unfolding primarily around the two axes of the United States and China, but its structure is not a simple bipolar system; rather, it takes the form of a complex, multi-layered competition. The US, with its companies leading global AI innovation, employs a strategy of restricting China's access to technology through export controls and investment screening. China, in turn, is catching up based on its state-led innovation ecosystem and vast domestic market, while seeking to break through US encirclement via an open-source strategy and partnerships with the Global South. Europe has taken on the role of a norm-setter, seeking a third way, while AI middle powers like South Korea, Japan, Singapore, and India are pursuing a hybrid strategy of maximizing their niche capabilities while maintaining strategic autonomy between the US and China.

It is noteworthy that the AI competition has begun to take on the character of a competition over values, moving beyond a simple technological race. At a congressional hearing, OpenAI's Sam Altman directly articulated a confrontation between AI based on US-led democratic values and authoritarian AI. Vice President JD Vance strongly criticized the weaponization of AI by authoritarian regimes at the Paris AI Summit. This discourse elevates the AI competition from a simple economic and technological level to a civilizational and ideological one, further complicating the strategic choices for middle powers. Unlike in the past, these middle powers now face a situation where they have no choice but to secure AI competitiveness to ensure a certain level of strategic autonomy.

II. The Proliferation of AI in Military, Economic, Knowledge, and Governance Domains

1. Comparative Analysis of Major Countries' AI Strategies

(1) The United States: The Evolution and Internal Contradictions of its AI Strategy

The US AI strategy has evolved through three transitional phases since 2017. The first Trump administration (2017–2021), while being the first to define AI as a strategic technology, showed a reliance on private-sector leadership without a concrete national strategy. The 2017 National Security Strategy (NSS) identified AI as a key emerging technology for economic growth and national security and designated China and Russia as strategic competitors, but this remained at the level of a preliminary recognition of risk (Trump 2017). In 2019, Executive Order 13859 (the American AI Initiative) established the first comprehensive AI ecosystem strategy by presenting five priorities: investment in AI R&D, opening up resources, setting governance standards, developing human capital, and international cooperation (Executive Order 13859 2019).

The Biden administration (2021–2025) elevated AI to a standalone, core security priority. This period was characterized by the coexistence of two threat perceptions. One was the external security threat posed by the rapid growth of China's AI capabilities. The other was the threat to democratic values inherent in AI itself, such as the spread of disinformation through deepfakes, the bias of AI systems, and the ethical issues of autonomous weapons. Executive Order 14110 (EO 14110), signed in October 2023, integrated these two dimensions, declaring "safe, secure, and trustworthy AI" as a national policy and mandating safety test reporting for high-performance AI models and the development of content provenance standards (Biden 2023). Additionally, Biden actively utilized multilateral alliance frameworks such as the Quad, the Trade and Technology Council (TTC), the IPEF, and the Fab 4 to build solidarity among allies. The semiconductor export control package announced in October 2022 was considered the most comprehensive in decades and was a comprehensive decoupling measure that included a ban on US citizens and permanent residents working for Chinese semiconductor companies.

The second Trump administration (2025–present) is showing a different direction from the Biden administration. The Stargate project, announced the day after his inauguration, is a four-year, $500 billion AI infrastructure investment plan involving OpenAI, SoftBank, and Oracle, with the goal of building AI data centers across the US in states like Texas, New Mexico, and Ohio, and securing up to 10 GW of computing capacity. The AI Action Plan, announced in July 2025, is composed of three pillars: removing regulatory barriers, building large-scale AI infrastructure, and a foreign and security policy strategy to spread AI as a global standard. The Genesis Mission (EO 14363, November 2025) aims to double the productivity of US scientific research within 10 years using AI, with its core being the establishment of an integrated research platform connecting 17 national laboratories under the leadership of the Department of Energy.

