107.4 Trillion Won on a Revenue Miss (Samsung Q3 Preliminary, 8 October 2026): Why Memory Prices Made the Record — and Why the Same Prices Squeeze Samsung's Phones

Samsung Electronics guided third-quarter operating profit to a record 107.4 trillion won (about $80 billion) on 8 October, up 782.5% from a year earlier, on revenue of about 195 trillion won. Profit came in slightly above consensus while revenue fell short of it. The record is a memory-price story: the same shortage that makes Samsung's chip division one of the most profitable businesses in the world is raising costs at its own phone business, which posted its first loss last quarter.

What Samsung actually reported

The preliminary release is a short guidance statement, not a full earnings report. It gives two numbers and nothing else. According to CNBC and the Korea Herald, Samsung guided:

Three facts give the scale. It is the first time the company's quarterly operating profit has passed 100 trillion won. It is the fourth record quarter in a row. And at about $80 billion, Seoul Economic Daily reports it is above the highest quarterly operating profit Nvidia has posted. The 783% growth rate overstates the momentum, because it is measured from a weak base: the same quarter of 2025 earned roughly 12 trillion won. The quarter-on-quarter figure — another 18 trillion won added in three months — is the better guide to where the cycle stands.

What the release does not say matters just as much. It has no split between the chip (DS) division and the device (DX) division, no memory pricing commentary and no guidance. Everything in this post about why the number looks like this comes from market data and the second-quarter report, not from today's statement.

A profit beat and a revenue miss in the same release

Korean brokers had published their forecasts ahead of the release. TrendForce's round-up of Korean press estimates puts Yonhap's consensus at 105.41 trillion won of operating profit on 200.26 trillion won of revenue, with JoongAng Ilbo citing about 106 trillion won and Chosun Ilbo 106.94 trillion won.

So profit beat by roughly 2% and revenue missed by roughly 2.6%. Profit up while revenue falls short means each won of sales earned more than analysts had modelled. That is consistent with the mix shifting further towards memory, where margins are highest, and away from finished devices, where they are thin or negative.

There is a second detail in the expectations. As recently as August, Chosun Ilbo reported, brokers were forecasting more than 113 trillion won. Estimates were cut by roughly 6–7 trillion won in the run-up — partly because the won had strengthened sharply against the dollar. In Seoul the dollar fetched 1,340.4 won on 7 October, per the Asia Business Daily's 3:30 p.m. reading. Memory is priced in dollars globally; Samsung reports in won. A stronger won means fewer won per dollar of chip sales. So the result beat a bar that had already been lowered.

The engine: memory prices, not memory volumes

The cause of the record is the price of memory. Seoul Economic Daily, citing DRAMeXchange, reports that a benchmark DDR4 DRAM chip averaged $26 at the end of the third quarter. That is more than four times the $6.30 a year earlier and 23.8% above the $21 at the end of the second quarter. A benchmark NAND flash chip rose to $30.61 — about eight times its year-earlier level.

Why price has this effect on profit is the core mechanism. A memory fab is a huge fixed cost: the equipment, the cleanroom and the engineers cost roughly the same whether a wafer sells for a little or a lot. When the selling price quadruples and the cost of making the chip doesn't, almost all of the increase is profit. That is how Samsung's chip division can run margins that TrendForce's sources estimate at around 75% for the division and above 80% for memory alone — numbers that would be extraordinary for a software company, let alone a manufacturer.

Why prices are this high comes down to where the wafers go. AI accelerators need high-bandwidth memory — stacked DRAM sitting next to the processor — and HBM uses far more wafer area per bit than ordinary DRAM. Every wafer turned into HBM for an AI server is a wafer not turned into memory for a phone or a PC. Demand from AI pulls supply away from everything else, and the shortage shows up in the commodity price.

