Intensifying Semiconductor and AI Competition Among South Korea, China, and Japan and the Asian AI Boom: The Sustainability of Bottleneck Positions and Strategic Implications for South Korea

Executive Summary

The fact that South Korea and Taiwan surpassed Japan in export value for the first time ever in the first half of 2026 is a result of the shift in AI computing architecture from a race for process miniaturization to one for bandwidth and packaging. This has concentrated value-added in South Korea and Taiwan, which control High Bandwidth Memory (HBM) and advanced foundries. At the same time, the rapid growth of China's CXMT and Cambricon suggests that the bottleneck position secured by South Korea and Taiwan may not last indefinitely. The United States is also increasing pressure on South Korean companies to invest in the U.S. to secure a domestic memory production base. This situation presents a dilemma for South Korean firms' resource allocation, as it combines risks from dependency on critical items in trade and economic security, the tech hegemony contest within the U.S.-China strategic competition, and the inter-state rivalry in AI. Over the next 12-18 months, a gradual erosion of this bottleneck position is expected, depending on three variables: whether South Korea and Taiwan can maintain their technological supremacy, the pace of China's internalization, and the intensity of U.S. pressure. South Korea will require a dual-track strategy of persuading the U.S. while simultaneously investing in domestic production bases in Yongin and Honam. From the perspective of a middle power's AI strategy, South Korea must leverage its hardware advantages while keeping pace with technology standards to avoid falling behind the Taiwan-U.S. alliance in the race to capture next-generation bottlenecks like optical modules and silicon photonics.

I. Issue Analysis

Intensifying Semiconductor and AI Competition Among South Korea, China, and Japan, and the Sustained Asian AI Boom: An Issue Analysis

1. Background and Developments

The division of labor in East Asia's semiconductor industry is being reshaped in the age of AI. According to data compiled by Nikkei, South Korea's exports in the first half of 2026 reached $496.3 billion, while Taiwan's reached $416.6 billion [3][7]. Japan's exports amounted to only $384.4 billion [3][7]. This marks the first time in history that South Korea and Taiwan have surpassed Japan in export value [3]. The gap was even more pronounced in year-on-year growth rates. Japan's growth remained around 10% [3][7], whereas South Korea and Taiwan both approached 50% [3][7].

The root of this reversal lies in the long-standing division of labor. Japan has traditionally held strengths in materials like fluorinated polyimides and photoresists rather than finished chips [3][7]. The expansion of investment in AI data centers has allocated greater value-added to South Korea and Taiwan, which control HBM and advanced foundries [3]. However, Japan has not been entirely excluded from the AI boom. The profit margins of materials and components companies like Ibiden and Rohm have also improved [3]. The head of Tokyo Electron's Taiwan branch noted that AI is driving demand across the semiconductor spectrum, including advanced packaging, DRAM, HBM, and NAND flash [11].

2. Current Situation

Following the surge in SK hynix's stock price, there are assessments that AI investment in the South Korean and Japanese stock markets has entered an overheating phase [1]. Consequently, a trend has been observed of investors turning to Chinese equity derivatives as a means of diversification. Trading desks at Barclays and UBS have confirmed an increase in client demand for call options and swap contracts linked to the CSI indices in recent weeks [1]. UBS has suggested the CSI500 as an alternative for Asian AI investment [1]. A fund manager at PIMCO also stated that due to valuation concerns with U.S. Big Tech, they are shifting their portfolio toward Asian equipment stocks and Chinese financial and healthcare stocks [15].

In the memory semiconductor market, Samsung Electronics reclaimed the top spot in the DRAM market in the second quarter of 2026 [8]. At the same time, it was reported that China's ChangXin Memory Technologies (CXMT) is rapidly increasing its market share [8]. China's semiconductor exports surged by 99.5% during the same period, and Cambricon's revenue grew by 108% [7]. This is interpreted as a signal that the current bottleneck position held by South Korea and Taiwan may not last indefinitely [7]. The Chinese state-run media outlet Global Times criticized the U.S. tariff strategy for failing to solve the trade deficit issue, while also noting that Samsung Electronics and SK hynix are emerging as key suppliers meeting the demand for AI infrastructure expansion in the United States [16].