<Figure 4> Status of US DOE's AI for Science Collaboration

Source: Kim Young-kee (2026)

However, the second Trump administration's strategy contains a key internal contradiction. While the AI Action Plan proclaims a competition for AI leadership with China, it reveals a disconnect between declaration and implementation, for example, by deciding to resume sales of advanced NVIDIA chips (H20) to China (The White House 2025). More fundamentally, the security logic that "allies are needed to contain China" clashes with the "America First" doctrine that "allies are also competitors," resulting in AI cooperation with allies taking the form of unilateral dissemination and directives. Trump is pursuing a self-contradictory strategy of demanding cooperation on export controls while imposing high tariffs on allies, which is creating policy confusion and failing to elicit their active participation.

(2) China: The Sprint of State-Led Innovation and Expansion of Global Influence

China's rise in AI is the result of concentrating its developmental state capabilities—which achieved an average economic growth rate of 9.4% for 41 years since the 1978 reform and opening-up—on the AI sector. In 2024, China's research and development (R&D) expenditure reached approximately $1.03 trillion, surpassing the United States ($1.01 trillion) for the first time. This indicates not just a matter of input volume but that China's technological capabilities have reached a structural turning point. China's ranking as number one in the 2025 Nature Index and Tsinghua University's position among the top institutions for AI patent applications worldwide are prime examples supporting this qualitative shift.

China's AI strategy began in earnest with the announcement of the New Generation Artificial Intelligence Development Plan in 2017. This plan set out phased development goals for 2020, 2025, and 2030, adopting a vision to become the world leader in AI by 2030, with the core AI industry reaching 1 trillion yuan and related industries reaching 10 trillion yuan (State Council 2017). Subsequently, the AI Plus strategy, unveiled in 2024, aimed to simultaneously create domestic added value and expand global influence through the comprehensive application and proliferation of AI.

<Figure 5> AI Factories Established in Europe through the AI Connection Action Plan

Source: State Council (2025)

The most prominent feature of China's AI innovation system is the complex symbiotic relationship between the state and the private sector. While it appears state-led on the surface, in reality, over 70% of government procurement goes to private AI companies, and a sophisticated transaction between the public and private sectors occurs between regulation and industrial development. Under the so-called "innovate first, regulate later" principle, the BATH companies (Baidu, Alibaba, Tencent, and Huawei) lead the AI ecosystem, while the state maintains constant competitive pressure through the "catfish effect." The Military-Civil Fusion (MCF) strategy is the security dimension of this structure, institutionalizing the pipeline through which private AI innovation is converted into military capabilities. A clear example of this structure is private AI companies like iFlytek acting as major AI contractors for the People's Liberation Army.

The unique competitive characteristics of Chinese AI are as follows. First, its frugal innovation capability—achieving more with fewer resources—was dramatically demonstrated by the DeepSeek case. Second, its holistic innovation capability through complex multi-models and its orientation toward extensive internalization and self-sufficiency provide structural resilience against US export controls. Third, its vast AI-native consumer market serves as the most powerful testbed for companies to rapidly validate and improve AI systems in real-world environments. The Open AI Innovation Platforms, including 13 AI innovation and development pilot zones in cities like Beijing, Shanghai, and Hangzhou, constitute the geographical and technological infrastructure of this innovation ecosystem.

However, the structural limitations of China's AI development must also be faced. The shortage of creative talent produced by a rigid education system, the institutional suppression of original thinking, and the structural contradiction of an innovation system that lacks freedom and tolerance for failure are major risk factors for China's long-term AI competitiveness. The intensification of US-China competition, a slowdown in economic growth, and the uncertainty of the medium- to long-term economic outlook due to population decline are also external factors constraining the sustainability of China's AI development strategy. Furthermore, an overconfidence in techno-governance—the belief that the Party can effectively manage any technology—also carries the risk of triggering unpredictable catastrophic accidents caused by AI.

Nevertheless, China's short-term strategic consistency is noteworthy. Compared to the unpredictable tariff policies and pressure on allies of the Trump administration, China tends to be perceived as a more predictable partner for cooperation by maintaining a relatively consistent carrot-and-stick strategy. The fact that many middle powers adopt a dual position—adopting US AI technology while also being drawn to China's economic incentives—reflects the effectiveness of this strategic consistency.

