For about thirty years the most reliable early warning in memory was a ratio you could compute in ten seconds: inventory divided by quarterly revenue. When it climbed, the makers were producing into weakening demand, and a price collapse followed within two or three quarters. It called the 2022 top. It called several before that.
At SK hynix (000660) that ratio is now 22.7%, down from 82.5% in the March quarter of 2025. On the old reading that would be the most bullish print in the company's history.
It isn't. The ratio has stopped measuring what it used to measure, and the reason is worth understanding because it changes which numbers tell you where this cycle is.
June-quarter revenue was ₩79,318.7bn against ₩22,232.0bn a year earlier. Up 256.8%.
June-quarter cost of revenue was ₩13,327.4bn against ₩10,248.6bn. Up 30.0%.
Sit with that gap. SK hynix sold roughly 30% more stuff, measured at what it cost to make, and collected three and a half times as much money for it. Gross margin went from 53.90% to 83.20%. Operating margin went from 41.44% to 76.33%.
Inventory is carried at cost. Revenue is carried at price. When price and cost diverge by a factor of nearly nine, any ratio with inventory on top and revenue on the bottom stops describing physical stock and starts describing the gross margin. Watching it fall from 82.5% to 22.7% and concluding that inventory has been drawn down is a category error — the numerator went up.
Inventory at 30 June was ₩17,985.7bn. A year earlier it was ₩13,408.3bn. It rose 34.1%.
Put inventory over cost of revenue instead, and you get days of stock in units the factory would recognise.
June quarter 2025: ₩13,408.3bn over ₩10,248.6bn, times 91 days. That's 119 days.
June quarter 2026: ₩17,985.7bn over ₩13,327.4bn, times 91 days. That's 123 days.
Four days of difference across a year in which revenue more than tripled. Physically, essentially nothing happened. The company is shipping about what it makes, and the extra ₩4,577.4bn of inventory on the balance sheet is roughly what you'd expect a business with 30% more cost throughput to carry.
Run the same calculation on the annual figures and you get the history that gives the measure its authority:
FY2018: 106 days. FY2019: 103. FY2020: 106. FY2021: 136. FY2022: 197. FY2023: 148. FY2024: 141. FY2025: 136.
FY2022 at 197 days is the build. SK hynix reported an operating loss of ₩7,730.3bn in FY2023, the year that followed. The metric did its job. Today's 123 days is the lowest reading since FY2020, and it is falling rather than rising.
There's a complication, and it undercuts the simple reading.
FY2019 showed 103 days — a lower number than today — and SK hynix earned operating income of ₩2,712.7bn that year, less than a twentieth of what it earns in a single quarter now. Low inventory days did not mean good times. They meant a company that had cut production hard into a slump.
So days of inventory measures supply-demand balance, not profitability, and a low number is consistent with either a shortage or a shutdown. What distinguishes them is what's happening to price, and the price here is unambiguous: 83.2% gross margin on the June quarter.
Combine the two and the reading is a genuine shortage of something the customer badly wants, being cleared at close to the rate it's produced. That's HBM, and it is not the DRAM commodity cycle wearing a new hat. Commodity DRAM has never sustained an 83% gross margin.
If inventory-to-sales is dead, what replaces it?
The first thing is days of inventory on cost. If that number moves from 123 toward 150 over two quarters, the shortage is easing whatever the revenue line says, and it will move before pricing does.
The second is receivables. Trade receivables at 30 June were ₩47,821.4bn against ₩13,125.2bn a year earlier, up 264%. That sounds alarming until you set it against revenue growth of 256.8%. Receivables were 59.0% of quarterly revenue a year ago and 60.3% now. Nothing is being pushed into the channel on soft terms — this growth is being collected.
The third is capex, which is the supply side. ₩10,671.4bn in the June quarter against ₩4,331.5bn a year earlier, up 146%. Property, plant and equipment reached ₩88,889.1bn, up 37.9% year over year, and the company filed two new facility investment notices on 6 August. That capacity arrives in 2027 and 2028, and it arrives alongside everyone else's, which is the mechanism that has ended every memory shortage there has ever been.
One warning about the income statement. Second-quarter net income was ₩93,922.6bn, larger than the ₩60,542.6bn operating income, because finance income was ₩65,858.8bn against ₩1,719.4bn in the same quarter of 2025. Total equity rose ₩98,313.4bn in three months.
I don't know from the summary lines what that gain is, and neither does anybody quoting a trailing price-to-earnings ratio for this company. Trailing net income including it is about ₩162,112.0bn, which would put the shares at roughly 7.5 times earnings and is not a real number. Trailing operating income is ₩128,705.9bn, and the ₩1,222,844.2bn market value is 9.5 times that. Use the second one.
The page carries an earnings per share figure of ₩58,796 and book value per share of ₩359,607, which put the stock at 28.5 times and 4.66 times respectively. The earnings figure corresponds to FY2025 rather than the last four quarters, so it lags badly in a period like this.
The most serious is competition, and it isn't visible in any ratio above. Chinese memory capacity aimed at HBM changes the supply picture on a timeline shorter than most people assume, and I've written about the estimate that CXMT could serve a large share of China's own HBM demand by 2028. A shortage that ends because someone else builds is still a shortage that ends.
The second is that 76.33% operating margins invite exactly the capital that destroys them, including SK hynix's own — the capex line above is the company voting on how long this lasts.
The third is concentration. An 83.2% gross margin on a product with a small number of buyers is a negotiated margin, not a market-clearing one, and negotiations get renewed.
The September quarter's inventory and cost of revenue lines, in that order. If days of stock stay near 120, the balance is intact regardless of what happens to the headline revenue growth rate, which cannot possibly repeat 256.8%.
And the note explaining the ₩65,858.8bn of finance income in the third-quarter report. Until that's on the page, the only earnings number worth quoting for this company is the operating one.
kstock reads DART every morning and writes up what moved — the contract, the buyback, the number that does not add up. The daily post and a Saturday roundup, by email.