The United States is making an all-out effort to secure a domestic production base for memory semiconductors. This stems from the judgment that its own production capacity is insufficient in the memory sector, which has become a bottleneck for AI innovation [13]. Within South Korea, concerns are being raised that U.S. pressure to invest in America could stifle domestic investment and distort companies' strategic resource allocation. Experts point out that, in parallel with efforts to persuade the U.S., the government must accelerate support for domestic production bases in areas like Yongin and Honam [13].

The supply chain competition is expanding beyond memory into the realms of optics and packaging. In the first half of 2026, Nvidia invested over $6 billion in its optical interconnect and silicon photonics supply chain in just three months [18]. Marvell acquired Celestial AI to secure its photonic fabric technology [18]. Taiwanese media have reported that both TSMC and Nvidia consider silicon photonics a key battleground in their strategic rivalry [18]. Chinese packaging firm SJ Semi is also in pursuit, expanding its 2.5D packaging platform for HBM integration [19].

3. Key Actors and Stances

South Korea has secured an export advantage by controlling two bottlenecks: HBM and advanced foundries. SK hynix is constructing an advanced packaging plant in Indiana, USA, with a target start date of late 2028 [5]. However, U.S. pressure to invest in America acts as a constraint on its capacity for domestic investment [13]. From the perspective of South Korea's Ministry of Trade, Industry and Energy and Ministry of Foreign Affairs, the country faces the dual challenge of persuading the U.S. while simultaneously promoting support for domestic production bases.

Japan has fallen behind in the competition for finished chips but maintains its position in the materials and equipment sectors. The Nikkei's direct reporting of this export reversal with the phrase "for the first time in history" can be interpreted as reflecting a sense of crisis within Tokyo's policy circles [3]. Kioxia has announced a strategy to strengthen its relationships with U.S. Big Tech and counter Chinese competitors by promoting high-performance NAND flash as an alternative to DRAM [10]. Tokyo Electron is adjusting its local strategy in Taiwan to meet the growing demand for packaging and memory driven by AI [11].

Taiwan led by TSMC, is maintaining its advanced foundry status and seeking to preserve its lead by adopting ASML's latest equipment. TSMC plans to use ASML's newest machines starting in 2030, while Samsung Electronics aims for 2028, and Intel has already begun mass production [12]. The Belgian chemical company Solvay announced plans to more than double its production capacity for high-purity semiconductor chemicals in Taiwan by 2026, a case that illustrates Taiwan's emergence as a key hub in the AI supply chain [17].

China is aiming to erode the current bottleneck positions by accelerating its semiconductor self-sufficiency. The 99.5% increase in China's semiconductor exports and the 108% rise in Cambricon's revenue support this trend [7]. CXMT is rapidly expanding its share in the DRAM market [8]. At an East Asia Institute forum, Professor Seok-June Kwun pointed out that "whereas a few years ago China was primarily focused on producing low-value-added semiconductor chips, recently we are beginning to see phases of a qualitative shift," and that "for AI semiconductors, the internalization of supply is now practically complete within China" [9]. The Chinese state-run media outlet Global Times is criticizing U.S. tariff policies while advancing the argument that the AI boom highlights the need for global cooperation [16].

United States is focused on establishing a domestic production base to secure the memory semiconductors needed for its AI infrastructure expansion [13]. Concurrently, it seeks to check China's semiconductor self-sufficiency through technology controls, but there are assessments that the pace of China's progress is eroding the effectiveness of these controls [7][9].

Global Investment Banks and Asset Managers (UBS, Barclays, PIMCO) are rebalancing their portfolios toward Chinese stocks and Asian equipment stocks, citing valuation concerns for AI-related stocks in South Korea and Japan [1][15]. This indicates that market judgments on the sustainability of the Asian AI boom are divided.

4. Key Issues

The first issue is the sustainability of the bottleneck position held by South Korea and Taiwan. Opinions are divided on whether their technological supremacy in HBM and advanced foundries is a structural advantage or a temporary phase. The surge in China's semiconductor exports and the rise of CXMT and Cambricon suggest the possibility that this bottleneck position will gradually erode [7][8].

The second issue is the direction in which the bottleneck is shifting. As new potential bottlenecks emerge—such as optical modules, silicon photonics, and next-generation packaging—companies like Nvidia and Marvell are concentrating massive investments in these areas [18]. If South Korea becomes complacent with its current dominance in HBM and foundries, it risks being preempted at the next bottleneck point.