(3) Europe: From Normative Leadership to Securing Industrial Competitiveness

Europe's AI strategy has centered on positioning itself as a "regulatory superpower." The EU has consistently maintained the strategic will to establish itself as a third AI power between the United States and China. To secure the same global digital norm leadership in AI governance that it achieved with GDPR, it enacted the EU AI Act in 2024. This law adopts a four-tiered regulatory framework based on the risk level of AI systems. The "unacceptable risk" category, which applies to social scoring systems and unauthorized biometric identification, is prohibited in principle. The "high risk" category applies to areas such as medical devices, education, employment, and critical infrastructure, imposing strict transparency and human oversight requirements (European Union 2024). The law's extraterritorial application provision also applies to all foreign AI companies providing services in the EU market, effectively intending to create a "Brussels effect" that sets the standard for global AI norms.

However, a significant gap exists between the EU's ambitions and reality. After the EU AI Act came into effect, its stringent penalty provisions deterred global Big Tech and other major world companies from investing in and entering the European market. Europe's homegrown AI companies also reached a point where they could not even consider developing new services due to the strong regulations. To overcome this loss of innovation momentum caused by over-regulation, Europe has been actively promoting its leap to become an AI innovation hub by suspending the implementation of the AI Act, creating new exemption clauses, and, in 2025, announcing the AI Continent Action Plan to support the use and innovation of artificial intelligence.

However, according to expert assessments, a significant portion of the EU's announced AI investments is a repackaging of existing programs rather than new funding. A serious investment gap exists, with US software companies spending 10 times more on R&D than their EU counterparts. Looking at individual country strategies, France is pursuing a leap to become an AI powerhouse based on human networks centered on elite educational institutions like École Polytechnique and ENS. Mistral AI, founded in 2023 by researchers who returned to France from Google DeepMind and Meta AI, became Europe's largest AI unicorn, reaching a valuation of $13.8 billion in just 29 months.

<Figure 5> AI Factories Established in Europe through the AI Connection Action Plan

Source: European Commission. (2025). AI Continent Action Plan.

Germany has adopted an "AI+X" model that pursues the AI transformation of a manufacturing powerhouse, fully supporting a strategy to integrate AI into precision manufacturing, automotive, chemical, and logistics sectors. However, the still-low AI adoption rate among German companies and the fact that Aleph Alpha, once the nation's leading AI startup, abandoned frontier model development to pivot to enterprise AI software, reveal the structural limitations of Germany's strategy. Furthermore, with recent moves such as selling off business units, the company is effectively in the process of abandoning its AI business, making future prospects even bleaker.

Since Brexit, the United Kingdom has pursued a pro-innovation regulatory hub strategy, intentionally differentiating itself from the EU's pre-emptive regulation model. It is positioning itself as a hub for global AI safety governance by hosting the Bletchley Park AI Safety Summit and establishing the world's first AI Safety Institute (AISI). Additionally, it is strengthening its strategic connectivity with the US AI ecosystem through DeepMind and key talent, while also simultaneously pursuing its own technological capabilities by creating a "Technology Sovereignty Fund."

(4) Asian Middle Powers: Seeking Strategic Autonomy Between the US and China

Major Asian middle powers are commonly pursuing a two-track strategy in the AI era. They are employing a dual strategy of maintaining connectivity with the United States and the West on normative and security fronts, while preserving a minimum of industrial autonomy and economic ties with China. While all have adopted the discourse of Sovereign AI as the core language of their national strategies, their specific approaches differ considerably.

South Korea accelerated its efforts to formulate an AI strategy following the "AlphaGo shock" of 2016. Since establishing its National AI Strategy in 2019, it has championed the goal of becoming one of the world's top three AI powers (AI G3). The core lever of South Korea's AI strategy lies in its semiconductor industry, particularly High-Bandwidth Memory (HBM). The country supplies approximately 90% of the global HBM market, a strategic asset that has secured its structural indispensability in the NVIDIA-centric AI computing paradigm. SK Hynix, as NVIDIA's largest HBM supplier, along with Samsung Electronics' foundry business, has established itself as a key pillar of the AI infrastructure supply chain. The AI Framework Act, which passed the National Assembly in December 2024 and took effect in January 2026, is Asia's first comprehensive, horizontal AI law. While it imposes transparency and human oversight obligations for high-risk AI, it adopts a more innovation-friendly design than the EU's AI Act. Furthermore, quickly following the United States and China, South Korea has established its own AI action plans and allocated large-scale budgets, fully supporting the development of an independent AI ecosystem (National Artificial Intelligence Strategy Committee 2026).

