Two Memory Stocks Pull the KOSPI Down 6.6%

South Korea’s KOSPI was down 6.64% at 10:46 a.m. on Oct. 13 as Samsung Electronics and SK hynix led a memory-chip slide. The Korea Exchange halted program selling at 9:47 a.m. after KOSPI 200 futures dropped 5.08%.

ChangXin Memory Technologies, China’s largest DRAM maker, had told a Shanghai conference on Oct. 10 it would bring a denser server memory module by year-end. Samsung had just filed early third-quarter operating profit of 107.4 trillion won ($80.2 billion). The index still sold off as if two stocks were the whole market, because in Seoul they nearly are.

Two Chipmakers Still Make Up Half the KOSPI

On June 19, Samsung Electronics and SK hynix together accounted for 54.76% of KOSPI market value, according to Korea Exchange figures carried that day. Samsung’s slice was 28.34%. SK hynix’s was 26.42%. Their combined value was 4,162.749 trillion won, against a KOSPI cap of 7,601.1748 trillion won.

JUNE 19 MARKET WEIGHT

That weight first cleared 50% at the end of May. By then SK hynix had crossed 2,000 trillion won in value, the second Korean name after Samsung to do so. A national index built like that does not need a new factory to move 6%. It needs a headline that hits both names at once.

Prices have already come in from those June prints. Samsung common closed Oct. 12 at 250,500 won, down 4.39%. SK hynix closed at 1,605,000 won, down 4.52%. GlobalData put Samsung at $1.27 trillion at the end of September, up 225.6% year on year, and SK hynix at $956.3 billion, up 432.8%. Samsung’s shares had already given back 25% from the June peak by early October. The weights move with the prices. The two-stock structure does not.

What CXMT Promised for Year-End

CXMT President Cao Kanyu spoke at the Integrated Circuits and Chiplet Conference in Shanghai on Oct. 10. He said the firm had made “key progress” on a 4F² DRAM cell and planned a next-generation DDR5 registered module for servers by the end of 2026. Mainstream DRAM still uses a 6F² layout, in which each cell takes six times the square of the process’s minimum line width. A 4F² cell takes four.

If the node stays the same, that cut is about one-third of the cell’s area, and CXMT describes a 30% to 50% rise in cells per chip. The design uses a vertical surrounding-gate transistor, which the company says trims leakage and short-channel effects. Cao said CXMT had more than 300 published patent families on the 4F² work, a portfolio he called industry-leading. He also said the company would use the layout to get over engineering barriers in server memory, a market it has treated as a follow-on to consumer chips.

THE 4F² CLAIM, ITEM BY ITEM

• The cell: 4F² versus today’s 6F², about a one-third smaller footprint at a given line width.
• The switch: a vertical surrounding-gate transistor, plus talk of hybrid bonding and split array and logic processing.
• The SKU: a DDR5 RDIMM aimed at servers, scheduled for year-end 2026.
• The gaps: no public yield, no module capacity, and Korean coverage that the first parts may only partly use 4F².

SK hynix put a 4F² roadmap out in 2025. Samsung showed a 16Gb 4F² test chip with wafer-to-wafer hybrid bonding at ISSCC 2026 and is not expected to sell the structure before 2028. An announced date is not a qualified, high-volume line. Server modules still have to clear reliability, ECC behavior, power, and platform tests that CXMT has not published.

The Product on Offer Is DDR5, Not HBM4

The year-end SKU is working memory for ordinary servers. High-bandwidth memory, the stacked DRAM that sits beside AI accelerators, is a different product with a different yield problem. CXMT had already started mass production of its fifth-generation DRAM platform and of LPDDR6 in September, which is how it got to this conversation. TrendForce ranked it fourth in the second quarter with 9.5% of global DRAM revenue, up from 7.6% in the first quarter, behind Samsung at 39.4%, SK hynix at 24.9%, and Micron Technology at 23.3%.

DRAM REVENUE SHARE, 2Q 2026

HBM is where the three incumbents still look furthest ahead. A September account said CXMT had begun small-quantity HBM3E output and was testing it with Alibaba’s T-Head unit and Cambricon, with a possible 2027 ramp. CXMT, T-Head, and Cambricon did not confirm that. People familiar with the firm put it three to five years behind on HBM, with low yields and no public customer qualification. Neil Shah, vice president of research at Counterpoint, called CXMT’s HBM “unproven at scale.” SK hynix began mass-producing HBM3E in 2024. By 2026 both Korean makers were in HBM4 production and had sent 12-layer HBM4E samples, with pin speeds up to 16 gigabits per second.

That split is why selling SK hynix harder than Samsung on this headline gets the product mix backwards. More of SK hynix’s growth sits in HBM. The module CXMT put on the calendar competes with commodity and server DDR5, the pool Samsung has been riding on price. Mark Calderhead, an innovation-center manager at ASML, still called even the HBM3E step “a major step forward for China’s domestic chip supply chain,” which is the fear the tape is pricing: once the process is in hand, China has scaled batteries and solar faster than rivals.

When China figures out how to manufacture a product, their ability to scale is often greater than the rest of the world combined. We saw this with lithium batteries and solar, for example. Maybe RAM will be different, but RAM was a low margin commodity only ~3 years ago. LP5 and LP6 are designed to a common standard that China seems to be solving fast. Newer HBM architectures and things like an SRAM layer below logic and 3D stacking RAM on top of logic while finding a way to keep the logic cooled will be harder to solve.

