share_log

Plunge in South Korea's stock market fails to mask the strong momentum in memory chips! South Korea, the 'canary in the global economic coal mine,' is expected to see exports surge by 59% year-on-year in July.

Zhitong Finance ·  Jul 30 19:31

Economists stated that exports in July are expected to have risen 59.0% year-on-year (compared with 70.7% in June), and fundamentals in the semiconductor industry remain robust. Despite record profits reported by memory chip manufacturers, stock markets still declined sharply.

Zhitong Finance APP learned that a survey of economists released on Thursday indicated that South Korea's exports—often dubbed the 'canary in the global economic coalmine'—are likely to sustain robust growth in July. Although the pace of expansion is expected to moderate from the near-record high seen earlier, which marked the fastest export growth in nearly half a century, July could still deliver the second-strongest year-on-year export increase within the current cycle of sustained memory chip exports driven by surging global demand for AI computing power.

Memory chips are crucial to South Korea's exports and, by extension, its economy. South Korea is home to the world’s two largest memory chip manufacturers— $SK hynix (SKHY.US)$ and $Samsung Electronics (005930.KR)$ —including SK hynix, the global leader in high-bandwidth memory (HBM), which has become NVIDIA’s primary HBM supplier in recent years. Samsung, South Korea’s other memory giant, is the world’s largest supplier of both DRAM and NAND memory chips and has recently also become an HBM supplier to NVIDIA—specifically serving as the HBM memory system provider for NVIDIA’s current flagship AI computing platforms, the GB200 and GB300 series.

As breakthrough AI applications such as AI agents permeate industries worldwide, they are generating massive 'AI inference-end computing demand,' signaling vast future potential in infrastructure sectors including AI chips, HBM memory systems, enterprise-grade SSDs, and high-performance networking and power equipment. Additionally, the rise of on-device AI is poised to drive consumer electronics-grade DRAM and NAND memory demand onto a new growth trajectory.

South Korea’s July exports projected to surge by 59%, with stock market plunge diverging sharply from trade boom

According to the median forecast from a survey of 15 South Korean economists, South Korea—the fourth-largest economy in Asia and a key barometer of global trade—is expected to see its total exports in July jump by 59.0% year-on-year.

This growth rate would be slightly below June’s near-frenzied surge of 70.7%—the strongest export growth since 1978—but July’s export growth is still projected to mark the second-highest year-on-year increase since the current cycle of consecutive export gains began in June 2025.

Analysts noted that rising memory chip prices, fueled by surging global investment and construction of AI data centers, continue to drive substantial export growth for Samsung Electronics and SK hynix.

Ha Keon-hyeong, an economist at Shinhan Securities, stated: 'A significant portion of this growth reflects higher semiconductor trade prices. The increase in shipment volumes, by comparison, has been considerably more modest.'

Data compiled by South Korea’s statistical agency showed that in the first 20 days of July, the country’s exports surged 52.3% year-on-year, with semiconductor exports—led by memory chips—soaring by approximately 180.6%.

Park Sang-hyun, an economist at iM Securities, remarked: 'Given that semiconductor demand and export strength remain intact, the underlying momentum in South Korea’s exports is unlikely to change significantly.' He added that part of the slowdown stemmed from fewer working days in July compared to the same month last year.

The release of these survey findings coincides with a sharp sell-off in South Korea's stock market. The benchmark Korea Composite Stock Price Index (KOSPI) has declined by more than 30% this month, driven by market concerns over extreme deleveraging, unwinding of crowded positions, and the delayed realization of optimistic returns on AI-related capital expenditures.

Since the beginning of this year, South Korea’s stock market has repeatedly experienced sharp, circuit-breaker-triggering plunges. On July 28, the KOSPI index plummeted by 10.84%, and on July 29 it plunged as much as 12.6% intraday, triggering a 20-minute market-wide trading halt before closing down 5.98%. These two consecutive days of steep declines have pushed the index down roughly 40% from its June peak, wiping out more than USD 2 trillion in market value. Meanwhile, the Philadelphia Semiconductor Index had already entered a technical bear market by July 17, having fallen 20% from its record high on June 22, and by around July 29 its maximum drawdown neared 30%. This is no longer merely a correction confined to South Korea—it reflects a cross-market liquidity shock driven by deleveraging in Korean leveraged ETFs, global quantitative momentum unwinds, crowded semiconductor positions, and reassessments of AI valuations.

The 'funeral wreaths' placed before the National Assembly have escalated this financial deleveraging event into a political and regulatory crisis: Samsung Electronics and SK Hynix together account for nearly half of KOSPI’s market capitalization, while retail investors have heavily concentrated their bets through daily 2x single-stock leveraged ETFs. Declines in the underlying stocks forced these funds to reduce derivative exposures, creating a feedback loop of 'falling prices → ETF rebalancing sales → margin calls → further price declines.'

In response, the South Korean government plans to cap individual investments in such leveraged products at 20% of total investable assets, increase transaction costs and simulated trading requirements, and raise the minimum cash threshold to KRW 30 million effective July 31. Previously, authorities had already suspended new product launches and advertising for these instruments and begun establishing the legal basis for emergency market stabilization measures. While these actions do not equate to an official announcement of deploying the 'stabilization fund,' they signify a policy shift—from investor education toward directly curbing incremental leverage and, if necessary, providing market backstop tools.

