The AI Semiconductor Rally and US-China Tech Rivalry: The Future of South Korea and Taiwan’s Bottleneck Status

Executive Summary

In the first half of 2026, the export values of South Korea and Taiwan surpassed Japan’s for the first time ever. This is a structural outcome of the bottleneck in AI computing architecture shifting from advanced processes to HBM and advanced packaging. While the KOSPI and TWSE surged in September following the release of a new OpenAI model, this phenomenon is a combination of strong earnings fundamentals and a one-off event, making its sustainability difficult to confirm. Considering the pace of technology indigenization by Chinese firms like CXMT and Cambricon, it is most likely that South Korea and Taiwan’s bottleneck status will be gradually eroded over the next 12 to 18 months, rather than collapsing abruptly. South Korea needs a dual-track strategy: simultaneously defending its technological lead in HBM and foundries while preemptively investing in next-generation bottlenecks, and using the expansion of its domestic production bases as a bargaining chip against US pressure to invest.

I. Situational Analysis

Situational Analysis: The AI Semiconductor Rally and US-China Tech Rivalry

Background and Developments

The roots of the recent surge lie in the restructuring of the East Asian semiconductor division of labor that has been underway since the first half of 2026. 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, 6]. During the same period, Japan’s exports were only $384.4 billion [3, 6]. This is the first time ever that South Korea and Taiwan have surpassed Japan in export value [6, 9]. The gap was even more pronounced in terms of year-on-year growth rates. Japan’s growth remained around 10% [9]. South Korea and Taiwan both recorded growth rates approaching 50% [6, 9].

This reversal is intertwined with changes in AI computing architecture. As investment in AI data centers has expanded, the competition over bandwidth and packaging has replaced the competition over advanced processes as the core driver of value-added [3]. This is why value-added has become concentrated in South Korea and Taiwan, which control HBM and advanced foundries [3, 6]. Japan has an industrial structure specialized in materials like fluorinated polyimides and photoresists, rather than in finished chip production [6, 9]. Therefore, a key EAI diagnosis is that Japan has been relatively excluded from the direct benefits of the AI boom [6].

Within Taiwan, this trend is already materializing as industrial restructuring. The president of Tokyo Electron (TEL) Taiwan stated that AI is pushing semiconductor demand into new areas such as advanced packaging, DRAM, HBM, and NAND flash [1]. Belgian chemical company Solvay has decided to double its production capacity for high-purity semiconductor chemical materials in Taiwan by 2026 [13]. Taiwan’s Tongtai Group announced that it has doubled its semiconductor-related orders, driven by demand from AI data centers [19].

Current Situation

The rally steepened in September. According to a Yonhap News report, on September 7, the Seoul stock market surged 4.6% to a five-week high, driven by demand for AI-related stocks triggered by the release of a new model from OpenAI [12]. Portugal’s Jornal de Negócios reported that on the same day, the KOSPI rose by 4.37%, with SK Hynix and Samsung Electronics leading the gains, soaring by 8.14% and 5.28%, respectively [15]. On the same day, Taiwan’s TWSE rose by 1.67%, and the MSCI Asia Pacific Index increased by 1.4% [15].

Singapore’s The Business Times reported that over the past six months, the Taiwanese stock market has risen by about 35% and the South Korean market by about 15% [10]. Standard Chartered economists Jonathan Koh and Edward Lee questioned how long this rally could be sustained [10]. Meanwhile, a perception is spreading among some investment banks that bets on South Korean and Japanese AI stocks have become overcrowded. Trading desks at firms like Barclays and UBS are seeing increased client demand for options and swap contracts linked to China’s CSI Index [4]. UBS is presenting the CSI 500 as an alternative investment to South Korea and Japan [4].

China’s state-run Global Times interprets this trend as a sign of the limits of US tariff strategy. It pointed out that the surge in imports of computers, computer components, and semiconductors, driven by the AI investment boom, was behind the expansion of the US trade deficit in July to a 16-month high [7]. Chinese experts emphasize the contradiction that at a time when the US is trying to reduce its dependence on its own manufacturing, the AI industry is becoming increasingly reliant on global supply chains and cross-border cooperation [7]. This leads to the argument that global cooperation, not confrontation, is necessary [7].

The next battleground in the technological competition is shifting to silicon photonics. Taiwan’s United Daily News (UDN) reported that both TSMC and Nvidia have entered the race to secure silicon photonics technology [16]. In just three months during the first half of 2026, Nvidia invested over $6 billion in the optical interconnect and silicon photonics supply chain [16]. Marvell, after investing $3.25 billion, completed its acquisition of Celestial AI to secure Photonic Fabric architecture technology [16].

