- Official figures confirm that chips and computers drove export growth. They accounted for 82% of October's increase; semiconductors alone generated 72% of growth in the first ten days of November.
- Taiwan is more concentrated on comparable product codes. Hanwha Securities used the same HS codes to measure AI-linked products: 45% of exports for Korea versus 72% for Taiwan. Taiwan also ships a greater share directly to the U.S. Neither measure isolates final AI demand.
- Korea is especially sensitive to memory prices. Memory supplied 70% of October's export increase. Prices drove roughly two-thirds of electronics export growth, though volume also rose 51%.
- Yet Korean stocks fell furthest in H2. From June 30 to Nov 16, KOSPI lost 39.8% versus 11.0% for Taiwan's TAIEX. After doubling in H1, KOSPI remains 21% above year-end 2025. Samsung Electronics and SK hynix together make up nearly half the index—more concentrated than exports.
- Watch NVIDIA for server demand and HBM procurement. Verify memory prices separately through suppliers and contract data. Cosmetics, biopharma and food exports are growing, but their combined October value was only 7% of chip exports.
Hover over dotted terms (tap on mobile) for definitions. Red dotted terms are central to the analysis.
NVIDIA reports Q3 earnings at 7 a.m. Wednesday, Nov 18, Korea time—its first since the Nov 12 U.S. announcement of the CEO's resignation and the CFO's appointment as interim CEO.1, 2
The question: if NVIDIA weakens, how much does Korea suffer? Semiconductors made up 49% of Korea's $115.28 billion in October exports.3 Hanwha Securities estimates that AI-linked products accounted for roughly 80% of Korea's export growth this year.4
Yet Taiwan is even more concentrated on like-for-like product codes. The bigger difference is what each economy sells: Taiwan ships servers; Korea ships memory.
Chips and computers drove
82% of export growth
AI-linked products drove growth, consistent with Hanwha's analysis. But these figures use different periods and categories, so they are not an exact replication.
Hanwha defines AI-linked goods as memory and other chips plus AI-server computing equipment and components (six-digit HS codes). The closest industry-ministry categories are semiconductors and computers.4 H1 exports rose 48.3%, with memory contributing 25.7 percentage points and AI infrastructure equipment 13.6 points (growth contribution). Together they made up about 80% of the gain. Here is the comparison with official ministry and customs data.
| Source | Scope · Period | Share of Increase | Share of Total Exports |
|---|---|---|---|
| Hanwha Securities | AI-linked productsH1 2026 | About 80%Memory 53% · AI infrastructure 28% | 45%2026 |
| Industry Ministry | Chips + computers (incl. SSDs)October | 82%Semiconductors alone: 73% | 54.0%28.0% a year earlier |
| Customs Service | SemiconductorsNov 1–10 | 72% | 49.7% |
Sources: Hanwha Securities, Ministry of Trade, Industry and Resources, Korea Customs Service, The Korea Herald3, 4, 5, 6 · Ministry and Customs contributions calculated by reconstructing last year's values from published growth rates
These are product-concentration measures, not direct AI end-demand estimates. Chips also include memory for smartphones and PCs.
Anatomy of the export gains
Korea was driven by memoryYear-over-year increase (US$100m) · Korea's Industry Ministry; Taiwan's Finance Ministry
Sources: Korea's Industry Ministry; Focus Taiwan (Taiwan Finance Ministry)3, 7
- Memory accounted for 70% of the gain. Memory rose from 18% of Korea's October exports last year to 44% this year. Of the US$55.77bn total year-on-year increase, US$39.25bn—or 70%—came from memory.3
- Other brokerages reach a similar conclusion. Eugene Investment & Securities estimates chips drove 73% of October's increase, down from 79% in September.8 Kiwoom Securities also sees semiconductors as the main growth engine.9
- Korea's concentration has risen faster. In Hanwha's basket, AI-linked products rose from 15.8% of Korea's exports in 2023 to 21.5% in 2024, 25.1% in 2025 and 45.0% this year. Taiwan was already at 66% in 2025.4
- Calendar effects matter. October had two more working days than a year earlier. Growth was 76.1% on a daily-average basis, not the 93.7% headline rate.3
Korea is less concentrated
than Taiwan
Across five measures, Taiwan (black squares) ranks higher than Korea (orange circles). Each measure uses its own denominator.
