NextFin News - South Korea’s memory-chip darlings are running into the one thing that has always humbled the sector: prices and sentiment can outrun fundamentals, but they rarely outrun the cycle for long. SK Hynix and Samsung Electronics have spent months benefiting from the AI buildout, yet on July 28 South Korean chip stocks fell as much as 10.9% and 9.5%, respectively, as investors started to question whether the market had already priced in too much of the AI memory story.
The move matters because it came after a spectacular rerating. On June 22, SK Hynix overtook Samsung Electronics to become South Korea’s most valuable listed company, a symbolic turn that captured how far the memory trade had run. By early July, Samsung was expected to report an operating profit of 86 trillion won for the second quarter, according to a consensus tracked by 30 analysts, while SK Hynix reported record quarterly revenue of 79.3187 trillion won and operating profit of 60.5426 trillion won in its second quarter. Those are not numbers that look like a sector in collapse. They are numbers that look like a sector whose profitability has become so elevated that the market is no longer rewarding it for merely being good.
That is the tension. The underlying business is still strong, but the stocks have become increasingly vulnerable to any sign that investors are comparing the next six months with the previous six months and concluding that the easy money has already been made. The July 28 selloff was triggered by worries about AI infrastructure financing and intensifying competition from China, but the larger issue is broader: the memory market has entered a phase where expectations, not just earnings, now drive price action.
This is where the story splits. In the near term, the pullback looks cyclical, because memory chips are among the most cyclical products in semiconductors and because sharp rallies often reverse when positioning gets crowded and valuation expands ahead of results. In the longer term, however, the supply-demand backdrop still has a structural edge. SK Hynix chief executive Kwak Noh-jung said the industry is heading toward its worst year from a supply perspective in 2027, and Samsung’s own profit outlook implied that AI-related demand was still pushing conventional memory prices higher. The question is not whether demand exists. It is whether the market has already discounted a lot of it.
What Actually Broke?
The first-order explanation for the slide is straightforward: investors sold AI-related semiconductor exposure when financing concerns around hyperscaler spending collided with renewed competition from China. That matters because memory chips sit one step removed from the headline AI model race. They do not get paid for hype alone; they get paid when compute demand translates into orders for HBM and DRAM. When investors begin to worry about how the AI infrastructure is being funded, they are effectively asking whether the capital intensity of the buildout will slow, and whether that slowdown will hit memory demand before the next capacity step-up arrives.
But the deeper mechanism is more revealing. Memory equities are leveraged not just to shipments, but to expectations of scarcity. When the market believes supply will stay tight, it assigns outsized value to future pricing power. That is why a company can post record revenue and still disappoint: the stock is not merely discounting current profit, it is discounting the slope of future profit. If investors decide that the next increment of upside is smaller than the last one, the multiple compresses even if the quarterly results are still exceptional.
That is also why SK Hynix’s and Samsung’s market moves can diverge from the fundamental trajectory. SK Hynix said second-quarter revenue reached 79.3187 trillion won and operating profit 60.5426 trillion won, both record highs, while Samsung was expected to report 86 trillion won of operating profit for the quarter based on a 30-analyst SmartEstimate. Those are extraordinary figures. They also create a higher bar. Once profits reach that scale, the market stops asking whether the business is good and starts asking whether it is better than the new expectation embedded in the share price.
Kwak Noh-jung, chief executive of SK Hynix, said the company forecasts “next year will be the worst year in the industry’s history from the supply perspective.”
That line matters because it shows the industry still sees scarcity, but it also reveals the trap. If supply is set to remain tight, then the long-run thesis is intact. If, however, the market has already moved to price an even tighter world than management itself is forecasting, the stock can still fall. In that sense, the July rout is less a verdict on the business than a repricing of the distance between reality and the supercycle narrative.
The cyclical call is the easier one. Memory stocks have always traded in waves because the industry expands capacity aggressively, overshoots demand, then works through a correction when supply catches up. Even in a healthy phase, the stocks tend to mean-revert faster than the fundamentals. That pattern is visible in the speed of the recent move: the sector had already enjoyed a huge rerating before the latest drop, so the selloff can happen without a deterioration in end demand. The market only needs a reason to stop paying up.
Is This Just a Cyclical Shakeout?
Short term, yes. The evidence points to a cyclical pullback rather than a structural break. The industry is still being pulled by AI infrastructure spending, and the supply side remains constrained enough that Samsung’s second-quarter operating profit was expected to rise to 86 trillion won. SK Hynix’s second-quarter results were even more striking: 79.3187 trillion won in revenue, 60.5426 trillion won in operating profit, and an operating margin of 76%. Those numbers do not describe a demand cliff. They describe an industry still enjoying extraordinary pricing conditions.
What makes the move cyclical is the transmission mechanism. AI infrastructure spending is lumpy, customer ordering is concentrated, and memory pricing amplifies every change in sentiment. When investors fear that hyperscalers may slow the pace of spending or turn to funding structures that reduce near-term chip orders, they do not need to prove a collapse in demand. They only need to believe the next phase of growth will be less explosive than the last one. For memory stocks, that is often enough to trigger a sharp de-rating.