Analysis of Key Actors

South Koreais the biggest beneficiary of the current rally but also faces structural vulnerabilities. As demonstrated by SK Hynix and Samsung Electronics leading the KOSPI surge, its dominance of the HBM supply chain is the substantive basis for the stock market gains [15]. However, an EAI analysis warns that this bottleneck status “may not last indefinitely” [3]. The rapid growth of China’s CXMT and Cambricon is the basis for this warning [3, 9]. The United States is pressuring South Korean companies to invest in the US to secure domestic memory production bases [3]. From the perspective of South Korean firms, this creates a resource allocation dilemma, caught between investing in domestic production bases (in Yongin and Honam) and meeting US demands for investment [3].

Taiwanled by TSMC’s foundry dominance, is making clear moves to preemptively capture the next-generation bottlenecks in silicon photonics and advanced packaging [16, 14]. The expansion of AI computing infrastructure is being treated as a matter of national competitiveness, to the extent that the supercomputers at Taiwan’s government-funded research institutes have already reached their operational capacity limits [14]. Discussions are also taking place within Taiwan at the national security level. In an interview with the Brookings Institution, a Taipei official assessed that the AI boom is serving as a factor strengthening economic ties between Taiwan and the United States, even amid US-China tensions [11].

Chinais demonstrating the pace of its technological self-sufficiency with figures such as a 99.5% increase in semiconductor exports and a 108% rise in Cambricon’s revenue [9]. State-run media, including the Global Times, are highlighting the declining effectiveness of US tariff strategy and framing their country’s efforts to indigenize optical modules and AI accelerators as a counter-narrative to the US containment strategy [7]. An assessment from a CFR interview suggests that the US-China AI competition has entered a new phase of competition over new model releases [2].

United Statesis facing the limits of its policy tools, namely tariffs and pressure on China. The paradoxical result of a widening trade deficit is evidence of this [7]. At the same time, US companies like Nvidia and Marvell are pursuing a parallel strategy of aggressively investing to dominate the optical interconnect and silicon photonics supply chains, aiming to technologically capture the next-generation bottleneck [16].

Japanis an indirect beneficiary of the AI boom in the materials and components sector, but it is in a structural position of lagging behind South Korea and Taiwan in finished chip exports [6, 9]. According to an EAI assessment, the fact that the Japanese media directly reported this reversal using the phrase “for the first time ever” reflects a sense of crisis within Tokyo’s policy circles [6].

Key Issues

The first issue is the sustainability of the HBM and foundry bottleneck status held by South Korea and Taiwan. An EAI analysis views the next 12 to 18 months as a litmus test for whether this bottleneck status will erode [3]. The three key variables are the ability of South Korea and Taiwan to maintain their technological lead, the pace of China’s indigenization, and the intensity of US pressure [3].

The second issue is the trajectory of the bottleneck’s shift. Silicon photonics, optical transceivers, and next-generation packaging are emerging as the next battlegrounds [16, 9]. The key question is whether South Korea and Taiwan can keep pace with the US-Nvidia alliance in this domain [9].

The third issue is physical bottlenecks. Taiwan’s DigiTimes points out that supply constraints on raw materials, including copper, could threaten the AI supercomputing boom [17]. Indeed, copper prices on the London Metal Exchange have hit a record high, driven by a combination of concerns over US tariffs and demand for AI infrastructure [18].

The fourth issue is whether investor sentiment is overheating. As the perception spreads that bets on South Korean and Taiwanese AI stocks are already overcrowded, there is a move to diversify funds into the Chinese stock market [4, 10]. The question raised by Standard Chartered—“how long can it be sustained?”—is a key metric for distinguishing between the rally’s fundamentals and its speculative elements [10].

II. In-Depth Analysis

In-Depth Analysis: The Structural Background of the AI Semiconductor Rally and US-China Tech Rivalry

1. Root Cause: A New Bottleneck Created by a Shift in Computing Architecture

Explaining the current rally solely as a one-off event triggered by OpenAI’s new model release is a superficial interpretation [12]. The deeper cause lies in the shift in AI computing architecture itself. The axis of competition in the semiconductor industry has shifted from advanced processes to bandwidth and packaging [3]. This shift has relocated the bottleneck.

In the past, the core of semiconductor competitiveness was process scaling in nanometers. The competition progressed from 22nm and 14nm down to 7nm and 3nm [20]. AI has changed these constraints. While processor speeds are increasing, moving data fast enough to keep up is becoming progressively more difficult [20]. This is the context for the competition among Apple, Xiaomi, and Nvidia to secure terabyte-level bandwidth [20].

This shift is structurally advantageous for South Korea and Taiwan. This is because they control two key bottlenecks simultaneously: HBM and advanced foundries [3, 9]. Conversely, Japan, with its materials-focused industrial structure, has been relatively sidelined in this trend [6, 9]. The statement by the president of Tokyo Electron Taiwan that demand is expanding into advanced packaging, DRAM, HBM, and NAND flash is consistent with this context [1].