Japan adopts a strategy of positioning AI as a solution to structural challenges such as population decline, an aging society, and energy dependence, framed under its unique "Society 5.0" concept (Cabinet Office of Japan 2019). In terms of governance, Japan has taken a soft law approach, relying on principles and voluntary compliance rather than binding legislation. As the 2023 G7 president, it played a central role in leading the Hiroshima AI Process and drafting an international code of conduct for developers of advanced AI systems. On the technology front, Japan is pursuing initiatives such as establishing a 2nm logic foundry through Rapidus, developing a Japanese-language LLM by NTT, and exploring Sakana AI's evolutionary model merging methods. Additionally, since the launch of the Takaichi administration, the government has announced large-scale investment plans for key technologies including AI, humanoids, and semiconductors (E-Today, March 8, 2026).

Singapore employs a strategy of turning its physical limitations as a city-state into a paradoxical strength. It focuses on establishing governance leadership as a trusted AI hub rather than on developing domestic foundation models. As a key architect of AI Verify, the world's first AI governance testing framework; SEA-LION, an LLM specialized for Southeast Asian languages; and the ASEAN AI Governance Guidelines, Singapore wields an international influence disproportionate to its size (Smart Nation Singapore 2023). Singapore's strategy can be seen as a unique approach that defines AI sovereignty not by technological self-reliance but by governance capabilities.

India bases its AI strategy on its vast market of 1.4 billion people, the world's third-largest pool of AI talent, and its proven public digital infrastructure, including Aadhaar (a biometric ID system for 1.4 billion people), UPI (processing 18 billion payments per month), and DigiLocker (NITI Aayog 2018). The India AI Mission, approved in March 2024 with a budget of 103.72 billion rupees (approximately $1.25 billion), outlines seven pillars, including sovereign GPU clusters, the development of Indic language LLMs, and the creation of a public dataset platform (MeitY 2024). A unique ecosystem of Indian foundation models is emerging, with names like BharatGPT, Krutrim, and Sarvam AI. At the AI Impact Summit and other multilateral platforms, India has taken on the role of representing the Global South and leading the global AI agenda.

2. The Proliferation of AI in the Military and Security Sectors

The military proliferation of AI is redefining the very paradigm of modern warfare. As AI applications rapidly expand in areas such as Autonomous Weapon Systems (AWS), battlefield situational awareness, cyber operations, logistics optimization, and Intelligence, Surveillance, and Reconnaissance (ISR), military strategists predict that AI will compress decision speed to a level that transcends the limits of human cognition. Technologies like drone swarms, AI-powered cyber weapons, and cognitive warfare using deepfakes are already being deployed in combat or are under development.

As a dual-use technology, AI poses a fundamental challenge to the traditional arms control paradigm. The fact that AI capabilities developed in the civilian sector can be diverted for military purposes makes it extremely difficult to prevent the proliferation of military AI through export controls and restrictions on technology transfer.

In terms of multilateral AI governance, NATO adopted its first AI strategy in 2021 and released an updated version in 2024, establishing an institutional framework for the military application of responsible AI principles (NATO 2024). However, with the United States and China not currently engaged in any meaningful dialogue on military AI cooperation, the formation of international norms for autonomous weapon systems faces a serious vacuum. The fact that the network of AI Safety Summits focuses on civilian AI safety and does not cover military AI governance also contributes to this gap.

