How to Read Korea’s Semiconductor Concentration

Key Takeaways

• In the quarter ended June 30, 2026, Samsung Electronics reported KRW 89.5tn in operating profit. Its device solutions arm—the chip business—reported KRW 89.2 tn of that. The phone and network unit lost KRW 0.7tn in the same quarter. Almost the entire company is one division now.
• SK hynix reported KRW 79.3 tn in revenue and KRW 60.5 tn in operating profit for the same quarter, a 76% operating margin. Both companies flagged their figures as preliminary and unaudited.
• HBM — high-bandwidth memory, stacked DRAM (dynamic random-access memory) that sits beside an AI accelerator — is a small share of memory bits and a large share of memory factories. TrendForce puts HBM at roughly 22% of DRAM wafer input by end-2026 against about 9% of DRAM bit supply. That gap is the whole pricing story.
• On August 7, 2026, Samsung Electronics, its preferred shares, and SK hynix together were 48.95% of the market capitalization of the KOSPI, Korea’s main stock index, per Korea Exchange data. That figure counts three listed lines rather than two, and it had been 54.04% a week earlier. If you own broad Korean equity exposure, you’re holding a memory position whether you meant to or not.
• Semiconductors were US$41.01bn of Korea’s US$98.89bn of July 2026 exports. On our arithmetic from the ministry’s published growth rates, chips supplied roughly 69% of the year-on-year increase in total exports.
• The US Section 232 semiconductor duty, as written, doesn’t reach commodity DRAM or NAND flash. Its thresholds describe logic chips.

Contents

1. The quarter that doesn’t look like a quarter
2. Why HBM crowds out everything else
3. What it did to the index
4. What it did to the export account
5. What would change the picture
6. Four traps
7. FAQ

1. The quarter that doesn’t look like a quarter

Start with the number that does the most work, because it’s the one that reframes everything after it.

Samsung Electronics reported KRW 171.5tn in consolidated revenue and KRW 89.5tn in operating profit for the second quarter of 2026. That’s about US$121.3bn and US$63.3bn at 1,414.29 won per dollar, the rate for August 14, 2026. The Device Solutions Division, which houses memory, foundry, and system chips, contributed KRW 127.5 tn of the revenue and KRW 89.2 tn of the operating profit (Samsung Global Newsroom, July 30, 2026).

Divide those two profit figures and you get 99.67%.

The rest of the company was busy. On net, it earned approximately nothing. Samsung Display earned KRW 0.7tn. The MX and Networks business—the Galaxy phones, the thing most people outside Korea think Samsung is—posted an operating loss of KRW 0.7 trillion on KRW 33.2 trillion of revenue, blaming component costs. Televisions and appliances also ran a small loss.

SK hynix has no phone division to complicate the picture. It reported KRW 79.3187 tn of revenue and KRW 60.5426 tn of operating profit for the same three months (SK hynix Newsroom, July 29, 2026). That’s roughly US$56.1 billion and US$42.8 billion at the same rate, and a 76% operating margin. Revenue rose 257% against the year-earlier quarter, and operating profit rose 557%. Net income came in at KRW 93.9226tn, which is higher than revenue. That is not a typo, and the release does not explain it. Cash reached KRW 88tn against KRW 18.6tn of debt.

Two cautions belong right here, before any of this hardens into a thesis.

The first is that SK Hynix states plainly that these are preliminary figures. The quarterly review is not final, and the numbers may change in the audit. Companies write that sentence every quarter, and readers skip it every quarter. At this scale of profit, an audit adjustment is worth waiting for.

The second is arithmetic. A 557% increase is a comparison against a quarter when memory was cheap. It describes the slope of a price cycle rather than durable earning power. Memory has done this before, in both directions.

2. Why HBM crowds out everything else

The mechanism worth understanding isn’t “AI demand went up.” It’s how HBM consumes manufacturing capacity relative to what it delivers.

TrendForce’s figures make the point cleanly. HBM was around 18% of total DRAM wafer input at the end of 2025, and it’s projected at roughly 22% by end-2026 and about 30% by end-2027. Measured in bits shipped, HBM was about 8% of DRAM bit supply in 2025, with roughly 9% projected for 2026 and 13% for 2027 (TrendForce, June 2, 2026). The 2026 and 2027 figures are forecasts rather than results.

