The Reversal of Korea-Taiwan Exports to Japan Amid the AI Semiconductor Boom and Shifts in East Asia's Supply Chain Bottleneck Status

Executive Summary

The reversal of Korea's and Taiwan's exports to Japan in the first half of 2026 is a structural achievement stemming from holding two bottlenecks: HBM and advanced foundry. However, the fact that China's semiconductor exports surged 99.5% and Cambricon's revenue grew 108% over the same period shows that the shelf life of this bottleneck is not unlimited. Among three branching scenarios, the most probable path is a gradual erosion of bottleneck status, with China's indigenization of legacy and mid-tier AI accelerators likely to slowly erode the revenue base of Korea and Taiwan. The Korean government and firms must simultaneously pursue tax and investment support to defend the HBM/foundry lead, and an offensive strategy to preempt the next points to which the bottleneck may shift—optical transceivers, next-generation packaging, and power semiconductors. Both complacency about the current advantage and, conversely, overreacting to future uncertainty in a way that abandons the existing decisive lead are risks that must be guarded against.

I. Issue Situation Analysis

The Reversal of Korea-Taiwan Exports to Japan Amid the AI Semiconductor Boom: Issue Situation Analysis

1. Background and Development

According to a Nikkei tabulation, Korea's export value in the first half of 2026 stood at $496.3 billion[1][5]. Over the same period, Taiwan's export value was $416.6 billion[1][5]. Japan's export value amounted to only $384.4 billion[1][5]. This was calculated by combining Korean, Japanese, and Taiwanese trade statistics with JETRO and UN Comtrade databases[5]. The gap was even more pronounced in year-on-year growth rates. Japan's growth rate remained around 10%[1][5]. Korea and Taiwan both approached 50%[1][5].

The roots of this reversal lie in the longstanding division of labor within East Asia's semiconductor industry. Japan has long held roughly 90% of the global market in materials such as fluorinated polyimide and photoresist, rather than in finished chips[2]. It has positioned itself as a country controlling the upstream supply chain. In contrast, Korea has dominated the finished-chip stage through SK hynix's high-bandwidth memory (HBM), and Taiwan through TSMC's advanced foundry[2][4]. As generative AI-driven data center investment gained momentum from 2025, the distribution of value added within this division-of-labor structure tilted sharply. A structure in which profits concentrate in finished chips and modules rather than in materials and equipment has become entrenched[2].

The very fact that Nikkei reported this directly using the expression "for the first time in history" is itself evidence of the sense of crisis within Tokyo's policy circles[2]. Around Japan's Ministry of Economy, Trade and Industry, discussions have recurred that the center of gravity of semiconductor industrial policy should shift back from materials and equipment to finished-chip production, but aside from Rapidus, tangible results in rebuilding an advanced logic foundry have yet to materialize.

2. Current Situation

The Korean press has largely embraced this statistic as a qualitative transformation of its own export structure. Hankyoreh reported that surging demand for AI semiconductors drove the first-half export reversal, identifying SK hynix's expanded HBM production as the core driver[5]. Nikkei, by contrast, frames the same fact as "Japan's relative stagnation," while also noting improved performance at Japanese materials and components companies such as Ibiden and Rohm, highlighting that Japan has not been entirely sidelined from the AI boom[2][14]. Indeed, Ibiden raised its FY2026 profit forecast on the strength of demand for AI servers and substrates[14]. The tone of Japanese media leans toward a defensive balance: "acknowledge the reversal, but Japan retains its status as a materials powerhouse."

The Chinese perspective is markedly different. Global Times reported that China's integrated circuit exports from January to July 2026 reached $216.02 billion, up 99.5% year-on-year, characterizing this as the result of "technological breakthroughs and industrial chain upgrading" rather than AI demand[7]. Citing Chinese experts, it emphasizes that the primary driver of the export surge is not external demand but internal technological self-reliance[7]. Cambricon's 108.13% year-on-year revenue growth in the first half of 2026 is treated in the same vein[12]. Even as U.S. semiconductor export controls on China persist, Singaporean media have also assessed that China's export engine has not been significantly damaged. While a U.S. review of additional restrictions on data center components such as optical transceivers is under discussion, analysis suggests that such items account for a limited share of China's overall exports[10].

