Capital Departs, Factories Hum: The Dual Timelines Facing Samsung Electronics and SK Hynix

Executive Summary

As foreign capital outflows from Taiwanese and South Korean stock markets coincide with a shortage of memory semiconductors, a disconnect has emerged between financial market sentiment and the timeline of the physical supply chain. Micron's record-high earnings and the Phison CEO's projection of a "supply gap persisting for several years" support the view that memory supply and demand remain structurally tight; however, the reassessment of AI-related stocks is proceeding independently of these fundamentals. Amid this divergence, Samsung Electronics and SK Hynix should maintain their pace of capacity expansion while shifting their financing methods from equity-based options to internal reserves and corporate bonds. Simultaneously, they need to reduce customer portfolio concentration and adopt a strategy of disclosing specific details regarding their order backlogs and long-term contract proportions in investor communications. The two variables of China's localization pace and US pressure on investment in South Korea must be constantly monitored as key conditions shaping future scenario branches.

I. Issue Analysis

Capital Flees Taiwan and South Korea, But Why Is It Not Going to India?

Mumbai's Question: "Where Did All the Money Go?"

The question that began circulating in Indian financial circles in late September was simple: why was the foreign capital leaving Taiwanese and South Korean stock markets not flowing into India? Mumbai-based Mint noted that "outflows have begun in the AI market" and wrote that "India is not receiving that money"[1]. Commenting on this situation, StockKart's Agarwal diagnosed that "domestic absorption capacity remains the primary stabilizer until macroeconomic indicators improve," suggesting it is premature to expect sustained capital reallocation[17]. Just months after the KOSPI and the TAIEX were hailed as the biggest beneficiaries of the AI rally, the market's focus is now shifting toward the sustainability of that rally.

The starting point of this trend dates back to the first half of 2026. According to data compiled by The Nikkei, South Korea's exports reached $496.3 billion and Taiwan's reached $416.6 billion, surpassing Japan's $384.4 billion for the first time in history[3][11]. This was the result of AI data center investments shifting the computational bottleneck from advanced node competition to HBM and advanced packaging[3][7]. This structural shift provided the fundamentals for the surge in the KOSPI and TWSE, and the launch of OpenAI's new model in September added short-term momentum[3]. The problem is now, as that momentum begins to fade.

Memory Shortages, Yet Why Are Stock Prices Cooling?

The paradox is that the supply and demand of memory semiconductors themselves are actually tightening. K.S. Pua, CEO of Taiwanese PC storage company Phison, said in an interview with Digitimes Asia, "Every morning I tell myself tomorrow will be better"[19]. Nevertheless, he projected that the supply gap created by AI-driven demand "could last for several years"[19]. Micron broke its historical record in its quarterly earnings announced in October. The Hankyoreh analyzed this as being due to "surging memory demand centered on AI data centers, coupled with supply shortages"[12]. The same article added that "although calls for pacing AI development are being raised," the capital expenditure competition among hyperscalers remains hot[12].

Nevertheless, investment capital is showing signs of leaving Taiwanese and South Korean stock markets during the same period[1]. Nikkei Asia explains this divergence with the phrase "the supply chain that ignores AI slowdown talk." It cites as evidence that just three days after Anthropic CEO Dario Amodei's remarks questioning the pace of frontier model development triggered a sharp drop in AI-related stocks, thousands of suppliers gathered at a Silicon Valley convention center[5]. In other words, local media commonly observe that physical demand and financial market sentiment are moving on different timetables.

Three Countries, Three Calculations

In this phase, the interests of each actor diverge clearly. From Taiwan's perspective, there is confidence that a single company, TSMC, still holds the very top of the AI supply chain. According to The Business Times (Singapore), TSMC is considering a new multi-billion dollar campus in Texas[16]. This is an extension of its overseas production base expansion to meet US demand. At the same time, Taiwanese and Chinese PCB manufacturers are reportedly flocking to the Hong Kong stock market to raise funds to cope with the upgrading of AI server and optical module specifications[9]. This indicates that the channels for attracting investment capital are diversifying.

South Korea differs from Taiwan in that the sensitivity of capital inflows and outflows is concentrated on just two memory companies: Samsung Electronics and SK Hynix. On a Council on Foreign Relations (CFR) podcast, a panelist summarized, "Taiwan is making money from the AI boom with TSMC, and Korea is making money with good memory chips"[10]. In other words, South Korea's AI benefits are structured narrowly but with concentrated depth. Accordingly, the sensitivity of foreign capital is heavily focused on these two stocks. India is a complete outsider in this dynamic. Mint projected that while there is a possibility of capital returning to India if foreign investors reallocate funds beyond the "crowded Taiwan-Korea AI trade," external market pressures could block immediate or sustained capital movement[17].

