Risks to the Korean Semiconductor and AI Supply Chain from Expanded U.S. AI Export Controls and Strategic Response Directions

Executive Summary

Executive Summary

The White House's post-hoc mandatory blocking of foreign access to Anthropic's frontier AI models in June 2025 signals a structural shift in the U.S. export control system, expanding comprehensively from hardware to AI software models. The fact that the access blocking order was issued just five days after SK Telecom and NVIDIA announced their gigawatt-scale AI cloud partnership symbolically demonstrates that U.S. export control decisions are primarily driven by self-centered national security and industrial interests rather than alliance cooperation logic, starkly exposing the structural vulnerability of the strategy of Korean companies relying on the U.S. AI ecosystem. However, at the same time, the bottleneck in the AI supply chain is expanding beyond GPUs to include core hardware value chains such as memory, substrates, and optical components, which Korea effectively dominates, thus structurally strengthening Korea's bargaining leverage. Accordingly, this report recommends a two-track strategy for Korean semiconductor companies: strategically deepening interdependence with the U.S. to secure institutional access guarantees while simultaneously building independent AI capabilities and alternative ecosystems. Since excessive dependence on a single partner and complete separation from the U.S. ecosystem are both realistically untenable options, Korean companies must enhance their strategic indispensability within the U.S. AI ecosystem while consolidating their structural foundation for negotiation power through strengthening their own capabilities.

Phase 1: Issue Situation Analysis

Risks to the Korean Semiconductor and AI Supply Chain from Expanded U.S. AI Export Controls

Issue Situation Analysis

1. Background and Progress of the Issue

As the U.S.-China tech hegemony competition expands beyond semiconductors into the realm of AI software models, the U.S. export control system is entering a new phase. The U.S. government has expanded the scope of export regulations from hardware (semiconductors and equipment) to include control over access to frontier AI models to prevent security-concerned countries like China and Russia from misusing advanced AI systems for military and intelligence purposes [7]. The Trump administration is pursuing this trend more aggressively, moving towards institutionalizing government intervention in the release and distribution of cutting-edge AI models by major AI companies such as Anthropic and OpenAI [9].

A decisive event occurred within this context. On June 7, 2025, SK Telecom and NVIDIA announced a partnership to build gigawatt-scale AI cloud infrastructure in Korea, seemingly positioning it as a core pillar of Korea's national AI strategy [1]. However, just five days later, on June 12, the White House ordered Anthropic to immediately cease all foreign access to its cutting-edge AI models, Mythos 5 and Fable 5 [1][7]. This marked the first instance of the government forcibly recalling publicly released frontier AI models, starkly revealing the uncertainty of U.S. AI technology access to partners worldwide, including allied nations [13].

2. Current Situation (Latest Trends)

The Anthropic incident has been resolved, not through complete revocation, but through a form of limited restoration. On June 26, the U.S. government partially allowed access to the Mythos 5 model for over 100 'trusted' domestic U.S. institutions, including Fortune 500 companies [8]. However, this measure is strictly limited to U.S. domestic institutions, and companies in allied nations, including Korea, remain blocked from access. During this process, India received a verbal assurance from the U.S. side that 'access to technology once provided will not be blocked' [12], but it is understood that no equivalent official guarantee has been made to Korea.

Meanwhile, on June 25, the U.S. hosted the second 'Pax Silica' summit in Washington, involving approximately 20 countries, to strengthen alliance cohesion for AI supply chain security [2]. At this meeting, U.S. Under Secretary of State for Economic Affairs, Jake Helberg, praised Korea as a country that 'far exceeds its weight class' in semiconductor production and technology, expressing expectations for Seoul's role in the U.S.-led AI supply chain security initiative [2]. Concurrently, China is pursuing a plan to invest $295 billion over the next five years in AI data center networks, mandating that at least 80% of the supported technology be procured from domestic suppliers [4], further structurally narrowing market access for Korean companies to China.

Domestically, the Korean government is planning to announce new semiconductor cluster development plans to address the production capacity shortage driven by the AI boom, and a presidential advisory committee member emphasized that Samsung Electronics and SK Hynix should accelerate their cluster plans by a decade. As the AI infrastructure bottleneck expands beyond GPUs to encompass the entire supply chain, including memory, substrates, and optical components, the structure is forming where U.S. capital investment is inevitably concentrated in Korea, Taiwan, and Japan, which effectively dominate these core hardware value chains [6].

3. Key Actors and Their Positions/Interests

The U.S. Trump Administrationdefines AI technology as a national security asset and is shifting its policy to allow access to frontier models only conditionally, even for allies. While Under Secretary of State Helberg praises Korea as a key partner [2], he maintains a cautious stance on establishing a formal framework that guarantees the continuity of technology access, as revealed in negotiations with India [5]. The U.S.'s core interest is to prevent advanced AI technology from leaking to adversarial nations like China and Russia, while simultaneously incorporating allied nations into a U.S.-led technological order.

