Memory's Competitive Edge Has Weakened. What About Equipment Makers?

The KOSPI fell for seven consecutive trading days, from 7,003.74 on October 2 to 5,554.22 on October 15 (−20.7%). On 10/15, it was 0.7% below the July 30 bottom (closing level 5,593.56). The KOSDAQ closed at 760.82, 38.0% below its April 27 high of 1,226.18.

Today (10/16) The KOSPI closed at 5,668.05 (+2.05%) and the KOSDAQ at 794.78 (+4.46%). The WFE basket rose +6.1%, while the two memory stocks rose +1.7%.

How much of the previous day's fall was recovered? Over the two days from the 10/14 close through today, cumulative returns were −5.67% for the KOSPI and −3.53% for the KOSDAQ. They recovered 25.0% and 53.9% of their 10/15 declines, respectively. A 100% recovery means regaining the 10/14 close, rather than simply subtracting today's gain from yesterday's percentage loss.

What overlapped? It cannot be reduced to one trigger (#29). Here is the sequence.

• 10/12 (local time) CXMT's president said at a Shanghai conference that the company would launch 4F² server DDR5 by end-2026 (SCMP report). The market expects Samsung's 4F² mass production around 2028.
• 10/13 night (US) Equipment stocks fell sharply. Profit-taking after year-to-date gains of AMAT +80%, LRCX +74% and KLAC +51%, along with doubts over fab investment, were cited.
• 10/14 night (US) Concerns grew about inflation, rates and AI data-center debt (CoreWeave; #28 and #29).
• 10/15 Circuit breakers in both markets. TSMC announced record earnings and a new outlook. Strong current results do not automatically confirm next year's order volume.

Who sold and absorbed shares through 10/15? From 10/1 to 10/15 (9 trading days), foreign investors sold KRW 18.34 trillion on the KOSPI. Other corporations, buying KRW 1.6 trillion a day throughout September, fell to KRW 0.64 trillion on 10/15. This category includes transactions beyond share buybacks; it does not identify today's buyers. Debt is also heavy. KOSPI margin loan balance for 10/15 was KRW 25.53 trillion, or 0.557% of market capitalization that day. Actual forced selling of unpaid purchases was KRW 80.5 billion, a different statistic from liquidation of all margin loans. At its 10/15 emergency meeting, the Financial Services Commission urged financial firms to restrain credit provision, but announced no specific measures such as a market-stabilization fund or short-selling restrictions.

From 10/8 to 10/15, memory (−22.4%) and equipment (−23.8%) had similar declines. Before adding today's moves, we separate the down days within that period.

Today's memory/equipment comparison. The 10 WFE stocks rose +6.1% in a day, versus +1.7% for the two memory stocks. Since the 10/14 close, returns were −9.1% and −8.7%, respectively. Since the 10/12 close, they were −14.9% and −17.4%. Basket membership is unchanged, and returns place equal amounts in each stock at each period's start. A larger daily gain is different from regaining the price before the fall.

The equipment rebound extended beyond a few stocks. All 10 WFE stocks in this basket rose. Newspim's 16:24 closing report, citing KB Securities, said more than 80% of stocks rose in both markets. The KOSDAQ triggered a buy sidecar (program-trading halt) at 12:55:23 (Yonhap, citing KRX). It points in the opposite direction from the previous day's sell-side measures.

But intraday gains did not hold fully to the close. Samsung Electronics went from an intraday high of +5.95% versus the previous close to +2.62% at the close. SK hynix went from +4.99% to +0.79%. This compares highs and closes in KRX daily data. It does not mean the stocks peaked simultaneously, or fell because a particular investor sold.

• Act 1 (9/23–10/8). Equipment rose and memory fell. Their returns diverged. Returns alone cannot tell us where money actually moved from and to.
• Act 2 (10/13). The CXMT announcement day. Memory fell −8.7%, while WFE rose +0.6%. PSK rose +6.5% and Jusung Engineering +3.6%. Korean equipment held up even as US (−2.5%) and Japanese (−5.4%) equipment fell.
• Act 3 (10/14–15). Equipment followed lower. Its −6.9% on 10/14 came after US equipment's −4.2% the previous night. Its −14.3% on 10/15 coincided with Japanese equipment's −11.7%.

