At the Peak of Record Profit, Samsung: The AI Memory Tide and the Market's Blunt Doubt

In early October, Samsung Electronics issued a forecast that stands as a milestone in semiconductor history. For the third quarter of the year—July through September—the company's operating profit is projected at 107.4 trillion won, equivalent to roughly $80.17 billion. That is about nine times the figure from the same quarter a year earlier. Revenue rose to approximately 195 trillion won, a year-on-year increase of about 127 percent. Such a leap is rare in this industry. Yet immediately after the announcement, Samsung's share price sat more than 25 percent below its historic June peak. A record profit headline and a falling share price—set the two side by side, and the real question of the day becomes obvious. Having followed the chip market's rises and falls for years, I keep noticing one thing: the market's great turning points sometimes show up in the announced numbers, and sometimes in the gap between those numbers and the price. In Samsung's case, it is the second that is happening now. The question is not whether profit grew—it did, historically. The question is how long this pace of profit can hold, and exactly which engine is pulling it along. The background needs to be laid out clearly. Over the past year, the surge in building artificial-intelligence infrastructure has at its core high-bandwidth memory, or HBM. The advanced processors placed in AI data-centre servers cannot function without HBM chips beside them. That demand has lifted the prices of both DRAM and NAND—the two main types of memory—in a way not seen in about a decade. Samsung itself has acknowledged that the bulk of its profit came from the memory division. In other words, there is a single engine behind this record. Put simply, memory chips are a computer's memory. DRAM holds temporary memory, NAND stores data permanently, and HBM is a special stacked form of DRAM that sits beside an AI processor and feeds it enormous volumes of data at very high speed. The moment the world's largest technology companies began building AI data centres, demand for HBM and advanced DRAM exploded. Some analysts forecast that this supply-demand imbalance could last until 2028. But that is a forecast, not a certainty. There is another side to this single-engine dependence. In the same period, the company's mobile division lost more than one billion dollars in the third quarter. And the foundry division—contract chip manufacturing—remained loss-making throughout the quarter, with low utilisation of its manufacturing capacity. On top of that came currency pressure: as the won strengthened, the won value of dollar-denominated overseas sales fell, prompting analysts to trim their profit forecasts. In other words, record memory earnings are covering the shortfall of two loss-making divisions. This is where the real analytical work begins. However eye-catching the headline profit figure may be, its quality is decided by two questions—first, how broad the earnings are, or whether they are concentrated at a single point; second, whether the pace of those earnings is rising or falling. The answer to the first is uncomfortable: earnings are highly concentrated. The answer to the second is more uncomfortable still: the pace is slowing. Samsung's position in the HBM market is gradually strengthening. By analysts' reckoning, HBM bit shipments rose about 50 percent quarter over quarter, narrowing the gap with the leader, SK Hynix. That progress is real, and it is the company's greatest strength. But there is a subtle point here. In HBM, the market leader is still SK Hynix, while in conventional DRAM and NAND Samsung stands roughly level with Micron and SK Hynix. So Samsung is at once a lagging challenger and a beneficiary of rising prices—a dual role that makes the foundation of its earnings unstable. The curious thing is that three big players in the memory market—Samsung, SK Hynix and Micron—are all posting record profits almost simultaneously. That is no coincidence. When so few suppliers raise prices at the same time, the market effectively moves under a narrow oligopoly, where pricing power temporarily rests in everyone's hands. But that power is not permanent—one new entrant changes the equation. And that new entrant is coming from China. Chinese memory makers are gradually increasing supply, which over the medium term could erode the collective pricing power of Samsung, SK Hynix and Micron. If Chinese supply ramps up quickly, the imbalance currently expected to last until 2028 could end earlier. This is a medium-term risk that does not make headlines today—but whose impact is the largest. In foundry, the picture is entirely different. There Samsung has traditionally lagged TSMC, and although there is hope of a recovery in utilisation at advanced nodes, it is not yet clear. The question is how long this division can run on customers' money. And if memory prices soften one day, the foundry losses can no longer be hidden. The numbers themselves give the clearest warning. Analysts expect profit growth to slow in the fourth quarter to 8.2 percent quarter over quarter, a sharp drop from 20 percent in the third quarter. More specifically, DRAM contract-price growth could fall from about 60 percent in the second quarter to 10 to 15 percent in the fourth. That is, the pace of price increases—the real driver of profit—is clearly slowing. One more fact is worth remembering. Samsung's forecast of 107.4 trillion won is slightly above the LSEG SmartEstimate of 106.1 trillion won. Slightly—not spectacularly. In other words, the market had already priced in roughly this much. Where profit is only slightly above expectations, the price sits 25 percent lower—that gap is the real story. Now let us ask where this profit is actually being generated and where it stops. Across the AI supply chain, Samsung is both supplier and buyer. Upstream—in HBM, DRAM and NAND supply—price increases are a blessing. But midstream and downstream—where mobile phones and consumer electronics are made—the same chip price increases raise its input costs. So the company is simultaneously raising its own profit and its own costs. This self-competition is the hidden story. Now consider the conventional read. The natural explanation is: record profit means a strong company, so this is a chance to buy the shares. That argument has worked many times before, and for four straight quarters Samsung has posted record profits—a long winning streak. But precisely for that reason, the argument is weaker this time. Four record quarters are a backward-looking sample; a single up-cycle does not prove future durability. When the market pushes the price down despite record profit, it is really saying: the pace looks to me like it is running out, and the next engine has not yet been built. So the picture being drawn for Samsung is not a golden age—it is a moment at the peak. Celebration in the media headlines, caution in the market's behaviour. That contradiction is the biggest signal. The market is pricing future risk, not celebrating present achievement. Investors' attention is now on shareholder-return policy—that is, whether an announcement comes to buy back shares or raise dividends. That attention itself tells us the market is waiting for a new incentive. Taken together, the risk picture is clear. The biggest risk is the memory price cycle turning—and the likelihood of that turn is no longer a distant matter. The second risk is on the demand side: if AI-related investment slows, the memory price tide can also ebb quickly. The third risk is on the cost side: higher memory prices mean higher input costs for Samsung's own mobile division. And the fourth risk is the concentration of earnings—when the entire profit rests on a single division's shoulders, a small crack there shakes the whole picture. Here long experience says one thing: single-engine growth always stops fastest. When a single division carries the whole company's profit while two others run at a loss, the moment that one engine slows, the whole picture changes. For Samsung, that moment is probably not far off—though it is not inevitable either, because its HBM progress is real. The most important date ahead is October 29, when Samsung will release its detailed quarterly results. Three things must be watched that day: the true size of the losses in the mobile and foundry divisions, the announcement on shareholder returns, and the new rate of DRAM and NAND price growth. These three numbers will decide whether Samsung's record is in fact a peak, or the start of a new era.