Anthropicrevealed itself to be an actor subordinate to government directives rather than exercising independent corporate judgment, by blocking and then partially restoring foreign access following a government order [8][9]. This suggests limitations for U.S. AI companies in pursuing independent global business strategies and demonstrates that Korean companies relying on U.S. AI platforms are exposed to similar risks at any time [3].

The Korean Government and Companiesare in a structurally vulnerable position, deeply integrated into the U.S.-led AI supply chain system yet unable to secure stable technology access. SK Telecom is pursuing a large-scale AI cloud partnership with NVIDIA while simultaneously experiencing the shock of Anthropic's model access being blocked [1]. Samsung Electronics and SK Hynix maintain global leadership in AI core hardware such as HBM and are recognized as strategic partners by the U.S. [2][15]. However, they face a persistent dilemma between the reality of being unable to completely escape dependence on the Chinese market and the intensifying U.S. export controls on China.

Chinais accelerating the establishment of an independent technological ecosystem by mandating domestic procurement of AI infrastructure and making large-scale state investments in response to the U.S. technology blockade [4]. This acts as a factor that structurally reduces the opportunities for Korean companies to seek business in the Chinese market.

European and Other Allied Nationsare also expressing concern over the unilateral expansion of U.S. export controls. The Netherlands has opposed the MATCH Act targeting ASML and is engaged in negotiations with Washington [11], while Australia, prompted by the Anthropic access block, is emphasizing the need to build its own AI capabilities [10]. This indicates that U.S. export controls are triggering discussions on technological self-reliance across allied nations.

4. Summary of Key Issues

The first key issue is the collapse of predictability in technology access. The Anthropic incident empirically demonstrated that access to U.S. AI technology can be blocked overnight, irrespective of allied actions, based on internal circumstances of Silicon Valley companies or U.S. government security assessments [3]. This is a factor that fundamentally destabilizes the planning horizon for Korean companies that have built their business strategies on U.S. AI platforms.

The second issue is the contradiction between alliance partnerships and technology control. While the U.S. publicly praises Korea as a key partner in AI supply chain security [2], it does not provide formal, binding guarantees regarding access to frontier AI models. Compared to India, which obtained a verbal assurance [12], Korea's negotiation position appears relatively weak.

The third issue is the intensification of restrictions on access to the Chinese market. With the strengthening of U.S. export controls and China's mandate for domestic procurement progressing simultaneously, the space for Korean semiconductor and AI companies to operate in the Chinese market is being squeezed from both directions [4]. In particular, the lack of clear definition regarding what level of business relationship with China constitutes a 'suspicious criterion' for the U.S. side [4] forces Korean companies to make business decisions in an uncertain regulatory environment.

The fourth issue is the competition for technological leadership to resolve AI bottlenecks. Korea's global leadership in core AI infrastructure hardware, including HBM, is a strategic asset [6][15], but its dependence in software and models acts as a risk factor. The trend of technological bottlenecks expanding beyond GPUs to encompass memory, substrates, and optical components presents a new opportunity for Korean companies, but also serves as a warning that the scope of U.S. supply chain control may expand [6].

Phase 2: In-depth Issue Analysis

Risks to the Korean Semiconductor and AI Supply Chain from Expanded U.S. AI Export Controls

In-depth Issue Analysis

1. Analysis of the Root Causes of the Issue

The most fundamental cause of this incident lies in the U.S. redefinition of AI as a core strategic asset for national security, rather than merely a civilian technology. The Trump administration views the potential misuse of frontier AI models by military and intelligence agencies of security-concerned countries like China and Russia as an existential threat, thus undertaking a strategic shift to comprehensively expand the existing hardware-centric export control system to the software and model domains [7][9]. This signifies the emergence of a new paradigm where export controls on semiconductor equipment and advanced chips evolve into access controls on AI models themselves.

A second fundamental cause is the structural conflict between the dual-use nature of AI technology and the existing export control framework. Semiconductors or manufacturing equipment have physical substance, allowing for relatively clear demarcation of export controls. However, AI models exist in digital form and can be replicated and distributed instantly across borders, making it difficult to ensure the effectiveness of controls with existing export licensing systems. The unprecedented measure taken by the U.S. government to forcibly block access to already publicly released frontier models [13] can be considered a policy of last resort stemming from this technological specificity.

