SK Hynix Capex Up 2.52x YoY, 23.1% of Operating Cash
SK Hynix, Samsung, Micron and Kioxia put 5.6–27.3% of Q3 operating cash flow into equipment, below a year earlier. Spending rose at three; Kioxia's fell 7.2%.
South Korean memory chipmaker SK Hynix paid KRW 12.67 trillion in cash for plants and equipment in July–September 2026, recorded on its line for acquisition of property, plant and equipment (PP&E). That's 2.52 times the figure for the same quarter a year earlier (PRISM calculation). Operating cash flow — the cash its business brought in — was KRW 54.781 trillion over the same three months. Divide the equipment spending by that operating cash and you get 23.1%, below the 35.0% of a year earlier and above the 16.2% of the previous quarter (PRISM calculation). The figures sit in the cash flow summary on page 15 of the third-quarter earnings presentation the company filed on DART, the electronic disclosure system run by South Korea's Financial Supervisory Service.
We held the same yardstick up to Samsung Electronics, US chipmaker Micron and Japan's Kioxia. The sources are Samsung's third-quarter earnings presentation, the earnings press release Micron filed with the US Securities and Exchange Commission (SEC), and the kessan tanshin (決算短信), or earnings summary, that Kioxia filed on TDnet, the Tokyo Stock Exchange's disclosure system. All four companies' ratios were lower than in the same quarter a year earlier, and three of the four paid more cash for equipment. Against the previous quarter, the ratio rose at SK Hynix and Samsung (consolidated) and fell at Micron and Kioxia, so the direction splits. This piece takes in turn the arithmetic of a share that shrank from a year earlier while spending grew at three companies, and the causes the filings don't confirm.
Is 23.1% high? This piece has three yardsticks to work with: the halfway (50%) mark, the same quarter a year earlier, and the previous quarter. Its data can't show an industry average or a multi-year trend. A trend only emerges once the same two lines are measured again over the coming quarters and stacked up. There's one point of contrast. Five US Big Tech companies that buy the chips spent $31.8 billion more on capex than they brought in from operations in the same third quarter (PRISM, November 3; combined for the five, each company's third quarter of 2026; PRISM calculation). That's a gap between combined dollar amounts, a different yardstick from this piece's ratio. Still, the five buyers together spent more than their operating cash, while the four memory sellers put 5.6–27.3% of theirs into equipment.
Four companies, measured on the same two lines
There's one yardstick: cash paid for equipment divided by operating cash flow. Because both lines are in the same currency, amounts stay in each company's own currency, with no conversion and no ranking. In the table, "operating cash multiple" and "capex multiple" show how many times each line grew from the same quarter a year earlier.
Table 1. Four memory makers' Q3 cash paid for equipment ÷ operating cash flow
| Company | Period | Operating cash flow (local currency) | Cash paid for equipment (line name) | Ratio (PRISM calc.) | Year-ago ratio | Operating cash multiple | Capex multiple |
|---|---|---|---|---|---|---|---|
| SK Hynix | Jul–Sep 2026 | KRW 54.781 trillion | KRW 12.67 trillion (유형자산의 취득, acquisition of PP&E) | 23.1% | 35.0% | 3.81x | 2.52x |
| Micron | May 29–Sep 3, 2026 (14 weeks; year-ago quarter 13 weeks) | $43.973 billion | $11.11 billion (Expenditures for property, plant, and equipment) | 25.3% | 98.7% | 7.67x | 1.96x |
| Kioxia | Jul–Sep 2026 (Apr–Sep cumulative minus Apr–Jun) | ¥1,513.243 billion | ¥84.12 billion (有形固定資産の取得による支出, purchase of PP&E) | 5.6% | 79.5% | 13.27x | 0.93x |
| — Below: consolidated basis (includes smartphones and home appliances) — | |||||||
| Samsung Electronics (consolidated) | Jul–Sep 2026 | KRW 98.6 trillion | KRW 26.9 trillion (유형자산의 증가, increase in PP&E) | 27.3% | 47.8% | 4.36x | 2.49x |
Sources: SK Hynix third-quarter earnings presentation, p.15 (DART); Samsung Electronics third-quarter earnings presentation, "Appendix 4: Summary Cash Flow"; Micron 8-K Ex. 99.1 non-GAAP reconciliation; Kioxia kessan tanshin (July 31, 2026 and October 30). Ratios and multiples are PRISM calculations.