3. The Proliferation of AI in the Economy

In the economic sphere, the proliferation of AI is reshaping entire industrial structures and value chains. In manufacturing, AI-powered predictive maintenance, automated quality inspection, and supply chain optimization are dramatically improving productivity, with the convergence of AI and robotics accelerating, particularly in the semiconductor, automotive, and precision machinery sectors. In finance, algorithmic trading, AI-based risk assessment, and anomaly detection have become standard infrastructure, with Singapore's DBS Bank leading the field by operating over 300 AI use cases. In healthcare, innovation is underway in areas like AI-assisted diagnostics, new drug discovery, and clinical trial optimization, and Japan's Society 5.0 framework is gaining attention as a model for AI healthcare in response to an aging society.

The race to invest in AI infrastructure is having wide-ranging effects on global energy, real estate, and financial markets. The 10 GW target capacity of Project Stargate is equivalent to the electricity consumption of a small to medium-sized country, and the surge in AI data centers is creating unprecedented demand for power grid infrastructure, liquid cooling systems, and the entire semiconductor supply chain. Iran's threat to completely destroy the Stargate data center in the UAE, designating it a military target and releasing satellite photos, symbolically demonstrates that AI infrastructure has already become a strategic target in modern conflicts.

AI-linked industrial policy has become a standard policy tool for all major countries. The U.S. CHIPS Act, the EU's Chips Act, South Korea's K-Semiconductor Belt, Japan's semiconductor reconstruction strategy (Rapidus), and the India Semiconductor Mission (ISM)—all centered on semiconductor supply chain security—are examples of hardware-based industrial policies aimed at bolstering AI competitiveness. A common feature of these policies is a protectionist tendency that combines localization of supply chains through state subsidies with export controls on key technologies, which is intensifying tensions in the post-war free trade order.

AI-linked industrial policy has become a standard policy tool for all major countries. The U.S. CHIPS Act, the EU's Chips Act, South Korea's K-Semiconductor Belt, Japan's semiconductor reconstruction strategy (Rapidus), and the India Semiconductor Mission (ISM)—all centered on semiconductor supply chain security—are examples of hardware-based industrial policies aimed at bolstering AI competitiveness. A common feature of these policies is a protectionist tendency that combines localization of supply chains through state subsidies with export controls on key technologies, which is intensifying tensions in the post-war free trade order.

4. The Proliferation of AI in Knowledge and Governance

The international institutionalization of AI governance has advanced rapidly since 2019. The OECD AI Principles (2019) laid the foundation for international norms with five core values: human-centered AI, transparency, accountability, safety, and fairness, which were also adopted as the G20 AI Principles (OECD 2019). The Global Partnership on AI (GPAI), launched in 2020 by the EU and 14 other democracies, is a multilateral AI governance forum from which China, despite Baidu's initial participation, has withdrawn to pursue its own path of AI standardization. The UN adopted a resolution on AI governance in March 2024, and UNESCO issued its Recommendation on the Ethics of Artificial Intelligence in 2021 (UNESCO 2021).

The AI Safety Summit is the platform demonstrating the most dynamic changes in the institutionalization of governance. The first AI Safety Summit, held at Bletchley Park in the UK in 2023, brought together 28 governments and leading AI companies to adopt the Bletchley Declaration and led to the establishment of the world's first government AI Safety Institute (AISI) in the UK. Subsequent summits in Seoul (2024) and Paris (2025) have continued to expand the international agenda on AI safety. An international network of AISIs has now been formed, with participation from the United States, the UK, Japan, South Korea, Singapore, India, the EU AI Office, Canada, and Australia, to promote joint evaluations of frontier AI models and the sharing of red-teaming protocols (AI Safety Institute 2026).

However, underlying AI governance is a fundamental normative competition. The U.S. approach emphasizes innovation-first, voluntary regulation; under a hypothetical second Trump administration, it is projected to roll back the concepts of DEI (Diversity, Equity, and Inclusion) and AI safety itself, while prioritizing the global expansion of American AI as a core national strategy. The EU approach takes precautionary regulation and the protection of fundamental rights as its core principles, seeking to establish mandatory norms for AI companies worldwide through its AI Act. The Chinese approach centers on state-led technological development and cyber sovereignty, building its own AI standards framework in parallel with Western-led governance forums and promoting its export to the Global South. The competition among these three normative frameworks is not merely a contest over technical standards but takes on the character of a civilizational contest for the future of the international order and internet governance in the digital age.