The US data point lines up. Micron's fiscal fourth-quarter release on 30 September showed revenue of $54.23 billion against $11.32 billion a year earlier, a GAAP gross margin of 86.8% and an operating margin of 80.7%. It guided the next quarter to $61.5 billion, plus or minus $1.5 billion. Two of the three largest memory makers, reporting a week apart in different currencies, tell the same story. (Our Micron preview covers why the quarter just reported ran 14 weeks, which flatters it against the next 13-week guide.)

The same price, booked as a cost

The less obvious part: Samsung is not only the world's largest memory seller. Through Galaxy phones, TVs and appliances it is also a large memory buyer.

In the second quarter, Samsung's Device eXperience division — phones, TVs and appliances — posted an operating loss of 0.8 trillion won, its first quarterly loss since the division was set up in 2021, according to Seoul Economic Daily's report on the Q2 results. The mobile unit accounted for 0.7 trillion won of that, and the paper attributed it to "chipflation" — the cost of the memory inside each device.

Inside one group, the memory price is moving money from one division to another. The chip division books the higher price as revenue; the phone division pays it as cost. Netted out, the group gains enormously, because the chip division sells most of its memory to outside buyers and only some of it ends up in its own phones. But the split explains why Samsung's revenue can fall short while profit beats. It also means the full report later this month is likely to show a gap between the two divisions, not a uniform boom. Chosun Ilbo's estimate, cited by TrendForce, that the chip division alone earned about 110 trillion won — more than the group total — implies the rest of the company lost money again. That is a press estimate, not a reported figure. The full results will confirm or correct it.

The same transfer runs through the rest of the industry. Every phone maker, PC maker and server builder that doesn't make its own memory is on the buying side of this price. That is the channel by which a Korean chipmaker's record becomes a margin question for companies listed far from Seoul. We traced one version of it in how the shortage pushed PC makers towards Chinese DRAM.

What it means for the instruments you trade

Samsung trades in Seoul, so it is in neither the S&P 500 nor the Nasdaq 100. The result still reaches US indices through two channels:

• Sellers' margins. Micron is a NAS100 and US500 constituent. A Samsung record confirms that memory prices were still rising through September, the input behind Micron's guide. What would undermine it is evidence of prices flattening, not evidence that they are high.
• Buyers' costs. Hyperscalers, server makers and device companies pay these prices. Memory is now a large and rising line in AI capital spending. Higher memory costs mean either more spending for the same compute or less compute for the same budget — and how the market values that spending depends on how it converts into revenue.

In Seoul, the channel is more direct. Samsung and SK hynix together are a very large share of the KOSPI, so the Korean index trades in large part on memory pricing. Our running coverage of Samsung, SK hynix and the KOSPI follows how that concentration behaves when sentiment turns.

The currency channel to the eight majors is thin. The won is not one of them, and a Korean earnings release does not move Federal Reserve expectations. The only link is through broad risk appetite — one of the five factors the meter scores for USD, JPY, CHF and the commodity currencies — and one company's quarter is too small an input to register there by itself. See how the factors are framed on the about page.

What would change the picture

What has to be true: memory prices have to keep rising, or at least stop falling, for these margins to hold. Margins this wide come from a shortage, not from a structural cost advantage. When fab capacity catches up with demand, the same fixed-cost leverage that turned price increases into profit works in reverse.

What is already priced: the consensus had already put the quarter above 100 trillion won, so the record itself was expected. The surprises were the size of the profit beat and the revenue shortfall.

What to watch next: - The full Q3 report later in October. It will give the DS/DX split, HBM commentary and whether the phone division lost money for a second quarter. - Contract prices into Q4. TrendForce expects DRAM contract prices to rise 10–15% quarter on quarter in the fourth quarter and NAND 15–20%. A slowdown in those increases would be the first sign the cycle is maturing. - The won. Further strength trims reported profit mechanically, even with dollar prices unchanged. - Buyer behaviour. Device makers raising retail prices, cutting memory per device or switching suppliers are signs that the cost side is starting to push back.

None of these is a forecast. They are the variables on which the next number depends.

Educational macro context only — not investment advice.