Taiwan's DigiTimes tracks on a weekly basis the competitive landscape across the entire supply chain encompassing HBM, DRAM, advanced packaging, and next-generation lithography[15]. This shows that Taiwan's industrial community perceives this export reversal not as a one-off windfall but as a structural advantage spanning the entire foundry and packaging ecosystem.

3. Key Actors and Positions

Korea has, riding on SK hynix's dominant position in HBM, generally viewed this reversal as an achievement of semiconductor industrial upgrading[5]. However, as Professor Kwon Seok-jun pointed out at an EAI academic forum, "as the center of gravity of existing geopolitical logic begins to shift toward semiconductors and, more recently, AI, techno-geopolitical logics are developing in earnest," and policy authorities are also aware that bottleneck status itself is fluid[6]. Professor Kwon assesses that China's semiconductor industry is moving away from "low-value-added chip production" and is "entering in earnest a phase of qualitative transformation"[6].

Taiwan shares a similarly beneficial structure with Korea based on TSMC's advanced foundry advantage, but the nature of its geopolitical risk differs. The Taiwan Strait risk is a variable Taiwanese firms cannot control themselves, and it is driving demand for supply chain diversification among customer firms in the United States, Japan, and elsewhere[2].

Japan has lost ground in the competition for finished chips but is trying to preserve its position at the upstream end of the supply chain by maintaining profitability in the materials, components, and equipment sector (Ibiden, Rohm, and others)[2][14]. Tokyo's policy circles may use this reversal as an opportunity to strengthen the policy rationale for advanced logic semiconductor rebuilding projects such as Rapidus.

China has, despite U.S. export controls, accelerated the pace of technological self-reliance through the first (2014), second (2019), and third (2024) phases of its semiconductor "Big Fund"[6]. Global Times describes this as a "technological breakthrough," reinforcing the self-reliance narrative[7]. Professor Kwon Seok-jun assesses that "most of the technological sanctions measures the United States has taken against China... are now largely not working, or China is developing strategies to disable or circumvent them"[6].

The United States is considering additional export controls on data center-related components, but questions have been raised about their effectiveness given the limited share such items hold in China's overall exports[10].

4. Key Points of Contention

The first point of contention is whether this reversal is structural or temporary. Given that the surge in Korean and Taiwanese exports coincides with the peak of the AI data center investment cycle, the possibility that bottleneck status could be readjusted if the investment cycle enters an adjustment phase cannot be ruled out[3].

The second point of contention is the pace of China's technological self-reliance. Interpretations diverge sharply between Korean/Taiwanese and Chinese media over whether the 99.5% increase in semiconductor exports merely reflects riding the coattails of AI demand or actually reflects a genuine narrowing of the technology gap[6][7].

The third point of contention is Japan's response strategy. Whether Japan can extend its materials and components advantage into the finished-chip domain, or whether it will remain confined to the role of an upstream supply chain nation, is the crossroads facing Japanese semiconductor policy[2][14].

The fourth point of contention concerns the timing at which Taiwan Strait risk could function as a variable in supply chain restructuring. Pressure for supply chain diversification from countries heavily dependent on Taiwanese foundries represents both an opportunity for Korea and, simultaneously, a shared risk should geopolitical tensions escalate[2].

II. In-Depth Issue Analysis

The Reversal of Korea-Taiwan Exports to Japan Amid the AI Semiconductor Boom: In-Depth Issue Analysis

1. Root Cause: Restructuring of the Value-Added Distribution

The surface-level cause of this export reversal is the expansion of AI data center investment. But behind it lies a more fundamental change: a shift in the point of value-added distribution within the semiconductor value chain. Japan has held roughly 90% of the global market in materials such as fluorinated polyimide and photoresist[2]. This status had been a stable source of revenue for decades. The problem is that what the generative AI boom demands is not increased volume of materials, but the extreme advancement of specific process stages—namely, HBM and advanced logic chips.

