SK Hynix Set a Profit Record and the Kospi Crashed Anyway

SK hynix published the biggest quarter in its history on Wednesday morning in Seoul: 79.32 trillion won of revenue, 60.54 trillion won of operating profit, and an operating margin of 76%. Analysts had modeled roughly 64.2 trillion won of operating profit. Coming in 5.7% under that was enough to knock the stock down double digits and trip a market-wide circuit breaker on the Kospi for the second session running, freezing trading for 20 minutes just after 12:30 p.m. local time.

Two sessions have now wiped out about a fifth of the Korean market. That is not how investors react to a bad quarter. It is how they react to deciding that this was the good one.

RelatedSK Hynix HBM4 Warning Triggers Record Memory-Stock Selloff

• SK hynix Q2 revenue rose 257% year over year to 79.32 trillion won, with operating profit up 557% to 60.54 trillion won, both company records.
• The Kospi fell 10.84% on Tuesday to 6,023.63 and kept falling Wednesday, putting the two-day drop near 21% and July on track for roughly a 35% loss, worse than the 27% recorded in October 1997.
• Samsung Electronics dropped 14.4% Tuesday, its steepest single day since October 2008, and slid another 11.3% Wednesday.
• The selling started before the earnings, triggered by reports that China has begun limited mass production of its own immersion DUV lithography machines.

What SK hynix actually reported

The headline figures are extraordinary by any normal standard. Revenue of 79.32 trillion won was up 51% from the previous quarter and 257% from a year earlier. Operating profit of 60.54 trillion won was up 61% sequentially and 557% year over year. The operating margin reached 76%, which is 35 percentage points better than the same quarter of 2025. First-half revenue crossed 100 trillion won for the first time, and the company finished the quarter holding 88 trillion won in cash against 18.6 trillion won of debt.

The problem sits underneath the total. Growth in commodity DRAM pricing has been cooling while an ever larger share of profit concentrates in high bandwidth memory, and HBM revenue depends on a handful of buyers. SK hynix said HBM4 entered mass shipment during the quarter with a ramp planned for the second half, and that long-term supply agreements are now locked with around ten customers. It also said it will "maintain capital expenditure discipline." In a normal cycle that line reassures investors. In this one it reads as management declining to underwrite the demand curve everyone else has already priced in.

Why does a record quarter get sold this hard?

Because the number that matters is not profit, it is profit versus what the market already assumed. A 76% operating margin is not a sign of a business with room to improve. It is close to the physical ceiling of what a memory maker can earn, which historically marks the top of a cycle rather than the middle of one. When a company printing that margin still lands under consensus, the natural conclusion is that estimates were built on a slope the business cannot climb any further.

There is a second reading, less about SK hynix and more about its customers. Memory demand at these prices is a derivative of AI datacenter buildouts. Anything that makes those buildouts look less certain flows straight into memory pricing, and it flows in with leverage, because memory is the part of the stack with the most volatile margins.

What China's new lithography machines have to do with Korean memory

The selling started before SK hynix reported anything. On Monday and Tuesday, The Information reported, with corroboration from the Financial Times and Reuters, that a state-backed Shanghai manufacturer has begun limited mass production of domestic immersion DUV lithography scanners. The effort absorbed engineering teams from the startup Yuliangsheng and from Shanghai Micro Electronics Equipment. Roughly five machines are targeted for delivery in 2026 and about 20 in 2027, going to SMIC, Hua Hong and CXMT.

The specifications matter more than the headline. These tools reportedly resolve 28nm-class features in a single exposure and can reach 7nm-class geometry through multipatterning, at the cost of tighter overlay requirements and lower yields. They still trail ASML on throughput and reliability, and some critical components are still imported from Japan. Nobody is claiming parity.

For memory specifically, the relevant name on that delivery list is CXMT, China's domestic DRAM maker. Immersion DUV is the tool that gates how far DRAM scaling can go without EUV. A credible domestic supply of those scanners, even a slow and imperfect one, changes the long-run assumption that Chinese DRAM capacity stays capped by export controls. That assumption is embedded in every model of memory pricing for the back half of the decade.

How bad is the damage?

Foreign and individual investors sold a combined 2.6 trillion won on Wednesday while institutions bought about 2.54 trillion won, which is the shape of a market where domestic funds are absorbing an offshore exit rather than one where buyers have disappeared. Leveraged products tied to SK hynix fell about 32%. The damage was not confined to Seoul: Tuesday's session took Micron down 8.9%, AMD down 8.1% and Applied Materials down 7.8% in the United States, while Japan's Nikkei 225 slipped 1.8% to 61,038.63 and Taiwan's Taiex lost 3.6%.

RelatedTSMC Adds $100B to Arizona as Capex Guidance Jumps

1. Jun 22Kospi closes at a record 9,114.55 peak of the AI memory trade
2. Jul 27Reports of Chinese immersion DUV mass production deliveries to SMIC, Hua Hong, CXMT
3. Jul 28Kospi falls 10.84%, Samsung has its worst day since 2008 first circuit breaker
4. Jul 29SK hynix reports records, misses consensus, market halts again two-day loss near 21%
5. H2 2026HBM4 production ramp the number that decides whether this was a top
6. 2027China targets about 20 domestic DUV scanners first real test of capacity impact

Does any of this make your RAM cheaper?

Not soon, and the gap between the two things is worth understanding. We have covered the 2026 memory shortage driving DRAM and NAND prices and the price surge that came with it. A stock repricing does not undo any of that. Contract pricing is negotiated in advance, HBM output through next year is committed under long-term agreements with about ten customers, and SK hynix just told the market it intends to hold capital spending down rather than flood the market with wafers.

If anything, capex discipline at the top of a cycle keeps consumer memory tight for longer. The mechanism that would eventually lower what you pay for a DIMM is Chinese DRAM capacity expanding on domestic tools, and that path runs through five machines this year and twenty next year, at yields nobody outside those fabs has measured. Call it a 2028 question, not a back-to-school one.

What it means for the market

The signal for investors is that the AI trade is being re-sorted rather than abandoned. Memory names carry the most cyclical leverage in the semiconductor complex, so they move first and hardest in both directions, and the same two-day window that halved sentiment in Seoul left Hong Kong's Hang Seng higher. Names to watch for whether this is contained or spreading: Micron as the clearest western read on DRAM pricing, ASML as the direct read on whether Chinese lithography is treated as a real substitution risk, and Samsung as the one company exposed to both the memory cycle and foundry competition at the same time. The tell will be second-half HBM4 pricing rather than any single day's tape.

Our take

The market did not decide that SK hynix is a bad company this week. It decided that the memory supercycle has a visible ceiling, and it found two reasons on consecutive days to test that idea. The earnings miss was small. The margin was enormous. Both of those facts point the same direction: this is what the peak looks like from the inside, and the argument now is over how long the plateau lasts, not whether the numbers were good. For anyone building on this hardware, the practical read has not changed. Memory stays expensive through the HBM4 ramp, and the first genuine relief valve is Chinese capacity that does not exist at scale yet.

Original analysis by GenZTech. Index and share-price figures for July 29 are intraday, captured around the midday trading halt in Seoul, and may differ from the official close. Q2 2025 and Q1 2026 profit figures in Fig 1 are derived from SK hynix's reported growth rates.