Everything About AI
NextFin News - Samsung Electronics delivered its biggest quarterly profit ever, yet the number was not big enough. The world's largest memory-chip maker said its third-quarter operating profit rose to about 107.4 trillion won ($80.1 billion), a near-tenfold jump from a year earlier, but the print fell short of the 108.67 trillion won analysts had penciled in. Shares dropped 2.4% in Seoul and the benchmark KOSPI index slid 2.6%, a reminder that in the AI chip boom the bar keeps moving faster than the companies chasing it.
The miss was narrow in absolute terms — roughly 1.2% below the high end of expectations — but it landed in a market that had stopped rewarding good news and started demanding perfection. Revenue more than doubled to 195 trillion won from 86.06 trillion won a year ago, and the profit was up about 20% from the previous record Samsung set just one quarter earlier. The problem was not the size of the boom. It was the suspicion that the boom may have already peaked.
The Numbers: A Record That Wasn't Enough
Samsung's preliminary results, released early Thursday in Seoul, put third-quarter operating profit at around 107.40 trillion won. The figure was up nearly tenfold from the 12.2 trillion won earned in the third quarter of 2025 — growth of roughly 782% — and marked the fourth consecutive quarter of record operating profit. Sales of 195 trillion won represented more than a doubling of the top line year over year.
Against that backdrop, the shortfall against the 108.67 trillion won forecast reads as small. But the market's reaction suggests investors are no longer measuring Samsung against its own past. They are measuring it against a narrative of endless AI-driven scarcity, and anything less than a blowout gets punished. The memory trade's most crowded positions — Micron, SK Hynix, and the Roundhill Memory ETF — all came under pressure, with the ETF falling nearly 7% in U.S. premarket trading.
The timing matters. The preliminary print arrived just a week after Micron posted bumper quarterly earnings and signaled that AI-fueled tightness in memory supplies was likely to continue well into 2027. That should have been a tailwind. Instead, Samsung's modest miss became the excuse traders needed to take money off the table. The sequence is telling: good news from the category leader was not enough to lift the laggard. That is not how a bull market behaves when conviction is high.
"Micron's record fiscal Q3 financial results and even stronger outlook for Q4 reflect the strategic value of memory in the AI era," said Micron chief executive Sanjay Mehrotra.
But even Mehrotra's confidence could not stop the memory complex from selling. The gap between what the category leader said and how the market reacted is the story in one line: the market is no longer listening to the earnings; it is listening to the exit.
What's Really Driving the Profit: Chips Alone, Not the Whole Company
The uncomfortable detail beneath Samsung's headline is how concentrated the gain is. The semiconductor business is carrying the entire earnings story, while the rest of the conglomerate is under pressure. Estimates point to the mobile unit recording an operating loss in the quarter that exceeded $1 billion, while the foundry division is also projected to remain in the red as rising input costs squeeze margins.
That split is the crux of the investment question. In the second quarter, Samsung's Device Solutions division — memory and foundry combined — generated 127.5 trillion won in revenue and 89.2 trillion won in operating profit, while the mobile and networks unit posted a 700 billion won loss. If the third quarter followed a similar pattern, then essentially all of Samsung's record profit came from memory, and the consumer-facing businesses were a drag.
This concentration cuts both ways. On the upside, memory is the asset class of the AI era: scarce, priced in dollars, and demanded by every hyperscaler building data centers. On the downside, a company whose earnings depend almost entirely on one cyclical product line is exposed to the moment that cycle turns. Samsung's shares have already slumped more than 25% from their record high in June, even before this print, as investors began pricing in the possibility that the AI buildout is maturing. The stock's all-time high of 374,500 won now looks like a distant peak rather than a launching pad.
The competitive backdrop makes the concentration risk sharper. Samsung is playing catch-up in the one memory segment that matters most. SK Hynix commands roughly 58% of global HBM revenue, with UBS analysts expecting it to hold about half of total HBM bits through 2026. Samsung, by contrast, is still working through qualification of its HBM3E and next-generation HBM4 products with Nvidia. Until that qualification is complete and shipping at scale, Samsung is earning the commodity upcycle without fully participating in the AI premium.