The market has also been pricing the trade as if scarcity alone guarantees share-price resilience. It does not. A tight market can still disappoint if it is already fully capitalized into expectations. The June 22 milestone, when SK Hynix became South Korea’s most valuable listed company, is a case in point. That moment was not merely a triumph; it was also a sign that the market had moved from underappreciating Hynix to celebrating it as the clearest AI-memory winner. Once that happens, any wobble in the AI capex narrative becomes a reason to trim exposure, because the first leg of the rerating has already been captured.
The historical pattern supports that view. Memory markets have repeatedly gone through phases in which investors extrapolate a shortage, push valuations higher, then discover that supply and pricing can normalize faster than expected. The exact catalyst changes — smartphone demand, server demand, DRAM mix, NAND oversupply — but the structure does not. This time the catalyst is AI and HBM, not consumer electronics. Yet the market psychology is familiar: first the trade becomes obvious, then crowded, then fragile.
So the short-term verdict is simple: this is a cyclical reset inside an otherwise strong industry. The rally had become vulnerable because sentiment outran the next leg of earnings. That does not mean the cycle has peaked. It means the stocks are now much more sensitive to any sign that the earnings slope is flattening.
What Changes If The AI Memory Boom Really Is Different?
The longer-term case is more interesting, because it is the one bulls will use to argue that the drawdown is merely noise. Their argument starts with scarcity. SK Hynix’s chief executive said the industry faces its worst supply year in 2027, and Samsung’s profit expectations point to continued tightness in memory pricing. If those conditions persist, then the current selloff may simply be a pause in a multi-year earnings upcycle, not the beginning of a structural reversal.
There is a credible structural argument here. AI data centers consume memory in a way that older server cycles did not, especially through high-bandwidth memory and advanced DRAM. That changes the demand mix. It also changes the bargaining power of the leading suppliers, because a smaller group of qualified vendors can capture more of the value chain. In that sense, the AI memory market is not just another commodity cycle; it is a capacity-constrained, technically specialized market in which product differentiation matters more than it used to.
Still, a structural thesis needs more than strong quarterly profits. It needs evidence that the old mean-reversion logic is breaking down. That means three things: supply discipline that lasts long enough to prevent overshoot, pricing power that survives the next wave of capacity additions, and customer demand that remains broad-based rather than concentrated in a few hyperscalers. If those conditions hold, then the current weakness would be a tactical shakeout inside a changed industry.
The strongest counter-thesis is that the AI boom itself may be entering a more cautious phase. Investors are increasingly focused on how large technology firms will fund their AI buildouts, and whether the returns on those investments will justify the scale of capital expenditure. That concern is not trivial. If financing costs rise, or if customers delay orders while they assess the economics of AI deployment, memory demand can soften even if the secular AI trend remains intact. In that scenario, the memory trade could still look structurally attractive over several years while being tactically exposed over the next few quarters.
The key falsifying signal for the bullish structural view would be straightforward: if AI infrastructure spending slows materially and Samsung or SK Hynix begin to miss consensus by a wide margin for two consecutive quarters, the idea that this is merely a temporary wobble would weaken fast. If, instead, both companies continue posting outsized profits while supply remains tight into 2027, the structural case gets stronger and the market’s recent selloff will look more like a compression of expectations than the end of the story.
That is the second-order point the market may be missing. The first-order reaction is to sell memory stocks when AI financing concerns flare. The second-order implication is that the winners may become more selective, not less important. If customers keep ordering fewer but more specialized, higher-margin chips, then the trade shifts from broad beta to name-specific execution. In other words, the market may stop paying for the whole category and start rewarding only the suppliers that can prove they own the bottleneck.
Who Benefits, Who Is Exposed, And What Comes Next?
Over the short term, the beneficiaries of the pullback are the investors who had treated the memory trade as a one-way bet. A selloff after a huge rerating often clears out crowded positioning and resets valuation. It also reminds the market that even a structurally attractive theme can have a bad quarter, especially when the narrative depends on continuous AI capex growth.
The companies most exposed are the ones whose share prices now assume that every favorable trend in memory pricing will persist. Samsung and SK Hynix both still enjoy strong fundamentals, but the larger the rally, the more they become hostage to the next set of earnings expectations. Their stock performance will increasingly depend on whether they can show that demand is still outrunning supply after the current wave of excitement cools.
In the medium term, the key watchpoints are simple: whether AI infrastructure spending keeps accelerating, whether memory prices remain firm into the next earnings season, and whether management commentary continues to point to shortages rather than normalization. Any sign that order growth is slowing, or that customers are delaying purchases while financing scrutiny rises, would strengthen the case that the market has begun to price a cooling cycle rather than a durable supercycle.
In the long term, the structural question is whether AI has changed memory from a cyclical commodity into a scarce strategic input. If HBM and advanced DRAM remain bottleneck products with a concentrated supplier base, then the market may keep rewarding scale, technical depth, and customer access rather than simple unit growth. If capacity expands too quickly, however, the old cycle will reassert itself, and the current anxiety will end up looking like the first phase of a familiar correction.
The base case is that the latest drop is a cyclical reset inside a still-constructive structural trend. The upside case is that demand stays extreme and supply remains constrained enough to keep pricing power intact into 2027. The downside case is that AI capex slows, consensus earnings peak before the market expects, and the memory rally is revealed as a classic boom-bust trade with a new label.
That is why the selloff should not be read as the death of the AI memory story. It is the market asking whether the story has already been sold too many times.
As of July 29, 2026, the memory trade still looks powerful enough to matter and crowded enough to crack.
Explore more exclusive insights at nextfin.ai.