Hold those two rows next to each other. At the end of 2025, HBM took about 18% of wafer input and returned about 8% of bits. Stacking, testing, and yield losses eat the difference. So every wafer moved to HBM removes more conventional DRAM supply than the HBM bits replace. That tightens the ordinary DRAM market that phones and PCs and servers all buy from.

That’s why the price move showed up across memory rather than only in the AI-specific parts. TrendForce put first-quarter 2026 DRAM industry revenue at US$97bn, up 81% in a single quarter, with Samsung at US$37.32bn and SK hynix at US$27.98bn (TrendForce, June 1, 2026).

Here’s a ranking inversion that’s worth carrying around, because it catches people out. In HBM specifically, Counterpoint Research’s latest published quarter—the first quarter of 2026—has SK Hynix at 58%, Samsung at 21%, and Micron at 21% (Counterpoint Research). In DRAM overall, Samsung leads. The company that dominates the headline product is the smaller of the two in DRAM overall.

One more thing from that same Counterpoint data set tends to get skipped. China’s CXMT held about 8% of DRAM in the first quarter of 2026, against roughly 3% a year earlier. It’s nowhere in HBM. It’s no longer nowhere in DRAM.

3. What it did to the index

Korea Exchange figures reported on August 9, 2026, put total KOSPI market capitalization at KRW 5,159.9986 tn as of the August 7 close, or about US$3.65 tn. Samsung Electronics, Samsung Electronics preferred shares, and SK Hynix together came to KRW 2,525.7338 tn, about US$1.79 tn, or 48.95% (Financial News, August 9, 2026).

Read the basis before you quote the number. That 48.95% counts three listed lines: Samsung common, Samsung preferred, and SK Hynix. In English-language coverage you’ll more often see a two-company figure, which is a slightly different calculation of a similar idea. The two get mixed together constantly, and the difference is large enough to matter in a portfolio note.

Now read the volatility. The same three lines were 54.04% on July 31, 2026, and had peaked at 58.9% on June 25, 2026. A concentration measure that moves five percentage points in a week isn’t a structural fact you can cite for a year. It’s a snapshot with a date attached.

So treat the date as part of the number, and check it yourself before acting on it. The underlying data is published by Korea Exchange and updated daily. A figure from a news article printed months ago tells you about that day.

What survives the volatility is the shape. Whatever the exact reading, roughly half of Korea’s main equity index has been sitting in two companies whose earnings run on the same commodity price. Passive exposure to “Korea” is, to a first approximation, a leveraged memory position with a garnish of everything else.

That is a defensible thing to own deliberately. It is a poor thing to own by accident, and owning it by accident is very easy.

This is a separate question from the valuation gap covered in How to Read the Korea Discount as an Investor. It is separate again from how index providers classify the market, covered in Why MSCI Still Calls Korea an Emerging Market.

4. What it did to the export account

The same concentration shows up in the national accounts, where it is measured more carefully and revised less often.

Korea’s Ministry of Trade, Industry, and Resources reported July 2026 exports of US$98.89 billion, up 62.8% year on year, with semiconductor exports of US$41.01 billion, up 178.8%. Non-semiconductor exports rose 26%. Imports were US$68.56 bn, and the trade surplus was US$30.32 bn (MOTIR, August 3, 2026).

Chips were therefore about 41.5% of Korean exports by value that month.

The growth split is starker, and we work it here rather than assert it. Back out the year-earlier levels from the published growth rates, and July 2025 gives roughly US$60.7bn of total exports and roughly US$14.7bn of semiconductors. Total exports grew by about US$38.1bn. Semiconductors grew by about US$26.3bn. Chips supplied roughly 69% of the increase.

That’s our calculation from the ministry’s published levels and rates, not a ministry figure. Inferring base-year levels from rounded percentages carries about a point of error either way.

The honest summary is that non-chip Korea grew 26% year on year. That’s a good year by any normal standard, and it was still a rounding detail next to the memory line.