Cyclical Windfall or Structural Shift?
The central question for Samsung is whether the AI memory boom is cyclical — a supply squeeze that will mean-revert — or structural — a permanent repricing of the industry's earnings power. The evidence points to both, operating on different time horizons, and confusing them is the easiest way to get the stock wrong.
The cyclical leg is real and powerful, and it fits a pattern memory investors have seen before. In the 2016-2018 supercycle, DRAM prices tripled as smartphone memory content surged and supply lagged; producers printed money, capex soared, and then prices collapsed more than 60% in 2019 as new capacity arrived. The 2020-2021 pandemic cycle repeated the script: work-from-home demand tightened supply, margins expanded, and the downturn in 2022 wiped out most of the gains. Samsung itself swung from record profits in 2022 to an operating loss in its chip division by late 2023. Memory does not do gentle landings.
The current upswing fits the classic pattern: demand from AI server builders collided with constrained supply, prices rose, margins exploded, and every producer rushed to spend more. Samsung's profit has gone from 20.1 trillion won in the fourth quarter of 2025 to 57.2 trillion in the first quarter of 2026, 89.5 trillion in the second, and now 107.4 trillion. That trajectory — four straight records, each quarter materially higher than the last — is what a cycle peak looks like, not what a stable new normal looks like.
But the structural leg is equally real. AI inference and training require orders of magnitude more memory per unit of compute than traditional workloads, and high-bandwidth memory is not a product any competitor can simply turn on. The technology barriers, the multi-quarter qualification cycles with customers like Nvidia, and the capital intensity of leading-edge fabrication mean the industry's competitive structure has changed. UBS forecasts blended HBM pricing per bit to rise 18.5% year over year in 2026, driving HBM revenues to a projected $32.7 billion — a market that barely existed five years ago. Even when prices normalize, the floor for memory earnings sits higher than it did before 2025.
The right read, then, is a cyclical supercycle riding on a structural floor. Profits will fall from these levels — they always do in memory — but they are unlikely to fall back to the pre-AI trough. The risk is that the market is currently pricing the structural floor without leaving room for the cyclical drawdown. A stock that has tripled in a year does not need earnings to collapse to fall; it only needs them to stop accelerating.
The Second-Order Problem: What Happens When the Customers Blink
The first-order story is simple: AI spending drives memory demand, memory demand drives Samsung's profit. The second-order question is what happens to that chain if the hyperscalers slow down. And there are early signs they might.
A report on data center investment showed at least $260 billion of projects canceled in 2026, with the pace of cancellations picking up through the September quarter amid disruptions in funding and energy supplies. That does not mean AI demand is collapsing — Micron's guidance into 2027 argues the opposite — but it does mean the growth rate is no longer frictionless. Every canceled data center is one less order for HBM, one less reason for a hyperscaler to sign a long-term contract at a lofty price.
Samsung knows this. The company said in July it aims to secure long-term contracts covering about two-thirds of its memory output — a defensive move that locks in volume but caps upside if spot prices keep climbing. TrendForce expects the slowdown in memory price gains to continue into the fourth quarter. Put together, the picture is of a company trying to trade peak-cycle pricing for guaranteed demand, which is exactly what producers do when they suspect the top is near.
The market reaction is the transmission mechanism in action. Samsung's miss did not change the fundamentals of the AI boom. It changed the confidence that the boom can keep accelerating. That is why a 1.2% earnings shortfall produced a 2.4% stock decline, a 2.6% drop in the KOSPI, and a 7% premarket hit to the memory ETF. The repricing is not about this quarter. It is about the discount rate investors are now applying to next year.
