The AI hype hasn’t cooled yet—why did the South Korean stock market drop first?

Data centers are still being built, and GPUs are still being bought, but the market has already started to wonder how much money all this spending can ultimately earn back.

If you look only at financial reports, it’s almost the best era in the world for the storage industry.

Samsung Electronics’ first-quarter operating profit for its semiconductor business reached 53.7 trillion won; SK hynix’s operating margin is as high as 72%; Micron has just delivered one of the best quarters in its history, and expects next quarter’s revenue to continue growing; TrendForce expects DRAM contract prices in the third quarter to still rise by 13% to 18%, while NAND prices will rise by 10% to 15%.

In any traditional semiconductor cycle, this is almost a signal that stock prices will keep climbing. It’s just that, right at this moment, the South Korean stock market suddenly turned around.

When the good news starts to stop working

In early July, Samsung Electronics and SK hynix both fell one after another, and the Korea Composite Index saw one of the most intense rounds of adjustments in years, at one point triggering a programmed trading sell-off circuit breaker. A few days later, SK hynix’s home-market shares also logged one of the largest single-day declines since listing.

The market quickly found all kinds of explanations: Meta exploring offering some AI compute services externally, SK hynix completing its ADR listing, South Korea’s margin financing positions being too high, single-stock leveraged ETFs amplifying volatility… Almost every day a new reason appeared, and each reason sounded like a brand-new “ghost story.”

The problem is: why did they start affecting the market exactly at this time?

Just a month ago, SK hynix’s market cap briefly surpassed Samsung Electronics, becoming the largest listed company by market value in South Korea. This was a rare scene in South Korea’s capital markets in over two decades, and it also meant that global AI money had put nearly all the most optimistic expectations onto storage chips.

At that time, HBM was still in short supply, GPU orders were booked into next year, and cloud computing companies kept raising capital expenditures. The South Korean market effectively became the most direct beneficiary of global AI infrastructure investment.

But data from South Korean exchanges shows that in the first half of this year, among the net-sold stocks by foreign investors in the KOSPI, Samsung Electronics and SK hynix together accounted for about 87%. Money wasn’t exiting South Korea across the board—it was cashing in the biggest gains from the past year, concentrated in the two largest positions.

Valuation before the industry starts to shift

If this were the peak of the storage cycle, the industry’s chain data should weaken in sync. But the industry didn’t cooperate with the market. TrendForce still expects DRAM and NAND prices to keep rising in the third quarter; Samsung and SK hynix haven’t slowed their HBM capacity expansion plans; Micron’s latest quarterly performance continues to set records, and its revenue guidance for next quarter is also above market expectations.

It’s hard to believe that AI storage demand has started to cool. In June this year, TrendForce noted that as server DRAM prices keep rising, unit wafer revenue for DDR5 64GB RDIMM has begun to exceed HBM. Over the past year, the market put almost all its attention on HBM. This report was the first to remind the market that the high-cycle optimism brought by AI is spreading across the entire storage industry.

If this were last year, it would almost be a reason to keep chasing. The market didn’t understand it that way. If ordinary-server DRAM also begins to enter a high-profit zone, how long can HBM maintain today’s kind of outsized profits? If AI infrastructure buildout gradually moves into a mature stage, will investment at the same pace still be needed over the next few years? These questions don’t have standard answers. But the capital markets don’t need to wait for the answers to decide where the stock price will go.

For the past two years, Wall Street has pinned everything on one logic: the models are getting bigger, so GPU demand will keep growing; the scarcer the GPUs are, the more valuable HBM becomes; as long as Microsoft, Meta, Alphabet, and Amazon keep expanding data centers, the whole industry chain will benefit. At least today, this logic hasn’t been overturned by reality.

Meta continues to raise capital expenditures this year, and the AI infrastructure investment disclosed by Microsoft, Alphabet, and Amazon remains at historical highs. Samsung, SK hynix, and Micron also haven’t slowed their capacity-expansion pace due to the recent pullback.

Data centers are still being built, and GPUs are still being bought, but the market has already started to care about how much those dollars can finally be turned into profit.

In recent quarters, in the earnings calls of several cloud computing companies, analysts have asked much less about when GPUs will be delivered or when HBM will expand. Instead, they’ve repeatedly focused on other questions: How high is GPU utilization really? After models become more efficient, how much additional compute will still be needed in the future? After spending hundreds of billions of dollars, when can it be recouped?

For stock prices, these questions often matter more than a single piece of bad-news headline. After all, they’re not discussing next quarter anymore—they’re discussing whether profits can keep growing over the next few years. Samsung’s second-quarter results were still a handsome performance, but the market barely rewarded that growth; SK hynix is the same. Hitting new profit highs doesn’t mean valuation can keep expanding.

Why the South Korean market reacted the most fiercely

At least for now, there’s no data supporting that the storage industry has entered a downturn cycle. The stock price correction looks more like the market is trading changes that will happen over the next few years in advance. The South Korean market amplifies that shift.

Since this year began, Samsung Electronics and SK hynix have essentially carried the AI narrative for almost the entire South Korean stock market. Financing balances have climbed rapidly, single-stock leveraged ETFs have rolled out one after another, SK hynix has completed ADR issuance, introducing a new pricing anchor and more cross-market capital.

When stocks are rising, this kind of structure amplifies gains; even small changes in expectations magnify volatility. That’s also why, when AI hardware adjusts, the South Korean market often reacts most dramatically.

In the past few weeks, industry and capital markets have seemed to be looking at two different timelines. The industry is focused on the present: whether HBM orders are decreasing, whether DRAM prices can still rise, whether GPU demand remains strong. The capital market has already started looking ahead—it cares about how much of today’s high profits will remain once this round of AI infrastructure buildout enters a mature phase.

So that’s why the market has shown such a strange situation recently: the industry is getting better and better, but stock prices are starting to pull back—because the two sides weren’t even looking at the same time point.

Maybe years from now, looking back, this round of South Korean stock market adjustments will just be a normal pullback within the AI bull market. Or maybe it will become a turning point as the market reassesses AI infrastructure investment.

At least today, HBM orders have not shown any obvious decline, DRAM prices are still rising, and the four major cloud vendors haven’t cut capital expenditures. The industry is still answering the questions for this year, while the market has already started pricing the valuations for next year.