Third, the competitive dynamics among U.S. AI companies are intricately intertwined with government regulation. Questions are being raised as to whether the foreign access blocking of Anthropic stemmed solely from security assessments or if competitive interests among Silicon Valley companies played a role in the regulatory process [3]. The fact that export controls can be influenced not only by security concerns but also by inter-company competitive dynamics fundamentally undermines the predictability of technology access for allied companies. The reality that corporate competitive maneuvers can nullify the technology access rights of alliance partners overnight exposes the inherent vulnerability of the strategy of Korean companies relying on the U.S. AI ecosystem [3].

2. Structural Context

Political Structure

At the political level, this incident reflects the structural tension between the U.S.'s strategy to maintain technological hegemony and its approach to alliance management. On one hand, the U.S. praises Korea as a key semiconductor partner that 'far exceeds its weight class' and draws it into the center of the Pax Silica initiative [2]. On the other hand, it exhibits contradictory behavior by unilaterally blocking AI technology access for allied nations without prior consultation. This suggests that U.S. export control policy is primarily driven by self-centered national security and industrial interests rather than the logic of alliance cooperation.

Particularly noteworthy is the comparison with India. The U.S. provided India with a verbal assurance that 'access to technology once provided will not be blocked' [12], but no equivalent official guarantee has been made to Korea. This demonstrates the harsh reality that U.S. technology diplomacy operates based on differentiated negotiation power and strategic priorities by country, and alliance status itself does not automatically guarantee the stability of technology access. If the Korean government fails to secure more explicit and binding technology access guarantees within the Pax Silica framework, it risks remaining in an asymmetric structure where it pays the costs of alliance participation while the benefits are uncertain.

Economic Structure

Economically, Korea holds a unique position as a key node in the AI hardware supply chain. As AI evolves into the agentic stage, bottlenecks are expanding beyond GPUs to encompass the entire supply chain, including memory, substrates, and optical components, and these core hardware value chains are effectively dominated by Korea, Taiwan, and Japan [6]. Samsung Electronics and SK Hynix have secured a leading position in High Bandwidth Memory (HBM), and projections indicate that the combined operating profit of these two companies next year will reach an all-time high [15]. Thus, Korea possesses bargaining leverage as an indispensable partner supplying the physical foundation of AI infrastructure.

However, this strength is also the source of structural vulnerability. Korean companies have built their business strategies based on close integration with the U.S. AI ecosystem (NVIDIA GPUs, Anthropic/OpenAI models, etc.). The incident where Anthropic's model access was blocked just five days after the announcement of SK Telecom and NVIDIA's gigawatt-scale AI cloud partnership [1] starkly reveals the vulnerability of this integration strategy. Conversely, the Chinese market, while investing $295 billion in AI data centers over the next five years, mandates that over 80% of supported technology be domestically sourced [4], structurally shrinking the access for Korean companies to the Chinese market. Ultimately, Korean semiconductor and AI companies face the dilemma of increasing their dependence on the U.S. market while failing to secure guarantees for the stability of that access.

Security Structure

From a security perspective, the current situation demonstrates that technological control is emerging as a key instrument of new geopolitical power. The United States is building a technological alliance system centered on AI supply chain security through the Pax Silica initiative, and South Korea is incorporated as a core member of this system[2]. However, the internal logic of this system is designed to prioritize U.S. control over the autonomy of its allies. If export controls can be applied to allied companies based on security concerns and unverified suspicions, the criteria for allied access to U.S. frontier AI technologies will remain effectively unclear[4].

This security structure imposes a dual pressure on South Korea. By participating in the U.S. export control system for China, South Korea faces restrictions on access to the Chinese market, while simultaneously bearing the uncertainty that access to U.S. AI technology could be blocked at any time. China is responding by strengthening the exclusivity of its own AI ecosystem[4], making it practically impossible for South Korea to pivot towards the Chinese AI ecosystem[3]. Thus, South Korea is facing structural pressure with narrowing options at the intersection of U.S.-China technological bloc formation.

3. Comparison with Historical Precedents and Similar Cases

COCOM and the Wassenaar Arrangement

The current strengthening of AI export controls can be understood as a modern iteration of COCOM (Coordinating Committee for Multilateral Export Controls), established during the Cold War to control the export of strategic materials to the communist bloc. The COCOM regime was a mechanism for Western allies to collectively control the export of dual-use technologies to the Soviet Union and communist countries, and it transitioned to the Wassenaar Arrangement in 1996 after the end of the Cold War. The current Pax Silica initiative[2], as a new form of technological alliance control system centered on AI and semiconductor technologies, can be seen as a modern version of COCOM. However, a significant difference lies in the fact that while COCOM operated based on consensus among member states, current U.S. AI export controls tend to be unilaterally decided without prior consultation with allies.