The third issue is the U.S. pressure for domestic production and the strategic resource allocation of South Korean companies. The key question is whether concerns that expanding investment in the U.S. could erode domestic investment capacity will translate into actual policy responses [13].

The fourth issue is the potential shift in investment flows. A short-term market concern is whether the fears of overheating in South Korean and Japanese AI-related stocks will lead to a capital flight to Chinese stocks or merely result in a temporary correction [1][15].

II. In-Depth Analysis

Intensifying Semiconductor and AI Competition Among South Korea, China, and Japan, and the Sustained Asian AI Boom: An In-Depth Analysis

1. Analysis of Root Causes

The fundamental cause of the current semiconductor and AI competition among South Korea, China, and Japan lies in the changing nature of AI computing demand. In the past, the competition in semiconductor process miniaturization was a game of increasing transistor density. Miniaturization from 22nm and 14nm to 7nm and 3nm was the primary path to performance improvement. In the AI era, this formula has changed. The bottleneck is no longer the processor's computing speed but how quickly data can be moved to the processing unit [14]. This is why Apple, Xiaomi, and Nvidia are all pursuing terabyte-level bandwidth [14]. This structural shift has concentrated value-added in companies that control HBM and advanced packaging.

This change has shaken the long-standing hierarchy of the East Asian division of labor. Japan has maintained its superiority in materials like fluorinated polyimides and photoresists rather than finished chips [3][7]. This position was stable in the era of process miniaturization competition, as material supply was difficult to replace and gave Japan greater bargaining power over finished product manufacturers. However, the value-added created by the AI boom has flowed more toward HBM and foundries than materials. This is the background to South Korea and Taiwan surpassing Japan in export value for the first time [3][7]. The fact that Nikkei reported this with the phrase "for the first time in history" itself reveals the sense of crisis within Tokyo's policy circles [3].

The overheating on the investment side stems from a separate cause. While the surge in SK hynix's stock price partly reflects improved performance, it is also a result of a structural reallocation of capital flowing into Asia as skepticism about U.S. Big Tech valuations grows [15]. The move by a PIMCO fund manager to reduce exposure to the Magnificent Seven and shift to Asian equipment stocks and Chinese financial and healthcare stocks is a prime example [15]. This reflects a market judgment that the distribution of profits within the AI value chain is shifting from U.S. platform companies to Asian hardware suppliers.

2. Structural Context

The Economic Security Structure. The current supply chain reorganization is directly linked to the risk of dependency on critical items, a central axis of trade and economic security issues. The U.S., judging its domestic production capacity for memory semiconductors needed for AI infrastructure expansion to be insufficient, is increasing pressure on South Korean companies to invest in America [13]. This is not merely an industrial policy but an economic security strategy aimed at absorbing supply chain leadership within its own territory. Within South Korea, there are concerns that this pressure could stifle domestic investment and distort corporate resource allocation [13]. The call to pursue persuasion of the U.S. in parallel with supporting domestic bases starkly illustrates this structural tension [13].

The U.S.-China Strategic Competition Structure. The fact that China's semiconductor exports surged by 99.5% and Cambricon's revenue grew by 108% [7] shows that China is accelerating its semiconductor self-sufficiency in response to U.S. technology controls. The fact that ChangXin Memory Technologies (CXMT) is expanding its market share at the same time Samsung Electronics is reclaiming its lead in the DRAM market [8] means that the competition over bottleneck positions is already underway. The Chinese state-run Global Times criticized the limitations of U.S. tariff strategy while assessing that Samsung Electronics and SK hynix have emerged as key suppliers meeting U.S. AI infrastructure demand [16]. This suggests that Beijing is interpreting the strategic position of South Korean semiconductor companies within the framework of U.S. efforts to contain China. From China's perspective, the bottleneck position of South Korea and Taiwan is an obstacle to be overcome, not a permanent hierarchy to be accepted.

