Key points
- Korea's KOSPI fell about 10% and halted twice; SK Hynix and Samsung each dropped about 12%.
- It was a crowded trade unwinding, not a collapse in memory demand.
- Tech steadied, Jensen Huang urged buying the dip, and Micron's earnings were the test.
If you're long anything tied to AI, last night was rough. Asian markets sold off hard, and Korea's two big chip names took the worst of it. By the time the dust settled, though, the story looked less like "the AI boom is over" and more like "a very crowded trade got unwound in a hurry." Here's what happened, why it happened, and the one thing actually worth watching next.
The damage
South Korea's KOSPI finished down roughly 10% at 8,203.84, its lowest close since June 12. The drop was bad enough to trip the exchange's circuit breakers twice in a single session, which almost never happens. (Circuit breakers pause trading for a few minutes to let a panic cool down.)
Memory chips were the center of it:
- SK Hynix: down about 12%
- Samsung Electronics: down about 12%
- Kioxia in Japan: down more than 15%, closing at 92,290 yen
These are the companies that make the high-bandwidth memory (HBM) sitting next to basically every AI chip out there, so when they move this much, people pay attention.
So why did they tank?
There wasn't one clean villain, which is usually how these days go. The honest read is that this was a positioning problem more than a fundamentals problem. A lot of money was leaning the same direction, and once it started heading for the door, it moved fast. A few things piled on at the same time:
- AI valuations had gotten stretched. After the run these stocks had, they were priced for everything to go right, so any wobble stung more than it normally would.
- US semiconductors had already been pulling back, and that selling bled into Asia.
- SpaceX's big debt raise had people nervous about how all this AI infrastructure is actually getting paid for.
- Some regulatory uncertainty in Korea added a local layer of worry on top.
The timing made it worse. This came right after a massive rally. Just a day earlier, on June 22, SK Hynix passed Samsung to become the most valuable company on the Korean market for the first time since 2000, closing near 2,080 trillion won (about $1.35 trillion) after a 5.6% gain. When a stock goes vertical and everyone piles onto the same side of the boat, the tip-over tends to be ugly. That's pretty much what we got.
One thing to be careful with: early on, a few outlets pinned the whole thing on a weak Broadcom AI forecast. That explanation didn't really hold up once you looked closer, so I'd take it with a grain of salt.
How a 10% drop snowballs
A move that big is never one decision. It's a chain reaction. Once the selling kicked off and that first circuit breaker hit, the same crowding that fueled the rally ran in reverse. Leveraged positions got forced out, that forced selling pushed prices lower, lower prices tripped more stops and margin calls, and around it went. The selling fed on itself, and it was concentrated in exactly the names everyone already owned.
The pushback came from a voice worth listening to. Nvidia CEO Jensen Huang told people to buy the dip, arguing that AI is long-term infrastructure and that investors were getting a discount: "We are at the starting point... you should be very happy because you can now buy at a discount." When the guy selling the shovels says the gold rush is just getting started, the bear case loses a little air.
The bounce
Here's the part that matters. The "AI is dead" version of the story didn't survive the day. This looked like a forced unwind, not a broken business. As the selling burned itself out and buyers stepped back in, tech and semis steadied instead of spiraling. Nothing about the real demand for AI memory changed overnight. The price changed. The story didn't.
Futures were ugly before the open
The nerves carried into US futures, and tech took the brunt, no surprise:
- Dow futures: down 0.48%
- S&P 500 futures: down 1.32%
- Nasdaq 100 futures: down 2.70%, more than 800 points
Look at the spread. The Dow barely budged while the Nasdaq got hammered. That tells you this was a chip and AI repricing, not some broad fear about the economy.
The real thing to watch: Micron earnings
You won't learn much about the AI memory story from a circuit-breaker day. You'll learn a lot more from Micron (MU), which reports fiscal Q3 2026 after the close tomorrow, June 24. Micron is the cleanest US-listed way to read the same HBM and DRAM demand that drives SK Hynix and Samsung, and the expectations are huge:
- Revenue: about $34.66 billion expected
- EPS: about $19.95 expected
- Revenue growth: roughly 276% year over year
- Non-GAAP EPS growth: roughly 998% year over year
- DRAM alone: about $26.27 billion, up 271.6% year over year, and the main engine of the quarter
Those are big numbers, and they set up a real test. A strong report with confident HBM guidance would basically answer the "AI is rolling over" crowd, and it would probably drag the whole memory group, SK Hynix and Samsung included, back up with it. A miss or soft guidance would hand the bears their first piece of actual evidence.
A word of caution if you're trading the print, though. Micron has a habit of selling off right after earnings even when the numbers are good. It's a classic sell-the-news move, where traders take profits on a stock they already bid up going in. The recent record shows it:
- Q4 FY2025 (September 23, 2025): record results, but the stock still slid about 17% over the following week.
- Q2 FY2026 (March 18, 2026): a big beat, EPS of $12.20 against about $8.60 expected, and shares still closed down about 3.8% the next day.
- The exception was Q1 FY2026 (December 17, 2025), when blowout records sent the stock up about 10%. So it can break the other way when the surprise is big enough.
The point isn't to call the exact reaction. It's to not read too much into it. If the financials land as strong as expected, a knee-jerk dip that recovers over the next few days is a lot more likely than a real reversal. Don't let a sell-the-news wobble get treated as a verdict on the whole AI memory cycle.
Why these names still matter
Zoom out from the one-day chart and the picture gets pretty clear. The fight that actually decides these stocks is HBM4 for Nvidia's next platform, Vera Rubin, and the split is lopsided:
- SK Hynix: about 60 to 70% of HBM4 volume, with the first-mover lead
- Samsung: about 25 to 30%
- Micron: the rest
That's the real prize, and one rough trading day doesn't change who's winning it. It just changes the price you'd pay to get in.
Bottom line
Korea's chip leaders fell about 12%, the KOSPI lost roughly 10% and broke its circuit breakers twice, and US futures pointed sharply lower into the open. But the cause was a crowded, overheated trade unwinding under valuation and funding pressure, not a collapse in AI demand. Tech steadied, the most important voice in the space said buy the dip, and the real test lands tomorrow with Micron.
Days like this are loud. The fundamentals are a lot quieter. Tomorrow, Micron gets to settle the argument.
Sources
- TheStreet, Nvidia CEO frames the AI selloff as a buying opportunity
- Yahoo Finance, Global chip selloff coverage (June 2026)
This is for information only and isn't investment advice. The figures come from reporting around the June 22 and 23, 2026 selloff, so double-check current prices before acting on anything.