U.S. Pressure on Japan's Semiconductor Industry (1980s)

The U.S.-Japan semiconductor friction in the 1980s provides a structurally similar precedent to South Korea's current situation. At that time, the U.S. viewed the rapid growth of Japan's semiconductor industry as a threat to its national security and industrial interests, and through the 1986 U.S.-Japan Semiconductor Agreement, it imposed restrictions on the market behavior of Japanese companies. This case illustrates that the U.S. can exert unilateral pressure on its allies if it perceives its technological and industrial hegemony to be threatened. The fact that South Korean semiconductor companies currently possess unparalleled competitiveness in core AI infrastructure technologies such as HBM[15] makes it impossible to rule out the possibility of a technological competition dynamic with the U.S. forming in the long term.

Huawei Sanctions and Supply Chain Realignment

The U.S. sanctions on Huawei in 2019 serve as the most direct precedent demonstrating how AI and semiconductor export controls can rapidly reshape global supply chains. The Huawei sanctions went beyond targeting a single company, effectively reorganizing the entire global telecommunications equipment and semiconductor supply chain based on dependence on U.S. technology. At the time, Samsung Electronics and SK Hynix were forced to halt semiconductor supplies to Huawei, leading to significant revenue losses in the short term. Current export controls on AI models possess the potential to rapidly alter the business environment for South Korean companies in a similar manner.

ASML and the Netherlands' Dilemma

The case of ASML in the Netherlands offers another important precedent for South Korea to consider. ASML, the world's sole manufacturer of extreme ultraviolet (EUV) lithography equipment, has been restricted from exporting advanced equipment to China under U.S. pressure and recently faces further restrictions due to the MATCH Act[11]. The unprecedented diplomatic action by the Dutch Minister of Trade to personally visit Washington and express concerns to Congress and the Commerce Department[11] demonstrates that technological export controls have emerged as a core diplomatic issue between allies, beyond mere trade matters. It is suggested that South Korea, similarly to ASML, needs to leverage its position as an indispensable node in the AI supply chain to pursue more proactive diplomatic responses.

4. Key Variables in Issue Development

Variable 1: U.S. Differentiated Approach to Allies

The most crucial variable in the future development of this issue will be how the U.S. differentiates among its allies regarding access to AI technology. Currently, India has secured verbal assurances[12], Australia has formalized the need to build its own AI capabilities[10], and Europe is seeking a collective response[13]. Whether South Korea can negotiate explicit and binding guarantees for technological access within the Pax Silica framework will be a key determinant of the stability of South Korea's AI ecosystem. If the U.S. applies differentiated access criteria for each ally, South Korea's negotiating power and diplomatic positioning will directly define the scope and stability of its technological access.

Variable 2: Level of Institutionalization of AI Export Controls

The extent to which the U.S. government institutionalizes export controls on AI models in a systematic and predictable manner is also a significant variable. Currently, there is uncertainty where access to technology for allied companies can be blocked overnight due to competitive maneuvering or unverified security concerns[3][4]. If the U.S. establishes an institutional framework with clear criteria and procedures, the business planning possibilities for South Korean companies will increase. Conversely, if arbitrary and retroactive control methods persist, South Korean companies will have no choice but to revise their strategies to reduce dependence on U.S. AI technology[9].

Variable 3: Speed of South Korea's Acquisition of Independent AI and Semiconductor Technological Capabilities

The speed at which South Korea can reduce its dependence on the U.S. AI ecosystem and build its own technological capabilities is also a key variable. The government's plan to create new semiconductor clusters[3] and the competitive development of AI infrastructure technology by science and engineering universities like KAIST and UNIST[15] demonstrate movement in this direction. Just as HBM originated from an idea in a KAIST research lab[15], if South Korea secures original independent technologies in next-generation AI bottleneck-solving technologies, it can buffer the impact of blocked access to U.S. AI models while strengthening its negotiating leverage. However, as this is a long-term challenge that is difficult to achieve in the short term, a dual approach is required, combining short-term risk management with long-term technological self-reliance strategies.

Variable 4: Speed of China's Advancement in Technological Self-Reliance

The speed at which China achieves self-reliance in its AI ecosystem through AI data center investments ($295 billion) and domestic procurement mandates (80%) is also an important variable for South Korea[4]. As China's technological self-reliance accelerates, South Korean companies' access to the Chinese market will become increasingly limited, potentially leading to a vicious cycle of deepening dependence on the U.S. market. Conversely, if China cannot resolve its short-term dependence on certain South Korean components or materials during its technological self-reliance process, this could serve as a temporary negotiating leverage for South Korean companies. The speed of China's technological self-reliance is an exogenous variable that will determine the effectiveness of market diversification strategies for South Korean semiconductor and AI companies, requiring continuous monitoring.