Professor Kwon Seok-jun described this shift as follows: "as the center of gravity of the existing geopolitical logic we are familiar with begins to shift toward semiconductors and, more recently, AI, techno-geopolitical logics are developing in earnest"[6]. This is a structure in which profits concentrate at the downstream stage of finished chips and modules rather than at the upstream stage of materials and equipment supply. As a result of Korea's SK hynix holding an irreplaceable bottleneck in HBM and Taiwan's TSMC holding one in advanced foundry, both countries' export values simultaneously recorded growth rates approaching 50%[1][5]. Japan, over the same period, saw growth of only around 10%[1][5]. The fundamental cause of this gap lies in which stage of the value chain each country occupies. The demand shock of the AI boom is closer to having exposed the vulnerabilities of the existing division-of-labor structure.

2. Structural Context: A Triple Structure Entangling Politics, Economics, and Security

Economic Structure: Concentration and Dispersion of Bottleneck Points

The semiconductor supply chain has a multi-stage structure running from raw materials → materials → equipment → design → foundry → packaging → modules. At each stage, a bottleneck exists in which a small number of firms hold an oligopoly. Japan has formed a bottleneck at the materials and equipment stage, the Netherlands' ASML at the lithography equipment stage, Taiwan's TSMC at the advanced foundry stage, and Korea's SK hynix and Samsung Electronics at the HBM/DRAM stage. The AI boom has concentrated demand particularly on the foundry and HBM stages among these various bottlenecks. Roland Berger's analysis notes that the data center investment boom is expanding to a scale that is testing existing limits across power, cooling, and networking domains[3]. This suggests that the location of the bottleneck is not fixed but shifts according to the structure of demand. It is currently concentrated on HBM and foundry, but the possibility that the center of gravity could shift to other bottlenecks such as next-generation packaging or optical transceivers cannot be ruled out[15].

Security Structure: Taiwan Strait Risk and US-China Technological Decoupling

The fact that Taiwan constitutes one pillar of this export reversal simultaneously implies geopolitical vulnerability. The higher the dependence on TSMC, the greater the risk that the entire global semiconductor supply chain would be exposed in the event of a Taiwan Strait contingency. This is both a reflected benefit and a potential risk for Korea. If problems arise in Taiwan, demand could shift toward Korean foundries, but at the same time, a decline in overall regional supply chain confidence could also deal a blow to Korea.

US-China technological competition adds yet another layer to this structure. The United States has continued export controls against China under the "small yard, high fence" approach[8]. However, China has expanded its scale through the first (2014), second (2019), and third (2024) phases of its semiconductor Big Fund[6], and from the third phase onward has been broadening its application scope beyond AI semiconductors to mobility, power, telecommunications, and biotechnology across the board[6]. Global Times characterizes China's integrated circuit exports of $21.6 billion from January to July 2026, a 99.5% increase, as the result of "technological breakthroughs and industrial chain upgrading"[7]. Some also assess that the United States' targeted sanctions are being substantially neutralized by China's circumvention and self-reliance strategies[6]. Even the U.S. review of additional restrictions on optical transceivers is criticized as having limited effectiveness given the small share such items hold in China's overall exports[10]. It is a situation in which export controls, which began as a security-driven logic, are being offset by economic logic.

Political Structure: The Sense of Crisis and Differing Response Temperatures Across National Policy Circles

The very fact that Japan reported this reversal directly using the phrase "for the first time in history" reflects the sense of crisis within Tokyo's policy circles[2]. Discussions about rebuilding an advanced logic foundry through Rapidus have continued around Japan's Ministry of Economy, Trade and Industry, but visible results remain limited so far. Meanwhile, although the Korean government has put forward sovereign AI as a core national agenda item, academic circles point out that a sophisticated design for exactly what industrial breakthrough this strategy would open up remains lacking[6]. China has renewed its semiconductor industrial policy in five-year cycles through its state-led Big Fund, aligning policy cycles with the tenure of the Xi Jinping administration[6]. While all three countries treat semiconductors as a national strategic industry, they show clear differences in the sophistication of policy design and the speed of implementation.