The China variable also casts a shadow over this landscape. Digitimes Asia reported that China's AI infrastructure investment is expanding beyond large tech companies to mid-sized enterprises[14]. As the pace of China's localization of proprietary technologies—such as by CXMT and Cambricon—accelerates, the scarcity premium of the bottleneck positions held by South Korea and Taiwan will inevitably erode gradually[7][11].

The Issue Converges on One Question: When Will "Peak Memory Shortage" Arrive?

The core issue of the current phase ultimately converges on a question of timing: when will the memory supply shortage peak and when will it enter an easing phase? There is a clear time lag between the perspective of supply chain insiders, who believe the shortage will persist "for several years" like the Phison CEO[19], and the perspective of the financial market, which sees signs of investment capital already cooling[1]. In other words, the time it takes for the supply shortage to be felt in the real economy and the time it takes for that shortage to be priced into and then discounted from stock prices are moving at different speeds.

Macroeconomic variables also overlap here. The Peterson Institute for International Economics (PIIE) projected that global economic growth will slow to 3.0% in 2026 from 3.3% in 2025, while noting that the AI boom is acting as a key driver supporting growth despite this slowdown[13]. However, it added the caveat that high oil prices, trade tensions, and policy uncertainty are acting as variables weighing on this growth[13]. Whether AI investment will remain an exceptional growth engine even amid a macroeconomic slowdown, or whether it will be swept away by the waves of deceleration, is the next turning point that will determine capital flows in the Taiwanese and South Korean stock markets.

The Carnegie Endowment for International Peace takes this perspective a step further. It assessed that while the United States and China are widening the gap in AI talent production and attraction, "Asia, excluding China, is rising, while Europe is faltering"[2]. South Korea and Taiwan are the core pillars of this "rising Asia" and, at the same time, are positioned on the front lines of the US-China tech hegemony competition. The fact that this structural reorganization lies behind the short-term volatility of investment capital is why this phase is more than just a matter of simple profit-taking.

II. In-Depth Analysis

Between Stock Prices and Wafers: Why Do Memory Semiconductors Move on Dual Timetables?

A Rally Built on FOMO, Ended by FOMO

To understand the signs of this capital outflow, we must first look back at why that capital flowed in to begin with. A significant portion of the foreign capital that rushed into the KOSPI and TWSE did not enter in pursuit of structural reorganization in HBM and foundries. Instead, short-term events like the launch of OpenAI's new model served as triggers, and a substantial portion consisted of momentum capital riding those triggers[3]. Capital that enters on momentum is the first to leave the moment that momentum fades. Agarwal of StockKart pointed out that "domestic absorption capacity remains the primary stabilizer until macroeconomic indicators improve"[17]; viewed conversely, this means that foreign capital is reacting more sensitively to liquidity conditions than to fundamentals, whether in India or other emerging markets.

The root cause here is not a simple valuation adjustment. The core issue is that the reassessment of AI-related stocks overall is proceeding independently of the supply-demand fundamentals of the specific memory semiconductor sector. This is illustrated by the case where AI-related stocks shook in unison due to a single remark by Anthropic CEO Dario Amodei[5]. When doubts were raised about the pace of frontier model development, investors treated the entire sector as a single bet and sold off simultaneously. This sell-off was unrelated to actual order volumes or inventory levels of memory chips. It is the result of the financial market's price discovery mechanism operating much faster—and much more harshly—than the speed of information transmission in the physical supply chain.

Where Three Layers Overlap

Structurally, three different layers overlap at this single point. The first is the macroeconomic layer. The Peterson Institute for International Economics (PIIE) projected that global economic growth will slow to 3.0% in 2026 from 3.3% in 2025, analyzing that high oil prices, trade tensions, and policy uncertainty are weighing on activity[13]. Thus, the diagnosis that AI investment is virtually the only engine supporting global economic growth[13] coexists with a situation where doubts are being raised about the sustainability of that very engine. The lack of other viable growth engines to rely on if AI investment falters is the backdrop fueling investor caution.