There is a deeper layer still. The memory upcycle has been financed, in part, by the same hyperscalers who buy the chips. When those customers face their own funding and power constraints, the demand curve becomes endogenous: the buyer's balance sheet is the supplier's ceiling. Samsung's record profit is, in a real sense, a claim on capex that has not yet been fully committed. If that capex slows, the claim ages poorly.
The Counter-Thesis: The Miss Is Noise, the Trend Is Intact
The strongest argument against the cautious read is that a 1.2% miss against a record profit is noise, and that the underlying trend remains overwhelmingly positive. SK Hynix and Micron have both guided to continued strength, Nvidia's demand for advanced memory shows no sign of slowing, and the supply-demand gap in HBM is still wide. From this angle, selling Samsung on a near-record print is mistaking a speed bump for a wall.
There is real force to that view. Memory cycles are long, and the AI infrastructure buildout is still in its early innings by historical standards. Data center construction cancellations, while large in dollar terms, may be concentrated in speculative projects rather than core hyperscaler capacity, in which case the impact on memory demand would be minimal. And Samsung's own progress on next-generation memory — reports suggest the company is close to securing Nvidia certification for HBM4, with mass production potentially beginning in February — could re-rate the shares regardless of one quarter's print.
But this counter-thesis has a specific failure point. It depends on memory prices continuing to rise, or at least holding, through 2027. If contract prices roll over sooner than Micron and Samsung expect — if the long-term deals Samsung is signing turn out to have been struck below where spot would have gone — then the structural-floor argument weakens and the cyclical drawdown dominates. That is the signal to watch: not the profit number itself, but the pricing trajectory embedded in those new contracts. A single quarter of flat or falling contract pricing would do more damage to the thesis than any earnings miss.
What to Watch Next
Samsung's detailed results, due later this month, will matter more than the preliminary headline. Investors will be looking for three things: the semiconductor division's operating profit, the trajectory of HBM shipments and pricing, and any update on foundry losses and mobile profitability. A strong chip number with a clear path to Nvidia qualification would do more for the stock than the preliminary print did against it.
Beyond Samsung, the memory complex is now a referendum on AI capex durability. Micron's guidance into 2027 sets the anchor; any downgrade there would hit the entire group. SK Hynix's capital spending plans — raised to the high-40 trillion won range this year from 30.2 trillion won in 2025 — signal confidence but also add future supply that must be absorbed. And TSMC's September sales, up 54.6% year over year to take third-quarter sales 50% higher, remain the cleanest read on whether hyperscaler demand is still accelerating. If TSMC's momentum holds while memory prices stall, the divergence itself becomes the warning sign.
The scenarios are clear. In the base case, memory prices grind higher but at a slowing pace, Samsung completes HBM qualification with Nvidia in the coming quarters, and the stock trades in a range as investors wait for proof that demand can absorb the coming supply. In the upside case, AI capex re-accelerates, contract pricing surprises to the upside, and Samsung's chip division re-rates toward SK Hynix's multiple as the HBM catch-up trade takes hold. In the downside case, data center cancellations accelerate into 2027, memory prices roll over, and the market discovers that a near-tenfold profit jump was the peak, not the new floor.
The time-horizon split matters. The short-term view is cautious: expectations are too high, the cyclical leg is mature, and the stock has more to prove. The medium-term view is neutral-to-positive: the structural floor is real, and Samsung's valuation has already absorbed a lot of the pessimism after falling more than a quarter from its peak. The long-term view is where the structural argument wins — memory is a permanent, larger part of the AI stack, and Samsung remains one of only three companies that can make advanced memory at scale, alongside SK Hynix and Micron.
Samsung's record profit was real. The market's disappointment was also real. The difference between them is the price of perfection in an AI market that may have just started demanding it. The lesson of this quarter is not that the boom is over. It is that in a market built on the promise of endless acceleration, a record is only a record until someone asks what comes next — and Samsung, for all its nine-fold profit surge, did not have an answer big enough.
Explore more exclusive insights at nextfin.ai.