- Samsung: It doesn't disclose operating cash flow for memory alone, so the consolidated figures, which also cover smartphones, home appliances and displays, sit separately below the divider. Its equipment line is called "increase in PP&E" (유형자산의 증가), and the IR summary alone doesn't show whether that's gross acquisitions or net of disposals. Because the figures are rounded to KRW 0.1 trillion, the ratio can move within ±0.1–0.3 percentage points and the capex multiple within 2.47–2.51x (PRISM calculation).
- Micron: This quarter covers the 14 weeks from May 29 to September 3, 2026, and the year-ago quarter is the 13 weeks ended August 28, 2025. On a per-week basis, operating cash is 7.13x and capex 1.82x. Measured net of government incentive receipts ("Investments in capital expenditures, net"), the ratio is 24.5%, against 86.0% a year earlier (all PRISM calculations). The table uses the gross figure.
- Kioxia: It discloses only April–September cumulative figures, so the July–September values are a PRISM calculation that subtracts April–June from the cumulative total. Measured on the April–September cumulative figures without subtracting, the ratio is 5.7%, against 81.7% for the same period a year earlier (PRISM calculation).
The highest number in the ratio column, 27.3%, is Samsung's consolidated figure, which mixes in businesses outside memory. The three pure-play memory makers range from 5.6% to 25.3%.
Three questions about memory makers' capex ratios
Questions about this table come down to three.
- What share of the cash memory companies brought in from operations in the third quarter went back into factories and equipment? Is that up or down from a year ago?
- Did the ratio fall because equipment spending shrank, or because operating cash grew faster? Apart from equipment, where did big cash outflows go in the same quarter?
- I hold memory stocks or a chip ETF. Which lines on the cash flow statement should I look at in the next earnings report?
There are three pushbacks, too. Is it only a denominator effect? Did a small year-ago quarter exaggerate the decline? Can one quarter's figures carry the conclusion? The next section answers them in order.
The arithmetic of a lower ratio, and the causes PRISM couldn't confirm
No cause is established. What PRISM confirmed goes as far as the arithmetic breakdown. Part A is that breakdown. Part B covers company statements that clash with a "they cut investment" reading, and Part C looks at other cash that left in the same quarter.
A. Breakdown: the denominator grew by a bigger multiple than the numerator
A ratio is a fraction. If the numerator (cash paid for equipment) grows less than the denominator (operating cash), the ratio falls. The two multiple columns in Table 1 show by how much.
At SK Hynix, operating cash grew 3.81x from KRW 14.368 trillion a year earlier, and cash paid for equipment grew 2.52x from KRW 5.033 trillion. Samsung's consolidated figures show 4.36x for operating cash and 2.49x for equipment. Micron's are 7.67x and 1.96x (14 weeks against 13; per-week figures are in the Table 1 notes). All three paid more for equipment, and operating cash grew by an even bigger multiple.
Kioxia went the other way. Its July–September operating cash grew 13.27x. Cash paid for equipment came in at 0.93x, a 7.2% drop (PRISM calculation). The denominator got bigger while the numerator got smaller. In the cash flow overview of its kessan tanshin, the company gave no reason for the lower equipment spending.
The arithmetic in the two paragraphs above answers the first pushback, about a denominator effect.
The cash flow figures alone can't say why operating cash grew that much. In the narrative section (overview) of its kessan tanshin, Kioxia named higher pre-tax profit as the main driver of the April–September increase in operating cash. That sentence doesn't address July–September on its own. A single operating cash flow line can blend prices, volumes and money customers paid in advance, and the table doesn't break them apart. Micron recorded $12.747 billion in customer deposits under financing activities, on the "Proceeds from customer contract liability deposits" line. That's a full fiscal 2026 figure, so the fiscal fourth quarter's share is unknown, and it isn't part of the operating cash used as this piece's denominator. In its 10-K, Micron described these deposits as payments "primarily intended to secure future production capacity."