The second is the industrial layer, namely the shift of bottlenecks within the semiconductor supply chain. The memory shortage is the result of simultaneous demand surges and supply constraints, not a temporary supply-demand mismatch. Phison CEO K.S. Pua stated that the supply gap "could last for several years"[19], which means that memory manufacturers' capital expenditure cycles are failing to keep pace with the growth rate of AI accelerator demand. Converting production lines for HBM typically requires a lead time of over a year. A structural time lag exists between the confirmation of demand signals and the subsequent supply response. This lag serves as the basis for counterarguments that the "peak memory shortage" theory is premature.

The third is the geopolitical layer. The bottleneck positions held by Taiwan and South Korea are not geopolitically neutral. The United States continues to pressure South Korean companies to secure memory and foundry production bases within its borders[7], while TSMC is expanding its overseas production capacity, including by considering a multi-billion dollar new campus in Texas[16]. The Carnegie Endowment for International Peace diagnoses that in the competition over compute capacity, "whoever can get chips up and running fastest will have a decisive advantage"[6], meaning that the investment decisions of South Korean and Taiwanese companies are not driven solely by pure market logic. The geographical dispersion of production bases itself is acting as a political variable.

Differences from the DRAM Cycles of the 2000s

Boom-and-bust cycles in memory semiconductors are not a new phenomenon. The sharp fluctuations in DRAM prices in the 2000s and the supercycle of 2017–2018 are cited as precedents. Those cycles moved in tandem with relatively predictable end-demand, such as PC and smartphone replacement demand or server expansions. When demand cooled, inventories accumulated and prices plummeted in a repeating pattern.

Where this phase differs from the past is that the stagnation of end-demand remains unclear. Hyperscalers' data center investments are tied to the competition to develop frontier AI models themselves, rather than to specific product cycles. Bain & Company assessed that US data center capital expenditure as a share of GDP is comparable to the eras of building railroads, telecommunications networks, and electricity grids[4]. For an investment cycle of this scale to break, the competition to develop AI models itself would have to stop, but the scene at the Silicon Valley convention center reported by Nikkei Asia[5] is far from showing such signs. This means that, unlike past cycles, an "investment sentiment adjustment" is highly likely to precede any "disappearance of end-demand," which is another pillar explaining the current divergence between stock prices and physical supply and demand.

The China variable is also an element that was absent in past cycles. Digitimes Asia reported that China's AI infrastructure investment is expanding beyond large tech companies to mid-sized enterprises[14]. This acts as a double-edged variable: Chinese domestic server and data center demand serves as an additional source of demand for South Korean and Taiwanese memory, while simultaneously driving the technological catch-up of China's own memory companies like CXMT[7].

Reading the Reason India Cannot Receive Capital in Reverse

Reversing the analysis from the Indian perspective makes the structure of this capital flow even clearer. The Carnegie Endowment for International Peace points out that middle powers like Germany and India "cannot build frontier AI on their own and face the risk of being cut off if they rely on the US or China"[15]. While India possesses AI software and service capabilities, it does not hold the hardware bottlenecks controlled by TSMC, Samsung Electronics, and SK Hynix. A CFR analyst summarizes this structure succinctly: "Taiwan is making money from the AI boom with TSMC, and Korea is making money with good memory chips. The Netherlands has ASML lithography equipment. But outside of the top two or three countries, most countries have to successfully utilize AI to make a profit"[10]. The reason foreign capital is not moving to India is not a problem with the attractiveness of the Indian market, but rather because this rally was structurally tied from the beginning to a handful of countries holding hardware bottlenecks. When that capital is shaken by a sector-wide reassessment, its destination is other stocks holding the same bottlenecks, not markets with entirely different structures.

What to Watch Over the Next 12 Months

The key variables shaping future developments narrow down to four. First, whether the upcoming quarterly guidance from Micron and SK Hynix reaffirms the current supply shortage outlook. The market's judgment on whether their record-high earnings are a one-off surprise or the beginning of a multi-year trend will diverge here[12][19]. Second, the timing of hyperscalers' capital expenditure budget disclosures. Whether the Silicon Valley enthusiasm reported by Nikkei Asia[5] is confirmed by actual CapEx figures will be a turning point for a rebound in investment sentiment. Third, how US pressure on South Korean investment takes concrete shape[7]. Depending on how much the demand to relocate production bases erodes South Korean companies' domestic investment capacity, the assessment of the sustainability of their bottleneck status will vary. Fourth, the pace of technological localization by China's CXMT[7][14]. At the intersection of these four variables, it will be decided whether the current capital outflow is a temporary breather or the start of a structural reassessment.