Korea runs at 50–80% of Taiwan's levelPercent · Compare within rows, not across them · October categories differ by country
Sources: Hanwha Securities, Korea's Industry Ministry, Focus Taiwan (Taiwan Finance Ministry)3, 4, 7
Taiwan's October ICT-plus-components category includes phones and telecom equipment, making it broader than Korea's chips-plus-computers basket. The first row, using matched HS codes, is the better comparison.
Contribution to economic growth
- The growth gap is even wider. In H1, Korea grew 3.8% and Taiwan 14.1%. IT manufacturing contributed 1.7 and 8.0 percentage points, respectively (Hanwha estimates).4
- Korea's IT contribution grew in Q3. The Bank of Korea attributed roughly 60% of Q3 year-on-year growth (YoY) to computers, electronics and optical equipment, including chips, up from the 40% range in Q2.10, 11 But headline Q3 growth was 3.4% in Korea and 14.36% in Taiwan.12, 13
- Do not compare these with the U.S. figure of 28%. Hanwha's 28% estimate for AI's contribution to U.S. growth measures AI investment net of imported equipment. It is not the IT-manufacturing share of GDP growth used for Korea and Taiwan.4
Routes to the U.S.
- Taiwan ships servers to the U.S. Servers and other processing units account for 13.3% of Taiwanese exports; other computing devices incorporating GPUs account for 21.2%. About 60% of each category goes to the U.S.4
- Korea ships memory to Asia. The destinations for Korea's AI-linked goods are China (27%), the U.S. (18%) and Hong Kong (15%). Computer parts and SSDs send more than half to the U.S., but make up only 8% and 4% of total exports. Servers and other computing devices account for less than 0.1%.4
- Shipping destination is not final demand. Memory shipped to China, Hong Kong or Taiwan may ultimately reach U.S. data centers after assembly. Korea's real U.S. exposure could exceed its direct-export share.
Same AI demand, different routes
| Measure | Korea | Taiwan |
|---|---|---|
| Main products | Memory-led semiconductor exports | Servers, ICT goods and electronic components |
| Direct destinations | Mostly China, Hong Kong and other Asian markets | Larger direct share to the U.S. |
| Key metrics | Memory prices, volumes and contracts | Server shipments, orders and assembly volumes |
| Exposure to final AI demand | Not directly measured here | Not directly measured here |
Taiwan's server export value includes imported chips and memory. So “less exposed” here refers to export mix and routes—not a direct measure of ultimate AI demand.
Korea's exposure lies
in memory prices
Prices account for more of the semiconductor export gain under every approach. That does not mean volumes were flat.
Export value equals price times volume. We separate the two effects three ways.
Sources: Hanwha Securities (1, 4), Yuanta Securities (2), Bank of Korea and MTN (3, 4)4, 14, 15, 16 · For (3), the price effect is estimated from value and volume growth factors. Depending on the allocation method, prices account for 50–76%. See Methods.
Why the measures differ
- Measured by weight: Unit value is export value divided by kilograms. A shift toward higher-capacity, pricier chips appears as a price increase. In total exports, heavy goods such as oil products and steel dominate tonnage, masking growth in lightweight chips.
- Measured by BOK indices: The volume index strips out price changes to estimate real exports. Product upgrades can count partly as volume. Thus customs weight can fall even as real volume rises, as in (4).
- What the evidence supports: Prices mattered more. But (3) covers electronics, not just memory, and unit value per kg in (1) is not a pure like-for-like price measure. Hanwha makes the same qualification.4 The precise memory-price contribution cannot be isolated.
Why price dependence matters
Memory prices swing more sharply than volumes. Rising prices amplify export growth; falling prices can reverse it just as quickly.
- Early November slowed from October. Average memory exports per working day in Nov 1–10 were up 215% YoY but down 7% from October. DRAM alone rose 6%; NAND and SSDs declined (Hana Securities, preliminary).17
- October may mark the peak growth rate. The comparison base is higher in November and December 2025. Eugene calculates that even unchanged monthly exports would yield just 89.7% growth in November.8 Hanwha expects semiconductor export growth to slow sharply from Q2 2027, given its memory-price outlook.14
- There is a counterargument. Kiwoom says more long-term supply agreements could sustain chip exports for now, while warning that soaring memory prices may hurt PC and smartphone demand.9
That is why NVIDIA news affects the two countries differently: Taiwan asks, “How many servers will ship?”; Korea asks, “Will memory prices hold?”.