III. Changes in the International Order in the AI Era and Implications for the Korean Peninsula

1. Structural Changes in the International Order in the AI Era

The most significant structural change in the international order surrounding AI is the deepening of tech bloc formation. As U.S. export controls and China's self-reliance strategy intersect, the global technology ecosystem is fragmenting from a single market into multiple, parallel ecosystems. A "multi-homing" structure is becoming apparent, in which each bloc builds its own AI models, semiconductor supply chains, digital infrastructure, and data governance frameworks. While this represents a loss from the perspective of market efficiency, it also expands the space for countries to exercise strategic choices based on their national interests.

<Figure 6> The Decoupling of the U.S. and Chinese Full-Stack AI Ecosystems

Source: Cho Eun-kyo et al. (2021)

The rapid proliferation of the "Sovereign AI" discourse is an ideological expression of this structural change. At its core, technological sovereignty seeks to mitigate unilateral structural dependence on any single country or company. This means it can be achieved by securing access to multiple alternatives, not necessarily through complete indigenous development. From this perspective, Sovereign AI is redefined not as an ideal of complete technological self-reliance, but as a matter of geopolitical capability to actively design the structure of one's dependencies. The United States emphasizes that the desirable strategy for countries is to strengthen their AI sovereignty by leveraging its world-class full-stack AI. In contrast, China argues that securing AI sovereignty is better achieved by utilizing its cost-effective models and avoiding unilateral dependence on the U.S. In this context, it is necessary to comparatively analyze the AI sovereignty strategies pursued by key "AI Middle Powers."

Particularly noteworthy is the diversity of strategic types that middle powers can adopt. "Strategic Indispensability Seekers" like Japan, the UK, and Canada focus on securing an irreplaceable role within the U.S.-centric ecosystem through governance leadership and safety research. "Small-Scale Full-Stack Providers" such as South Korea, France, and Germany pursue a strategy of possessing both their own AI models and semiconductor capabilities. "AI for All Supporters" like India leverage their vast population and digital public infrastructure to pursue a leadership role for the Global South in AI. These types are not fixed and can shift fluidly in response to technological advancements and geopolitical changes.

<Figure 7> Country Groups by Artificial Intelligence Capability

Source: https://www.tortoisemedia.com/data/global-ai#rankings

2. South Korea's Strategic Positioning and Challenges

South Korea's AI strategy aims for a vertically integrated position spanning semiconductors, models, and applications, leveraging its structural strength in HBM semiconductors to connect to foundation models and sector-specific AI proliferation. While the political goal of becoming an "AI G3" nation is ambitious, it is true that many challenges remain. The absolute scarcity of Korean-language data compared to English and Chinese, an excessive reliance on large conglomerates like Samsung and SK Hynix, the brain drain of AI talent to U.S. Big Tech, and the limited global competitiveness of domestic foundation models are all cited as structural constraints.

Recognizing these limitations, South Korea's realistic strategy must be to simultaneously secure "Strategic Indispensability" and leverage it to achieve AI sovereignty. While South Korea is currently pursuing this strategy by leveraging HBM, this dominant position may not be permanent due to changes in AI computing architecture and China's attempts to develop alternatives. Therefore, South Korea must continuously discover new technological and ecosystem foundations to maintain its strategic indispensability in the next-generation technology paradigm.

Pursuing "cooperative technology sovereignty" is also crucial within South Korea's strategic framework. Given that complete technological self-reliance is practically impossible, South Korea must move toward securing collective leverage through strategic networks with other middle powers. This is a reciprocal sovereignty strategy of give-and-take, aiming for a structure that diversifies dependencies across various sectors through multiple partnerships rather than unilateral reliance, while mutually reinforcing cooperative relationships. South Korea's efforts to expand its presence in frontier AI research communities like NeurIPS, co-host the AI Seoul Summit, and bid to host a UN AI Hub can be seen as practical implementations of this direction.