3. Historical Precedents: A Recurring Pattern of Bottleneck Migration

This case overlaps substantially with prior instances in semiconductor industry history in which bottleneck points shifted. In the 1980s, Japan dominated the DRAM market and overwhelmed the United States. However, from the 1990s onward, it ceded memory and foundry leadership to Korea and Taiwan and retreated upstream into materials and equipment. This AI boom-driven export reversal can be seen as a moment, some 30 years after that retreat, in which Japan is once again being pushed out of the downstream stage. What differs this time, however, is that it is not a complete exit. As shown by the case of Ibiden raising its FY2026 profit forecast on the strength of demand for AI servers and substrates[14], Japan continues to maintain solid profitability at the upstream materials and components stage. This is closer to a shift of the center of gravity downstream rather than a complete disappearance of the bottleneck.

Another precedent is the rise of Taiwan's foundry industry in the late 1990s and early 2000s. At the time, TSMC and UMC created a new division-of-labor model called the foundry within an industry structure previously centered on IDMs (integrated device manufacturers). By separating design from production, they lowered entry barriers for fabless firms, and as a result, the foundry stage itself grew into a new bottleneck. TSMC once again becoming a key beneficiary in this AI boom demonstrates that this structural position has persisted for over 20 years. This suggests that the location of a bottleneck does not easily shift even as technology generations change.

China's current trajectory shows a pattern similar to Japan's path to semiconductor self-reliance in the 1970s-80s and Korea's in the 2000s: state-led capital injection, scale-up leveraging the domestic market, and a sequence of adopting advanced technology before replacing it with indigenous capabilities. The case of Cambricon's 108.13% year-on-year revenue growth in the first half of 2026[12] shows that China is entering a trajectory toward self-reliance even at the design stage. However, the conditions differ in that the past successes of Korea and Taiwan occurred in an environment relatively open to technology transfer from the United States and Japan, whereas China is attempting self-reliance amid comprehensive U.S. export controls. The pace of self-reliance could be slower than in past cases, or, conversely, faster than expected as controls amplify the incentive for domestic substitution.

4. Key Variables in the Development of This Issue

The first variable is the pace of China's self-reliance in advanced foundry processes. Currently, China's export surge is understood to be substantially composed of mature-process chips and applied products. If China secures stable yields at the sub-7nm node, the very scarcity of the bottleneck held by Korea and Taiwan would be shaken. The key question is how much U.S. export controls on China can delay this pace. However, trends thus far suggest that the effectiveness of these controls is gradually diminishing[6][7].

The second variable is whether the AI data center investment cycle will be sustained. Roland Berger also flags the risk that the data center boom could face an adjustment phase[3]. The current export reversal is heavily dependent on this particular demand cycle of AI infrastructure investment. If this investment cycle stalls sooner than expected, HBM and foundry demand could slow in tandem. In that case, the export advantage of Korea and Taiwan could turn out to be a temporary phenomenon.

The third variable is the shift of the bottleneck point itself. HBM and advanced foundry are currently the key bottlenecks, but there are signs that new bottlenecks are emerging in adjacent technology areas such as next-generation packaging, optical transceivers, and next-generation lithography[15]. For instance, if alternative technologies such as X-ray lithography reach the commercialization stage, the ASML-centered lithography equipment bottleneck structure itself could be restructured[15]. The reshuffling of beneficiary and marginalized nations each time the bottleneck shifts is a recurring pattern in this industry.

The fourth variable is whether geopolitical risk surrounding the Taiwan Strait materializes. Given the structurally elevated dependence on TSMC, if cross-strait relations deteriorate, the shock to the entire supply chain would be of a magnitude capable of instantly neutralizing the profit structure created by this export reversal. This risk differs in character from the other variables in that its probability is low, but its impact if realized would be overwhelmingly large.