For the memory-cycle debate, see Insights No.29 (Earnings Credibility) and No.34 (Memory Fab Timelines).
Yet Korean stocks fell
the furthest in H2
Korea's AI-linked export share is lower than Taiwan's, but its stock market has fallen most among the three since midyear.
KOSPI doubled—then gave back much of the gainClosing levels · Last trading day of 2025; Jun 30 and Nov 16, 2026
Sources: Korea Economic Daily, Kyodo News, Commercial Times, Yahoo Finance (closes), iM Securities18, 19, 20, 21, 22
Start with the full year. KOSPI gained 101% in H1, far more than TAIEX (+59%) or Nikkei (+39%). It then lost nearly 40%, but remains 21% above year-end 2025. TAIEX is up 42% YTD and Nikkei 19%. Taiwan leads on full-year returns; Korea has seen the biggest swings.
All three markets sold off on AI shocks. In the week after the U.S. midterms (Nov 2–6), as concern over data-center restrictions grew, TAIEX fell 12.2%, KOSPI 11.7% and Nikkei 9.3%.4 On Nov 16, amid AI caution and U.S. inflation worries, all three fell again: TAIEX 6.0%, KOSPI 3.6% and Nikkei 2.4%.21, 23 The wider divergence since June therefore needs more than an AI-shock explanation.
Why exports and stocks diverge
- Exports look back; stocks look forward. Export data measure past sales. Share prices discount expected earnings. Stocks can fall despite rising exports if investors expect price growth to peak.
- The stock index is even more concentrated. Samsung Electronics and SK hynix rose from 35% of KOSPI at the start of the year to around 56% near the June peak; they still accounted for more than 46% on Nov 6.24, 25 TSMC represented 41% of TAIEX at the end of August.26 So while Taiwan's exports are more AI-heavy, nearly half of KOSPI is concentrated in two companies whose profits hinge on memory prices.
Four possible drivers of the gap
| Candidate | Evidence | Further context |
|---|---|---|
| 1 · H1 rally | A bigger rally left more room for a pullback (iM Securities).22 | FIG.3 |
| 2 · Foreign selling | Net foreign outflows were far larger in Korea than in Taiwan; Japan saw net inflows this year (iM Securities).22 | NO.32 |
| 3 · Won appreciation | Since June 30, yen/won fell 12% and Taiwan dollar/won fell 17%.21 iM Securities says those currency moves coincided with changes in KOSPI's relative performance. It argues currencies may affect AI-sector profits more than price competitiveness.22 Converting KOSPI into Taiwan dollars reduces its decline to about 28%, narrowing the gap with TAIEX from roughly 29 to 17 percentage points. | NO.16 |
| 4 · Memory-price doubts | When export growth depends on prices, earnings expectations can overshoot on the way up and retreat sharply near a suspected peak. KOSPI's H1–H2 pattern is consistent with this—but does not prove causation. | NO.29 · NO.34 |
We cannot apportion the gap among these four factors. Even iM Securities says the link between won appreciation and KOSPI weakness remains hard to explain.22
Three non-AI growth engines—
still small beside chips
When chip risks rise, what else can support exports?
Cosmetics, instant noodles and pharmaceuticals have grown across markets outside the U.S., with both prices and volumes rising.
| Product | Export growth | Contribution from non-U.S. markets | Unit value · Weight contribution |
|---|---|---|---|
| Cosmetics | +31% | +23.1%pU.S.: +8.0pp | +8.2%p · +22.8%p |
| Instant noodles | +24.9% | +19.8%pU.S.: ~+5.1pp | +3.4%p · +21.5%p |
| Pharmaceuticals | +15.7% | +21.5%pU.S.: ~−5.8pp | +5.6%p · +10.1%p |
Jan–Sep · Hanwha Securities analysis of Korea Customs data4 · U.S. contributions for noodles and pharmaceuticals are approximated as total growth minus non-U.S. contribution
- Volumes drove the gains, unlike semiconductors. For all three, tonnage contributed more than unit values: more products were sold, not simply at higher prices.4
- Markets beyond the U.S. led growth. The U.S. accounts for less than 20% of noodle and pharmaceutical exports. Sales rose in about 90% of their non-U.S. destination markets.4 U.S. pharmaceutical exports contracted, subtracting roughly 5.8 percentage points from overall growth.