While the strategic implication of approaching the Sovereign AI-centric technology race from the perspective of strengthening absorptive capacity and fundamental competitiveness is valid, what is more important is to seize the opportunity of the next paradigm shift. Just as South Korea preemptively dominated the memory market at the inflection point of the digital transition, a similar strategy is needed in the AI era to capture a "Process Innovation Moment" when the limits of current technology become clear. Securing proactive positioning in next-generation technology areas such as post-transformer architectures, quantum-AI convergence, and energy-efficient AI is key to long-term competitiveness.

3. Implications for the Korean Peninsula

The implications of the AI strategy competition for the Korean Peninsula are complex and multi-layered. First, the intensification of the U.S.-China AI rivalry is likely to structurally entrench South Korea's strategic dilemma. As the United States pursues a strategy to contain the expansion of China's AI influence as much as possible, South Korea faces pressure from both sides: it must participate in the U.S.-led AI supply chain (full-stack) while also maintaining its economic ties with the Chinese market. Just as the market in the Middle East shrank after the U.S. removed Chinese technology and realigned the market around American tech, South Korea's export space for AI could also become progressively limited by geopolitical fragmentation.

Second, the structural implications of North Korea's AI capabilities for inter-Korean relations must be examined. Under international sanctions, North Korea's access to advanced GPUs and AI development tools is extremely limited. However, China's strategy of proliferating open-source AI could partially alleviate this constraint. The spread of Chinese open-source models like DeepSeek provides a pathway to building a certain level of AI capability even without high-performance GPUs. Indeed, there is growing evidence that North Korea is strategically using AI in areas such as cyberattacks, cryptocurrency theft, and propaganda and deception activities. This suggests that in the military and security domain, the AI capability gap between North and South Korea may evolve differently from general technology gaps.

Third, the conditions and prospects for South Korea's potential AI cooperation with the North must be structurally analyzed. Under the current international sanctions regime and the stalemate in inter-Korean relations, AI cooperation with North Korea is extremely limited. From a medium- to long-term perspective, however, AI has the potential to become a new medium for inter-Korean integration. The use of AI in North Korea's agriculture, healthcare, and education sectors could provide a new framework for inter-Korean economic cooperation. In the realm of knowledge and governance, changes in North Korea's information environment driven by AI could become a long-term driver of social change. However, all these possibilities are predicated on a highly sophisticated dual strategy that simultaneously restrains the military application of North Korea's AI capabilities while supporting their civilian use.

Fourth, the most critical aspect of a Korean Peninsula strategy in the AI era is for South Korea to establish its status as an irreplaceable AI partner for the United States, China, and North Korea alike. If South Korea can position itself as a key technology provider in the semiconductor supply chain for the United States, a trusted AI technology partner for various other countries, and, in the long term, a practical partner for AI-based development for North Korea, then AI can become more than just a tool of technological competition; it can be a new structural foundation for peace and prosperity on the Korean Peninsula. To achieve this, South Korea's AI strategy must go beyond strengthening domestic industrial competitiveness and be organically linked with the macro-level goals of unification and peace within the context of Northeast Asian and global AI governance.

Ultimately, securing technological capabilities alone is not enough for South Korea's strategic success in the AI era. Only when the continuous discovery of strategic indispensability, the implementation of cooperative technology sovereignty, the building of middle-power networks, and the establishment of a macro-strategy to position AI as a long-term tool for peace and prosperity on the Korean Peninsula are organically combined can South Korea become a true beneficiary and an active shaper of the AI age. This requires dynamic policy updates and continuous fine-tuning. Recognizing that no single actor can monopolize AI, a reciprocal sovereignty strategy is needed—one that leads AI proliferation through cooperation within international networks. ■

■ Author: Seo-In Baek_Professor, Division of Global Culture and Commerce, Hanyang University.

■ Editor: Jae-Hyun Lim_Research Fellow, EAI Contact: 02-2277-0746 (ext. 209) jhim@eai.or.kr