Reading (interpretation). CXMT reports and memory's sell-off appeared on the same day. Korean equipment's large falls overlapped global equipment weakness and Korea's broad market sell-off. Changes in global investment expectations, domestic investor flows and company-specific circumstances all need checking. From 9/23 to 10/15, equipment fell −14.4% and memory −29.3%. That comparison ends yesterday; cumulative returns including today's recovery appear separately at the last point of Chart 2. Today's rebound also does not establish any single factor's contribution.

CXMT competes with memory makers and buys from equipment makers. Stronger memory competition and more equipment orders can occur together. Which Korean equipment makers receive orders is a separate question.

But Korean equipment makers' share is limited. Reports cited by TrendForce put Chinese-equipment adoption at 35% in 2025, and above 40% for deposition and etching. A 50% domestic-equipment requirement for new capacity was also reported, but this research did not confirm the publicly available original regulation. Deposition and etching are central to Korean WFE makers. Total investment therefore cannot be translated directly into their order opportunities. Process-specific constraints and supply-contract counterparties need checking. Foreign-equipment demand lasts longer in areas with slower localization, such as lithography, inspection/metrology and ion implantation.

China exposure differs by company. Jusung Engineering's available 2025 disclosures showed China at 64.28% of revenue; two unnamed major customers together accounted for 84.36%. A China location alone does not identify the customer as CXMT. The disclosure ledger separately preserves region, customer anonymity and contract counterparty.

Korean equipment's 10/14–15 sell-off overlapped global equipment weakness. This historical comparison cannot simply be carried over as the cause of the 10/16 rebound.

• US (10/13 night). AMAT fell −7% intraday, while LRCX and KLAC fell −4%. They had risen +80%, +74% and +51% year to date, respectively. Falling several times more than the semiconductor-index ETF prompted doubts about customers' fab budgets (translated). The article cited profit-taking and investment concerns as the backdrop.
• Taiwan (10/15). TSMC reported record Q3 revenue of USD 46.7 billion and Q4 guidance of USD 52.3–54.3 billion. Its 2026 capex plan is USD 60–64 billion. This year's budget and next year's orders reaching Korean equipment makers require separate checks. That same day, US equipment stocks were the weakest part of semiconductors.
• Japan (10/15). Advantest −15.0%, Lasertec −13.6%, SCREEN −10.6%, Tokyo Electron −7.8%.

Countries moved in different directions today too. In the US regular session on 10/15 before Korea opened, the 3 equipment stocks averaged −3.15%. On 10/16, Korean equipment rose +6.07%, and the 4 Japanese equipment stocks rose +2.50%. All are equal-weighted daily returns with unchanged membership. Korea's rebound cannot be explained by gains in US equipment the previous day.

Reading (interpretation). Splitting the dates shows Korean equipment's sell-off overlapping global equipment weakness. Korea's market sell-off and investment-trust selling were also observed (Section 07). This comparison cannot separate each contribution to the decline.

Separate customers' investment plans from the last confirmed order backlog. Start with customer investment, the basis for equipment revenue.

• Signals from equipment makers. ASML reported Q3 revenue of EUR 11.7 billion and Q4 guidance of EUR 14.2–15.2 billion, saying supply would not meet all demand in 2027–2028. It plans to increase 2027 production capacity for low-NA EUV and DUV immersion equipment by 30% versus 2026. That is not the expansion rate for all equipment types.
• Korean figures. A 10/6 Newspim report cited September semiconductor-equipment imports of USD 3.44 billion (+50.8%) and semiconductor exports of USD 60.3 billion. The official monthly import ledger was not obtained in this collection, so these remain attributed press figures. Import growth cannot be read directly as order growth for a particular Korean equipment maker.

Backlogs were checked against original 6/30 half-year report figures. Jusung's KRW 233.635 billion is the total across all businesses; semiconductor manufacturing equipment accounts for KRW 214.358 billion. We do not sum backlogs with different business scopes. Revenue and operating-profit E figures are FnGuide estimates cited in the 10/6 report, not consensus estimates refreshed today. Today's rebound is a price change; the end-June backlog is the amount remaining from orders received earlier. New orders in between require separate checks of supply-contract disclosures and the next quarterly report.

Equipment earnings are recorded after the order.

Second-half orders can become next year's revenue, but recognition timing differs by company and contract. This lag must be considered when comparing expectations in share prices with actual revenue. Jusung Engineering's operating-profit consensus estimates are KRW 40.8 billion for 2026 and KRW 140.1 billion for 2027.

