A recent research report from the Bank of Korea reveals a core risk to South Korea's economy: if American technology giants slow their AI capital expenditure next year, South Korean semiconductor exports could fall short of earlier forecasts by $50 billion to $100 billion, dragging down 2027 GDP growth by roughly 0.4 percentage points. This, however, is a pessimistic-scenario estimate and outlook, not an export loss that has already occurred, nor is it a baseline forecast that exports will decline by $100 billion year over year. Meanwhile, Wall Street giant Goldman Sachs, despite cutting its KOSPI target by 25%, still believes memory chip valuations are excessively pessimistic. South Korean retail investors, for their part, are buying back into Samsung Electronics, SK Hynix, and those triple-leveraged ETF products closely tied to the chip theme. These three signals together constitute the current divergence in the AI computing trade theme 鈥?at the macro level, some investors remain persistently worried about a series of major negative shocks from cooling AI investment by North American tech giants amid escalating AI financing cost pressures, while the stock market is beginning to judge whether such pessimistic risk has already been overpriced after valuation metrics for South Korea's two memory chip giants have fallen sharply.

"South Korea cannot afford to lose the AI memory competition" has a realistic basis. South Korea's third-quarter GDP grew 1% quarter over quarter, beating market expectations of 0.7%. The Bank of Korea estimates that semiconductors contributed about half of economic growth, while private consumption fell 0.2% quarter over quarter. The chip boom has already deeply affected exports, corporate investment, and overall economic performance. This is why, in the view of South Korean President Lee Jae-myung and Bank of Korea officials, South Korea needs to consolidate its high-end memory advantage, expand real AI applications and customer sources, and reduce the economy's dependence on a single capital expenditure cycle.

The underlying mechanism of export pressure lies in the fact that shipment volume and average selling price together determine export value. South Korean media disclosed that in the pessimistic scenario, chip export volumes could still grow by about 10%, but below previous expectations, while price performance would be weaker than the baseline assumption. Therefore, even if demand for AI computing resources continues to expand, as long as procurement growth and prices fall below the previously extremely high expectations, it is enough to create a massive export gap. For memory stocks, the key question thus becomes: will future earnings merely experience slower growth, or will they suffer a deep decline under a recession cycle implied by current pessimistic valuations?

The Bank of Korea warns: if AI spending slows, South Korean chip exports could fall short of earlier forecasts by up to $100 billion

Bank of Korea data show that if large American technology companies slow their artificial intelligence capital expenditure next year, South Korean semiconductor exports could fall short of previous forecasts by $50 billion to $100 billion. In a written response submitted to lawmaker Ahn Do-geol, the Bank of Korea constructed a "pessimistic scenario" to assess the economic impact of American tech giants cutting large-scale infrastructure investment. Under this scenario, exports suddenly falling below expectations would push South Korea's 2027 current account surplus significantly below the central bank's initial forecast of $430 billion, and reduce next year's gross domestic product growth by about 0.4 percentage points. The scenario assumes a 10% decline in South Korean semiconductor unit prices. This latest Bank of Korea assessment highlights deepening market concerns over South Korea's heavy reliance on chip exports. Chips currently account for about half of South Korea's total exports. The report said combined capital expenditure by five large American technology companies 鈥?Amazon, Google, Microsoft, Meta, and Oracle 鈥?grew rapidly from $40 billion last year to a conservative estimate of $80 billion this year, and is conservatively expected to reach a staggering $110 billion next year. However, the Bank of Korea cautioned that considerable uncertainty remains regarding the trajectory of demand related to frontier AI technology and the overall memory chip business cycle driven entirely by AI computing demand.

The Bank of Korea calculates the export gap, while Goldman Sachs seeks to find the expectation gap in memory chip stocks