5. What would change the picture

Tariffs, but read the scope. The US Section 232 semiconductor duty of 25% was imposed by Proclamation 11002, signed January 14, 2026, with duties applying from January 15. It reaches logic integrated circuits under a narrow set of tariff lines that meet defined thresholds for processing performance and DRAM bandwidth. A companion tariff code exists specifically for goods in those lines that fail the thresholds and are therefore not dutiable (Morgan Lewis). Commodity DRAM and NAND don’t meet the stated thresholds. As of August 18, 2026, we found no public action extending the duty to them. Scope in this area has moved before, so verify against current US Customs and Border Protection guidance rather than against this paragraph.

The forecasts are forecasts. The wafer-input and bit-supply shares for 2026 and 2027 are projections. So is the expectation that HBM contract prices rise further. TrendForce publishes no target for how much, and neither will we.

Capacity is the whole cycle. Both companies describe supply as constrained and demand as running ahead of it. Memory has resolved that condition the same way every previous time, by building. SK hynix says it’s holding to capital-expenditure discipline while expanding, and Samsung’s release gives no capital-expenditure figure at all. Watch the disclosed spending in the quarters ahead rather than the damned adjectives.

Someone else is learning. CXMT going from roughly 3% to roughly 8% of DRAM in a year doesn’t threaten HBM leadership. It does put a floor under how tight the commodity end can stay.

6. Four traps

Trap 1: Quoting the concentration figure without its basis and its date. The 48.95% reading counts Samsung common, Samsung preferred, and SK Hynix at the August 7, 2026, close. It was 54.04% a week before. Both halves of that sentence have to travel with the number.

Trap 2: Treating record profits as run-rate. Growth of 557% measures distance from a cheap quarter. Both companies published preliminary, unaudited figures. Price cycles in memory have historically gone both ways.

Trap 3: Assuming HBM share equals DRAM share equals revenue. SK hynix leads HBM at 58%, and Samsung leads DRAM overall. Those are different products with different economics, and a single “memory market share” number usually hides which one is meant.

Trap 4: Assuming semiconductor tariffs hit memory. The Section 232 thresholds describe logic chips. Commodity DRAM and NAND fall outside them as the measure is written. Headlines about chip tariffs and headlines about Korean memory are often about different goods.

7. FAQ

How much of the KOSPI are Samsung and SK Hynix?

Korea Exchange data reported on August 9, 2026, put Samsung Electronics, Samsung Electronics preferred, and SK Hynix at 48.95% of KOSPI market capitalization as of the August 7 close. That was down from 54.04% on July 31, 2026, and from a peak of 58.9% on June 25, 2026. It moves daily, so check the Korea Exchange rather than an old article.

What is HBM, in one sentence?

High-bandwidth memory: DRAM dies stacked vertically and placed next to an AI accelerator so data reaches the processor faster than standard memory allows.

Who leads the HBM market?

On Counterpoint Research’s latest published quarter, the first quarter of 2026, SK Hynix held 58%, with Samsung and Micron at 21% each.

Why did ordinary memory prices rise if HBM is a small share of bits?

Because it isn’t a small share of factories. HBM consumed roughly 18% of DRAM wafer input at the end of 2025 while delivering about 8% of bits. Wafers moved into HBM, subtracting more conventional supply than they add in HBM bits.

Do US semiconductor tariffs apply to Korean memory chips?

The Section 232 measure imposed in January 2026 targets logic integrated circuits meeting specific performance and bandwidth thresholds. Commodity DRAM and NAND are outside those thresholds as written. Trade measures change, so confirm against current US Customs and Border Protection guidance before relying on it.

If I want exposure to this, what am I actually buying?

That depends on the instrument, and it’s worth checking the holdings rather than the name. A broad Korean equity fund has historically carried a very large weight in these companies. Mechanics of buying Korean equities as a non-resident are covered in How to Buy Korean Stocks as a Foreigner (2026 Guide).

---

Last updated: August 19, 2026

This article is for informational purposes only and does not constitute investment, tax, or legal advice. Company results cited here are preliminary and unaudited as published. Market share, index composition, trade data, and tariff scope all change; verify current figures with the primary sources linked above before relying on any of it. Currency conversions use KRW 1,414.29 per US dollar, the rate for August 14, 2026, per Federal Reserve statistical release H.10.