The Structure of Inter-State AI Competition. Discussions at the U.S.-China AI summit, as summarized by Brookings, suggest that the gap between the two sides is not narrowing [2]. An analysis of U.S.-China AI competition by the Council on Foreign Relations also points out that the emergence of new models is continuously reshaping the competitive landscape [6]. Within this inter-state competition, South Korea has presented "Sovereign AI" as a national agenda item. However, scholars in South Korea have pointed out that it remains unclear what specific industrial breakthroughs this strategy will yield in the international political context [9]. Professor Seok-June Kwun noted that "the existing geopolitical logic... has begun to shift its center of gravity toward semiconductors and, more recently, AI, giving rise to what can be called techno-geopolitical logic" [9]. He further diagnosed that "as you well know, the biggest threat to our country's semiconductor industry right now is China" [9].

Physical Constraints on New Technology Development. The expansion of competition into the optics and packaging domains is directly linked to the current state of new technology development. The fact that Nvidia has invested over $6 billion in its optical interconnect and silicon photonics supply chain [18], and that Marvell acquired Celestial AI to secure photonic fabric technology [18], shows that the data movement bottleneck has emerged as a new front in the semiconductor competition. The fact that TSMC, Intel, and Samsung Electronics have all announced plans to adopt ASML's latest equipment [12] is also a response to this structural shift.

3. Historical Precedents and Comparative Cases

The current reversal can be contrasted with the rise of the Japanese semiconductor industry in the 1980s and 1990s. At that time, Japan seized semiconductor hegemony by eroding the U.S. DRAM market, only to be later surpassed by South Korea and Taiwan. The ongoing reversal by South Korea and Taiwan against Japan is a repetition in the opposite direction, yet its cause differs from the past, as this time value-added is concentrated in finished products and packaging, not materials. The structure where Japan maintains its strength in materials while losing in the finished product competition is similar to the trajectory of the Japanese electronics industry since the 1990s, which specialized in parts and materials while ceding leadership in the finished goods market.

China's attempt at semiconductor self-sufficiency, as a response to U.S. export controls since 2018, can be compared to the Soviet Union's response to COCOM (Coordinating Committee for Multilateral Export Controls) restrictions. However, while the Soviet Union ultimately failed to close the technology gap with the West, China is narrowing the gap at a considerable speed, as seen in Cambricon's 108% revenue growth and CXMT's market share expansion [7][8]. This shows that the effectiveness of controls varies with the era and the maturity of the technology. The AI semiconductor field contains a mix of bottlenecks monopolized by a few companies, like EUV lithography equipment, and those with relatively higher potential for substitution, like HBM, making the effects of controls uneven.

The phenomenon of investor concentration also overlaps with past cases. The flow of capital from overheated U.S. tech stocks to emerging Asian markets during the dot-com bubble of the early 2000s and the current shift from the Magnificent Seven to Asian AI-related stocks [15] show a similar cyclical rotation pattern. The difference this time, however, is that the destination is not emerging markets in general but is concentrated in a specific value chain: the AI hardware supply chain.

4. Key Variables Shaping Developments

The first variable is the level of U.S. pressure on South Korea for investment. If the U.S. continues to press South Korean companies to invest in order to secure a domestic memory production base [13], the tension between South Korea's domestic production foundation and its investment allocation to the U.S. is likely to become a major policy issue. The speed of support for domestic bases like Yongin and Honam will be key to responding to this pressure [13].

The second variable is the pace at which China overcomes its bottlenecks. If CXMT's DRAM market share expansion and Cambricon's revenue growth continue [7][8], the lifespan of the current HBM and foundry bottleneck position held by South Korea and Taiwan could be shortened. This implies the possibility that the front line of competition will shift to preempting the next bottleneck points, such as optical transceivers, next-generation packaging, and power semiconductors.

The third variable is whether the concentration in capital markets will ease. If the overheating of investment in South Korean and Japanese AI-related stocks enters a correction phase and the flow of funds to Chinese stocks becomes more pronounced [1][15], this could lead to a market reassessment of the Asian AI value chain, going beyond a simple portfolio rebalancing. It remains to be seen whether the increased demand for CSI index-linked products at Barclays and UBS [1] will act as a leading indicator of this trend.

The fourth variable is the winner of the competition in optics and packaging technology. The outcome of the rivalry among Nvidia, TSMC, and Marvell over silicon photonics and next-generation packaging technologies [18], along with the expansion of 2.5D packaging technology by China's SJ Semi [19], will determine the direction of the next phase of supply chain reorganization after the memory bottleneck is exhausted.