Prices move before that. Expectations remain ahead in some cases even after the sell-off (10/15 close).

Comparing the 10/15 close with the preceding three-month high gives average declines of −32.2% for WFE, −30.3% for memory, −21.5% for US equipment, −23.0% for Japanese equipment and −15.1% for Micron. Many equipment stocks peaked on 10/1–10/6, at the end of Act 1.

Reading (interpretation). Even strong earnings can make release day a day when expectations are spent, if results fail to exceed what prices already reflect. The multiples above simply divide market capitalization by expected operating profit. They are neither P/E nor EV/operating profit, and the comparison years differ. They cannot alone rank companies by overvaluation.

Through 10/15, investment trusts sold KOSDAQ stocks every day from 10/7 to 10/15 (6 trading days), totaling KRW 769.6 billion. Selling centered on equipment and components, while foreign investors bought the same stocks. On 10/15, 13 of foreigners' top 20 purchases overlapped institutions' top 20 sales.

Chart 5 is the flow on 10/15, not the money buying today's rebound. Closing reports for 10/16 put foreigners' net buying of KOSPI cash equities at about KRW 343 billion (Newsis 16:26 revised version). This is a one-day turn; it does not show tag-level flows or intraday purchase times.

Selling also occurred through ETFs. On 10/15, investment trusts had net sales of KRW 184.7 billion in KODEX KOSDAQ 150 and KRW 344.6 billion in KODEX 200. Retail investors and securities firms (liquidity providers) took the other side. Pension funds and others bought about KRW 77 billion across 4 index ETFs that day. That was much more than the previous day's roughly KRW 3 billion, but small relative to foreign selling (KRW 3.6 trillion on the KOSPI).

Reading (interpretation). These data cannot distinguish whether investment-trust selling came from redemptions or risk management. One day also cannot establish whether foreigners bought because they examined equipment makers' books. But buyers and sellers met in exactly the same tags.

ETF trading checked separately today. KRX's 10/16 figures show KODEX 200 foreign investors at +KRW 186.182 billion and financial investment firms at −KRW 196.885 billion. KODEX KOSDAQ 150 went the other way: institutions in total +KRW 18.600 billion; foreign investors −KRW 14.472 billion. We cannot identify all financial investment trading as liquidity provision, or call the institutional total policy funds or pension buying.

Listed units rose from 10/15 to 10/16 by 3.75 million in KODEX 200 and 4.15 million in KODEX KOSDAQ 150. The link between creation/redemption request dates and changes in listed units was not confirmed, so these increases were not converted into today's cash inflows or actual shares bought today.

Will won strength and foreign buying continue together? The saved Naver exchange-rate data show KRW 1,331.8 per dollar on 10/16. We distinguish this from ECOS's Seoul 15:30 close. The last ECOS observation available by 18:00 was 10/15, so it is not labeled today's official close. Won strength, foreign cash-equity net buying and a broad rebound offer clues about whether leader concentration is easing. We do not add flows with different periods or cash/futures notionals to establish net exposure or causation.

We will check customers' earnings releases and equipment makers' quarterly reports over the next 6 weeks.

What we still do not know.

• Whether 2027 investment reaches equipment or buildings first. Micron said construction investment was growing faster than equipment.
• Korean equipment's share of CXMT orders: only the room left by localization.
• Customer concentration. Jusung's two unnamed major customers accounted for 84.36% of 2025 revenue. This ratio cannot be applied to other companies, but customer schedules and revenue concentration should be examined together.
• How much expectation is already priced in (Section 06).

Future comparisons. Using the 10/16 close as the base, compare the 10 WFE stocks, KOSDAQ and two memory stocks over the same periods. Record supply contracts, customers' 2027 investment plans, investor flows and backlogs too. We do not call success or failure using a threshold that has not been set.

Memory and equipment need separate comparisons of their down days and recovery sizes. A sharp stock decline alone does not establish lost competitiveness; a rebound does not confirm a technological gap or the next order. Equipment recovered more than memory today, and the rebound spread broadly across the market. To assess whether this is sustained rotation, we need the next trading day's foreign cash-equity flows, ETF and detailed institutional flows, and concentration changes. Customer investment plans, new supply contracts and backlogs are business records. We record price rebounds and order recoveries separately, checking whether they converge.