2. Structural Context: The Convergence of Three Layers of Pressure

This issue rests on a structure where three different layers are operating simultaneously.

The first is the layer of trade and economic security. The risk associated with South Korean firms’ dependence on key items falls into this category. The United States is pressuring South Korean companies to invest in the US to secure domestic memory production bases [3]. This pressure is not driven by simple market logic but is directly linked to Washington’s industrial policy objectives.

The second is the layer of US-China strategic competition. China is accelerating the pace of its technological self-sufficiency. China’s semiconductor exports surged by 99.5% during the same period, and Cambricon’s revenue increased by 108% [6, 9]. The rapid growth of memory companies like CXMT also shows that the bottleneck status may not last indefinitely [3]. In response, the US has employed a tariff strategy against China, but its limits were exposed when the US trade deficit widened in July to a 16-month high [7]. Chinese state media assessed this as evidence that “US tariff strategy has shown clear limits,” highlighting the AI industry’s structural dependence on cross-border cooperation [7].

The third is the layer of international competition in AI. There are assessments that the United States and China have already established a significant lead in the competition over frontier AI models [5]. The invitation of AI company CEOs to the G7 summit and parallel discussions at the UN level are also signals that this competition has transcended being a simple industrial issue [5].

These three layers converge into a resource allocation dilemma for South Korean companies [3]. Three demands arise simultaneously: whether to invest in maintaining the technological lead in HBM and foundries, whether to expand local production bases in response to US investment pressure, or whether to make preemptive investments in next-generation bottleneck technologies to prepare for China’s pursuit.

3. Historical Precedent: The Lesson of Japan’s Semiconductor Industry in the 1980s

Forecasts repeatedly suggest that the bottleneck status currently enjoyed by South Korea and Taiwan is not permanent [3, 9]. The basis for this concern can be found in the past experience of Japan’s semiconductor industry.

In the 1980s, Japan dominated the DRAM market. However, its position was gradually eroded by the combined effects of pursuit by South Korea and Taiwan and trade pressure from the United States. This history is also part of the background to why Japan now has an industrial structure specialized in materials rather than finished chips [6, 9]. The fact that Nikkei directly reported the export reversal by South Korea and Taiwan with the phrase “for the first time ever” reflects the sense of crisis within Tokyo’s policy circles [6].

The situation South Korea and Taiwan face today is the opposite of Japan’s at that time. However, the pattern of a bottleneck status being eroded by changes in industrial structure and pursuit by latecomers could operate in the same way. The possibility that China’s indigenization of legacy and mid-range AI accelerators could gradually erode the revenue base of South Korea and Taiwan is the contemporary version of this pattern [9].

The surge in copper prices is also a reenactment of a similar resource bottleneck competition. On the London Metal Exchange (LME), copper futures reached a record high of $14,510 per ton [18]. This was the result of concerns over US tariffs leading to a competition to secure inventory [18]. Given that the physical bottlenecks triggered by AI infrastructure build-out are spreading beyond semiconductors to raw materials markets, this phase should be seen not as a simple technological race but as a competition across the board in resources, materials, and equipment. Taiwan’s DigiTimes referred to this as a “Copper Wall,” identifying it as a physical constraint threatening the AI supercomputing boom [17].

4. Key Variables: Factors Determining the Three Potential Paths

The variables that will shape the development of this issue over the next 12 to 18 months can be condensed into three main factors [9].

First is whether South Korea and Taiwan can maintain their technological lead. The key is how long they can defend their technological gap in HBM and advanced foundries. Nvidia’s investment of over $6 billion in the optical interconnect and silicon photonics supply chain in just three months shows that the bottleneck is already shifting to the next stage [16]. Marvell’s acquisition of Celestial AI to secure Photonic Fabric architecture technology is part of the same trend [16]. Local Taiwanese reports that TSMC and Nvidia are competing in silicon photonics as the new battleground for AI suggest the possibility that the bottleneck could shift from HBM and foundries to the optical domain [16].

Second is the pace of China’s indigenization. If Chinese companies like CXMT and Cambricon grow faster than expected, the erosion of South Korea and Taiwan’s revenue base could be accelerated [3, 9]. China is leveraging this as both an achievement of its technological self-sufficiency and as evidence of the failure of the US containment strategy [7].

Third is the intensity of US pressure. As pressure to invest in the US intensifies, the burden on South Korean companies to divide their resources between domestic investments to maintain their technological lead and the expansion of overseas production bases will grow [3]. The growing perception in investment banking circles that bets on South Korean and Japanese AI are already overcrowded, coupled with rising demand for options and swaps on China-related indices [4], can be read as a signal that the market has already begun to price in the uncertainty of these three variables.