Elon Musk, on X, Oct. 12, 2026

When China figures out how to manufacture a product, their ability to scale is often greater than the rest of the world combined. We saw this with lithium batteries and solar, for example.

SpaceX is a rare counter-example, where its mass to orbit is roughly an order of…

— Elon Musk (@elonmusk) October 12, 2026

The memory selloff did not stay inside DRAM. SanDisk, a NAND name, dropped 6.0% in New York on Oct. 12. The Philadelphia Semiconductor Index fell 2.67%. Funds treated a China density claim as a sector event, then let the KOSPI’s two-stock weight do the rest of the damage.

A Record Quarter Could Not Hold the Tape

Samsung’s Oct. 8 filing put consolidated operating profit at 107.4 trillion won, up 782.5% from a year earlier, on revenue of 195 trillion won, up 126.6%. The operating margin was 55.1%. It was the first time a Korean company had cleared 100 trillion won in quarterly operating profit, and the print beat a Yonhap Infomax consensus of 105.7533 trillion won by 1.6%.

SAMSUNG’S EARLY THIRD QUARTER

• Operating profit: 107.4 trillion won ($80.2 billion), up 782.5% year on year.
• Revenue: 195 trillion won, up 126.6%.
• Operating margin: 55.1%, a company high.
• Monday follow-through: the common shares still closed down 4.39% at 250,500 won.

The profit did not buy a bounce. Foreign ownership of Samsung common had already fallen to 46.38% as of Oct. 8, the lowest since January 2008, with SK hynix at 49.59%. Lee Kyoung-min, an analyst at Daishin Securities, said foreign selling and weak supply and demand were limiting any rebound even with solid earnings. On Oct. 13 morning, Samsung was down more than 6% and SK hynix more than 8% as the KOSPI printed 5,959.08, the first trade under 6,000 in 47 sessions, since July 31.

Foreigners Had Been Selling for Weeks

CXMT spoke on Oct. 10. The KOSPI’s slide was already in motion. From Oct. 6 through the Oct. 12 close the index lost 619.50 points, or 8.85%, after it had been up through 7,000. Monday, Oct. 12, was the first full session after the Hangul Day break, and foreigners sold 2.02 trillion won ($1.49 billion) on the main board, with institutions adding 110.7 billion won of selling. Individuals bought 1.47 trillion won and could not hold the close.

THE EIGHT-DAY TAPE

1. Oct. 8: Samsung files early third-quarter operating profit of 107.4 trillion won. Foreigners still sell Korean shares that day.
2. Oct. 10: Cao Kanyu presents 4F² progress and a year-end DDR5 RDIMM plan in Shanghai.
3. Oct. 12: KOSPI closes at 6,384.24, down 3.65%. Foreigners sell 1.0431 trillion won of SK hynix and 533.1 billion won of Samsung.
4. Oct. 13, 9:47 a.m.: a sell-side sidecar, the first in 55 sessions and the 26th this year, pauses program sells after KOSPI 200 futures hit 955.02, down 5.08% from 1,006.22.
5. Oct. 13, 10:46 a.m.: KOSPI trades at 5,959.08, down 6.64%, with Samsung off more than 6% and SK hynix off more than 8%.

The Financial Supervisory Service’s running tally had foreign net selling above 227.6 trillion won this year as of Monday, against 9 trillion won in all of 2025. In September alone foreigners sold 21.51 trillion won of KOSPI shares, and the two chipmakers were 16.47 trillion won of that, or 76.6%. U.S. 10-year yields stood above 5.2% on Monday, which is the other bid against a five-times-earnings memory trade. CXMT gave that bid a name.

Micron Already Gave Back Part of Its Summer

Micron, the U.S. leg of the same three-maker DRAM group, closed Monday at $989.04, down 3.88%, with a market value of $1.12 trillion. After-hours trade printed $975.08. The stock is up 246.7% year to date from $285.25 at the end of 2025, and it still sits well below the $1,255 close on June 25. It first cleared $1 trillion on May 26, when shares last traded at $886.6. Samsung and SK hynix had joined that club by early summer, with SK hynix’s June 19 value at 2,008.3953 trillion won.

The fundamental print behind that rerating was as large as Samsung’s. For the quarter ended Sept. 3, Micron’s revenue was $54.2 billion, up 379%. Full-year revenue was $133.2 billion, with net income of $85.0 billion. Chairman and CEO Sanjay Mehrotra said on the Sept. 30 call that the memory shortage would last through 2028. Those numbers are why a 4F² slide looks, on a screen, like a sector event in Boise as well as in Seoul. They are also why a year-end DDR5 module, with no yield card, is a thin reason to mark HBM supply as solved.

CXMT listed in Shanghai in July and raised 57.9 billion yuan ($8.6 billion) for process work, research, and added output. That capital is real. So is a 9.5% DRAM-revenue share. The module Cao put on the calendar is still a plan, and the KOSPI’s two names are still the index.

Disclaimer: This article is news reporting and analysis of index moves, company filings, and product plans, and it is for information only. It is not investment advice, a recommendation to buy or sell any share, or a forecast of DRAM or HBM prices. Readers should consult a licensed financial adviser or broker before acting on any figure, valuation, or product timeline mentioned here. Index levels, share prices, market values, and shipment plans are those given by the named companies, exchanges, and data firms on the dates in the piece and can change.