- Broader categories also grew in October. Cosmetics reached US$1.44bn (+57.5%), biopharma US$1.53bn (+30.2%) and agricultural, forestry and fishery products US$1.11bn (+11.6%).3 The last two categories are broader than pharmaceuticals and noodles. Kyobo Securities says Jan–Oct cosmetics exports hit US$10.3bn (+32%), above all of last year, and forecasts another 27% rise next year.27 DS Investment & Securities highlights noodles, snacks, sauces and health supplements as overseas growth segments.28
- Some non-AI sectors held up in equities. According to Eugene, construction, transport, healthcare, IT appliances and batteries, and KOSDAQ remain above their July 30 lows.29 These figures do not isolate cosmetics or food stocks.
equal just 7% of chip exportsOctober export value (US$100m) · Korea's Industry Ministry
Source: Korea's Industry Ministry3
These categories cannot offset a major semiconductor slowdown alone. Total non-chip exports reached US$58.84bn in October, more than chip exports, but include large, differently driven sectors such as cars and petroleum products.3 Hanwha sees consumer goods and biotech as alternatives if concerns about AI spending intensify.4 For country-level cosmetics trends, see Insight No.23.
NVIDIA earnings this week:
what Korea should watch
Insight No.21 tracks the AI earnings scoreboard. Here we follow how NVIDIA's numbers transmit to each country.
- Data-center revenue and next-quarter guidance
- Server-rack and GPU-board shipment plans → Taiwanese ODM exports
- U.S. customers' capex plans
- HBM Comments on HBM procurement (volume and contract tenor)
- Whether rising memory costs appear in margin commentary
- Any changes in sourcing or orders after the leadership transition
NVIDIA earnings offer an indirect read on memory demand—not proof of memory selling prices. Check suppliers' results and contract-price data. If NVIDIA does not disclose procurement prices, its call alone cannot establish a trend.
Next data points
- 11.18 06:20 · 07:00NVIDIA Q3 results and earnings callKorea time · Release ~1:20 p.m. U.S. Pacific Nov 17; call at 2 p.m.
- 11.19 04:00October FOMC minutesKorea time
- Around Nov 21Korea Customs: Nov 1–20 exportsWill the chip share shift from 49.7% in the first ten days? Is memory still below October's pace?
- 11.26Bank of Korea rate decisionPolicy rate decision
- 12.1Korea's November trade releaseWatch export value and daily averages, not just YoY growth
- Early DecemberTaiwan's November exportsFinance Ministry forecast: US$94.2bn–96.7bn (+47–51%)
- Mid-DecemberBOK November trade indicesCompare electronics volume and value indices. Can volume offset weaker prices?
Sources: NVIDIA, Federal Reserve Board, Business Watch, Focus Taiwan, Eugene Investment & Securities1, 7, 8, 30, 31
What could change the conclusion?
October's DRAM contract price (DDR4 8Gb) hit another record at US$27.50. NAND stopped rising after 21 months.32 As long as prices hold, price sensitivity supports exports rather than undermining them.
October exports excluding chips rose 34.3% YoY, but daily-average growth slowed from 37% in September to 22% in October.8 Faster non-chip growth would dilute the semiconductor share. Hanwha's chip-plus-SSD share already edged down from 55.4% in September to 53.8% in October.14
Stable exchange rates would weaken explanation (3) in Section 04. Hanwha notes USD/KRW is down 8.7% YoY and that since 2008, declines have often halted near 10%.33 History is no guarantee; the question is whether appreciation continues or stops.
- Most October export growth came from chips and computers, chiefly memory
- Taiwan's U.S.-bound export share reached a 36-year high in Jan–Oct
- Both prices and volumes drove October electronics export growth
- AI-linked export share and U.S.-bound share (Hanwha HS definition)
- IT share of H1 growth: Korea 46%, Taiwan 56%; Korea Q3 ~60% (BOK commentary)
- Reasons for the equity divergence (iM Securities)
- When memory prices will turn
- Which end-use AI markets Korea's memory reaches through intermediaries
- How NVIDIA's management transition affects sourcing and orders