Samsung Electronics made 107.4 trillion won in operating profit last quarter, about $80.2 billion, according to CNBC's report on its October 8 preliminary results. No other quarter in the company's history has crossed the 100 trillion won line. Revenue was 195 trillion won, which puts the operating margin at about 55 percent. The surprising part is not the size of the profit. It is what produced it.

Prices did the work, not shipments

Samsung did not sell much more memory. It sold the same memory for a lot more money. Seoul Economic Daily, citing Meritz Securities estimates, reports that DRAM shipments rose 3.4 percent from the previous quarter and NAND shipments rose 4.2 percent, while average selling prices climbed 20.1 percent and 19 percent.

Combine those and DRAM revenue grew roughly 24 percent, with about 85 percent of that gain coming from price. That is why profit grew faster than sales: operating profit rose 20 percent from the second quarter on revenue growth of 13.7 percent, according to the company's preliminary figures reported by Malaysian Wireless. When a price rise falls straight to the bottom line, costs barely move.

Commodity memory is now the profit engine

The sharpest numbers are in ordinary chips, not exotic ones. According to DRAMeXchange data cited by Seoul Economic Daily, the contract price of a commodity DDR4 8-gigabit DRAM chip rose from $6.30 in September 2025 to $26 this September, a 4.1-fold increase. A 128-gigabit multi-level-cell NAND chip went from $3.80 to $30.60, about eightfold.

Meritz estimates Samsung's DRAM operating margin at 78.5 percent and NAND at 68.9 percent, for a combined memory margin near 76 percent, roughly 10 points above the 66 percent Nvidia recently reported. The same estimate puts memory profit at about 109 trillion won, more than Samsung's total. Subtract the company's reported 107.4 trillion won and the other businesses, which include devices, foundry and System LSI, appear to have lost around 1.6 trillion won between them. For one quarter, Samsung was effectively a memory company with a phone maker attached.

High-bandwidth memory squeezed everything else

The shortage in ordinary memory is a side effect of AI's appetite for the special kind. Seoul Economic Daily reports that high-bandwidth memory (HBM) consumes more than three times as many wafers as conventional DRAM of the same capacity. Every wafer that becomes HBM is a wafer that does not become memory for a laptop or a phone.

KB Securities projects that commodity products will fall to 59 percent of global DRAM capacity next year, from 65 percent this year and 73 percent last year. The same report says Samsung became the first company to mass-produce sixth-generation HBM4 and is expanding shipments to customers including Nvidia. Douglas Research Advisory's Douglas Kim estimates Samsung's HBM shipments by volume rose nearly 50 percent from the prior quarter, as Reuters reported. AMD chief executive Lisa Su told reporters in South Korea on October 7 that memory supply is tight. That demand is partly the pull from systems like the one we covered when Nvidia's Vera Rubin debuted at MLPerf, and TrendForce expects average HBM selling prices to rise 121 percent next year.

The forecasts assume customers will keep paying

Analysts expect more. FnGuide's consensus puts Samsung's 2026 operating profit at 366.4 trillion won and 2027 at 538.9 trillion won, a 47 percent rise, while Meritz forecasts 632.4 trillion won for 2027. Those estimates need prices to hold.

There are warning signs. A brokerage report cited by the Korea JoongAng Daily expects commodity DRAM and NAND prices to rise only 3 percent and 12 percent next quarter as customers push back and consumer demand stays weak. Samsung's revenue of 195 trillion won also fell short of the 206.8 trillion won FactSet had expected, as TradingKey noted, even though profit beat expectations. Seoul Economic Daily adds that some customers may cut the memory they install to save money.

That last point matters for devices. Microsoft and Nvidia recently launched PCs that run 120B-parameter AI models locally, and local models need large amounts of memory. My read is that device makers will face a choice between absorbing higher memory costs and shipping thinner configurations, though no source has said which way they will go.

What the October 29 results must show

The preliminary release is thin. It gave totals only, and KuCoin's summary notes Samsung has not said how much HBM4 contributed. The company's full report on October 29 is the first chance to separate HBM from commodity profit, to see how deep the non-memory losses run, and to learn how much of the margin is price rather than product. If commodity memory is carrying most of the profit, the boom depends less on AI chips than on how long the shortage lasts.