According to a Goldman Sachs research report, the institution lowered the 12-month target for South Korea's benchmark KOSPI Composite Index from 12,000 points to 9,000 points, mainly reflecting more conservative valuation assumptions. Based on the report's timing, the new target still implies more than 60% upside. Goldman Sachs estimates that current semiconductor valuations imply a roughly 65% earnings decline in 2027, a scenario the Goldman Sachs analysts consider overly pessimistic and a sign that market pessimism has gone far too far. Samsung Electronics and SK Hynix were each trading at only about 3 times forward price-to-earnings at the time. The implied return on equity corresponding to the memory chip sector's price-to-book ratio was only about 23%, well below the market consensus expectation of 45%. These figures reflect the distance between market pricing and earnings expectations. The implied 65% earnings decline was calculated based on KOSPI trading levels as an indication of excessive market pessimism, not Goldman Sachs' own forecast and outlook. The Goldman Sachs analyst team noted in the report that Samsung Electronics and SK Hynix are both trading at about 3 times forward price-to-earnings, the lowest level in history. Based on the relationship between price-to-book ratio and return on equity compiled from the past 20 years of monthly data, which shows a goodness of fit (R-squared) of as high as 72%, the current sector price-to-book ratio of 2.6 times implies a 2027 ROE of only 23%, while the market consensus forecast is as high as 45%. This divergence is equally prominent at the index level: the KOSPI has fallen about 40% from its June high, while forward 12-month earnings forecasts rose about 22% over the same period. The KOSPI's record-low forward price-to-earnings of about 4 times is at an extreme low among the 18 major trillion-dollar markets covered by the relevant statistics. However, the memory industry's low price-to-earnings ratio may also contain concerns that cycle-peak earnings are unsustainable. The key to Goldman Sachs' bullish view on South Korea's two memory chip giants and the benchmark KOSPI is that as long as the actual magnitude of earnings downgrades is smaller than the pessimism implied by share prices, combined with a decline in risk premiums, South Korea's stock market and the two memory chip giants that account for 50% of the market capitalization of South Korean stocks could achieve a strong valuation recovery during a period of slower earnings growth.

The Goldman Sachs research report also shows that funds in South Korea's stock market have already begun betting on this expectation gap. From October 1 to 23, South Korean individual retail investors heavily net-bought 14 single-target high-leverage ETF products related to Samsung Electronics and SK Hynix, totaling 315.4 billion won, including 226.3 billion won in Hynix-related products and 89.1 billion won in Samsung-related products. From October 1 to 22, South Korean investors heavily net-bought about $4.8 billion in U.S. market stocks, of which about $3.548 billion flowed into a triple-long U.S. market Philadelphia Semiconductor ETF 鈥?the U.S.-listed ETF with the ticker "SOXL" 鈥?accounting for 73.9%. This shows that bottom-fishing funds are highly concentrated in the memory chip and semiconductor industry chain driven by AI computing demand, but products that reset leverage daily also amplify volatility losses, and the return of funds alone cannot prove that the market has bottomed.

The industry-chain basis for the broader global memory chip sector counterattack is that as frontier AI agents such as Muse become popular worldwide, demand for computing resources at the AI high-performance computing level is spreading to more memory and larger-capacity NAND enterprise-level storage tiers 鈥?high-bandwidth memory (HBM) serves accelerator computing, server DRAM carries data and task environments during operation, and enterprise SSDs save indexes, task records, and persistent states. From the perspective of AI inference system architecture, more complex AI tasks led by Meta Muse and OpenAI's GPT-6 Astra often contain longer contexts, multi-turn model calls, tool execution, and result verification: prefill needs to process input, decode continuously generates output, and KV cache occupies more HBM memory as context and concurrency scale expand, requiring coordinated improvements in computing throughput, memory bandwidth, and enterprise-grade high-performance storage capacity. SK Hynix has disclosed that HBM4 began mass shipments in the second quarter and has established long-term supply agreements with about ten customers. Bank of America, after recent exchanges with Micron management, still emphasized that supply constraints and long-term agreements support demand visibility. Market research firm TrendForce expects fourth-quarter conventional DRAM contract prices to rise 10% to 15% quarter over quarter on a high base, and NAND contract prices to rise 15% to 20%. On the DRAM side, agent workloads drive supporting general-purpose server and memory procurement, while constraints exist in advanced processes, packaging and testing, and capacity allocation. On the NAND side, real-time retrieval, caching, and vector databases expand enterprise SSD demand. TrendForce expects enterprise SSD bit demand to grow more than 80% year over year in 2026, with most new supply already booked, so the price increase for this category in the fourth quarter may even widen compared with the third quarter. TrendForce also sharply raised its 2027 HBM blended ASP forecast to 121% year-over-year growth, based on HBM and conventional DRAM competing for limited advanced process and wafer capacity, the time needed for yield ramp-up of new-generation products, and the rising share of higher-priced HBM4 and the gradual volume ramp of HBM4e.

In early trading on October 27, South Korea's KOSPI rose more than 4%, after the Philadelphia Semiconductor Index gained about 2.8% overnight. Anthropic's official IPO filing in the U.S. market and Meta's equity financing plan also improved market expectations for the sources of AI construction funding. The core factor that will determine whether the large-scale counterattack in the U.S. semiconductor sector on Monday can continue will be the formation of persistently stronger procurement expectations after financing is secured, which is expected to ultimately translate into earnings and cash flow for memory manufacturers.

For more major Hong Kong stock news, download the Zhitong Finance app. For more Hong Kong and overseas wealth management news, please visit www.zhitongcaijing.com (search for "Zhitong Finance"). To join the Zhitong Hong Kong stock investment group, please add the Zhitong customer service WeChat (ztcjkf).