SK hynix reported record second-quarter profits on Wednesday, but the results fell short of investors’ exceptionally high expectations, further exacerbating the market downturn.

Samsung Electronics, despite announcing on Thursday that its semiconductor business posted a record profit surge of over 250-fold in the second quarter, projected sustained strong demand for AI chips and memory chips this year alongside continued supply shortages. Nevertheless, South Korea’s stock market closed lower on Thursday.

The survey of economists also indicated that they expect South Korea’s July imports to rise 26.6% year-on-year, down from the 30.0% increase recorded in June. The median forecast for the July trade surplus stands at USD 29.59 billion, suggesting it will likely fall short of June’s USD 36.09 billion surplus.

South Korea is scheduled to release its closely watched July export data on Saturday, August 1, at 9:00 a.m. local time (00:00 GMT).

South Korea—the quintessential beneficiary of the memory chip supercycle

The underlying driver behind South Korea’s surging chip exports is not simply that 'more GPUs are being sold,' but rather that AI computing architectures are shifting from compute-centric designs toward coordinated expansion across compute, memory bandwidth, capacity, and interconnects. Large model training requires repeated movement of model weights, activations, gradients, and optimizer states. As the industry transitions into the era of inference and intelligent agents, long-context processing, concurrent users, and continuously operating agents rapidly expand KV Cache requirements.

Tensor Core compute speeds are increasing faster than the data delivery rate from conventional external memory, causing system bottlenecks to shift increasingly toward 'whether data can be delivered to the compute cores in time' rather than raw peak compute performance. High Bandwidth Memory (HBM), by vertically stacking multiple layers of DRAM, expanding I/O channels, and integrating closely with GPU packaging, achieves per-watt bandwidth levels unattainable by traditional memory. Consequently, HBM has evolved from a peripheral GPU component into a core determinant of actual AI accelerator utilization.

This demand exhibits a clear dual multiplier effect: 'increased content per server' multiplied by 'rising server deployment volumes.' NVIDIA’s H100 is equipped with approximately 80GB of HBM and offers over 3TB/s of bandwidth; the Blackwell Ultra supports up to 288GB of HBM3E with 8TB/s bandwidth; and the upcoming Rubin platform can accommodate 288GB of HBM4, pushing peak bandwidth to 22TB/s. In other words, each new AI platform generation not only increases GPU deployment counts but also continuously raises the required HBM capacity, bandwidth, stack layers, and packaging complexity per accelerator. SK hynix’s HBM4, featuring 2,048 I/O channels, delivers roughly 2.54 times the bandwidth of its predecessor and improves energy efficiency by over 40%, specifically addressing the growing 'memory wall' between model scale and data movement speed.

More importantly, HBM’s growth is spilling over to boost demand for commodity DRAM, server DDR5, low-power server memory, and enterprise SSDs. Producing HBM requires multiple high-specification DRAM dies, significant wafer area, through-silicon vias (TSVs), and advanced packaging. As manufacturers prioritize capacity allocation toward higher-margin HBM, effective supply of conventional DRAM naturally tightens. Meanwhile, the CPU side of AI servers still demands large-capacity DDR5 or SOCAMM modules, while inference workloads, vector databases, and KV Cache requirements continue to drive enterprise SSD demand. SK hynix explicitly noted that HBM investments are improving the supply-demand balance for commodity DRAM, and demand for server DDR5 and enterprise SSDs from AI data centers is simultaneously emerging as a new growth pillar.

Based on Samsung Electronics' closing price of approximately KRW 207,000 and SK Hynix's closing price of KRW 1,322,000 on July 30, 2026, Nomura’s target price of KRW 670,000 for Samsung Electronics implies a potential upside of 225% over the next 12 months, while its target price of KRW 4,700,000 for SK Hynix suggests a potential gain of approximately 255.5%.

Nomura’s analyst team stated that Samsung and SK Hynix should no longer be viewed merely as traditional memory stocks dependent on PC and smartphone cycles, but rather redefined as structural growth assets underpinning AI infrastructure. This assessment rests on three key points: (1) AI training, inference, and data center expansion are driving memory demand at a pace consistently outstripping supply growth; (2) high-bandwidth memory (HBM) and general-purpose memory have entered a synchronized supercycle; and (3) both companies’ forward price-to-earnings ratios of approximately 6x have been significantly lower than Taiwan Semiconductor’s ratio of around 20x, failing to adequately reflect their improving earnings sustainability and return on equity (ROE). Consequently, Nomura has substantially raised its target price for Samsung from KRW 590,000 to KRW 670,000 and for SK Hynix from KRW 4,000,000 to KRW 4,700,000.

Editor/Deng

The translation is provided by third-party software.


The above content is for informational or educational purposes only and does not constitute any investment advice related to EleBank. Although we strive to ensure the truthfulness, accuracy, and originality of all such content, we cannot guarantee it.