SK Hynix's third-quarter earnings presentation said its long-term supply agreements (LTAs) include "substantive enforcement mechanisms such as deposits" (PRISM translation). The summary cash flow statement doesn't show whether those deposits landed in third-quarter operating cash. So there's no basis for reading all of the growth in operating cash as sales receipts.
B. The counter-reading: payment timing and next year's plans
Reading the lower ratio as "they cut capex" runs into the companies' own words. SK Hynix's third-quarter earnings presentation states: "2027 investment: substantial increase expected versus 2026" (PRISM translation). It gave the purpose as "equipment investment to expand the share of leading-edge processes and ramp up Yongin Phase 1 Fab, and full-scale domestic and overseas infrastructure investment" (PRISM translation).
Micron's fiscal 2026 10-K lays out estimates for next year's equipment spending. Net of government incentive receipts, it puts fiscal Q1 2027 at "approximately $11.5 billion" and the first half at "approximately $25 billion," with the second half "higher." The Q1 estimate is above the actual net figure of $10.774 billion for fiscal Q4 2026. That comparison is net against net, and fiscal Q4 2026 ran 14 weeks.
Micron attached a caveat to the same passage: "Actual amounts for 2027 will vary depending on market conditions and may vary from quarter to quarter due to the timing of expenditures and proceeds from government incentives." The equipment line on a cash flow statement is booked when the payment goes out. So a single quarter's ratio can swing with the timing of equipment bills.
This evidence has limits, too. Both companies' statements are plans and estimates. Actual spending can be checked only in the next quarter's cash flow statements. Nor do the filings show that any of this quarter's payments slipped into the next. PRISM wasn't able to check Samsung's or Kioxia's statements on next year's investment amounts. So this piece keeps the "capex cut" label off the lower ratio and narrows its conclusion to this quarter's figure. That's the answer to the third pushback, about relying on one quarter. A supporting point: widen Kioxia to the April–September cumulative period and its ratio is still 5.7% (81.7% a year earlier).
C. Other cash that went out the same quarter
A cash flow statement doesn't trace where cash not spent on equipment ended up. What can be checked is the other large outflows listed in the same quarter, on the same statement.
SK Hynix is the only company where PRISM confirmed this with same-quarter figures. On its third-quarter summary cash flow statement, KRW 30.845 trillion went out on the 장기운용상품의 증감 (change in long-term investment products) line. That's the line name in the company's summary statement, which doesn't say what the money was invested in. The amount is 2.43x the KRW 12.67 trillion paid to acquire PP&E in the same three months (PRISM calculation). Dividends paid came to KRW 272 billion. The 자사주 취득/처분 등 기타 (treasury share acquisition/disposal and other) line shows an inflow of KRW 10.308 trillion, so it can't be read as share buybacks. Cash at quarter-end, including short-term financial instruments, stood at KRW 114.46 trillion (per a footnote in the presentation).
For the other companies, the periods don't line up. Micron's "Purchases of available-for-sale securities," at $34.871 billion, is a full fiscal 2026 figure. Kioxia's purchase of other financial assets (その他の金融資産の取得による支出), at ¥78.22 billion, is an April–September cumulative figure, and the company attributed it to buying shares in Taiwan's Nanya Technology. For both, the current quarter's share is unknown. PRISM didn't check Samsung's same-quarter outflows into financial assets.
Scale check: the previous quarter and alternative readings
The second pushback is about the base. The year-ago quarter was one in which operating cash was small at Micron ($5.73 billion) and Kioxia (¥114 billion). So the drops from 98.7% to 25.3% and from 79.5% to 5.6% are heavily shaped by that base. This piece sets the size of the drop aside as evidence and switches the baseline to the previous quarter.
Table 2. Previous-quarter ratio vs. this quarter's ratio
| Company (previous quarter) | Previous-quarter ratio (PRISM calc.) | This quarter's ratio | Direction |
|---|---|---|---|
| SK Hynix (Apr–Jun 2026) | 16.2% | 23.1% | Up |
| Samsung Electronics, consolidated (Apr–Jun 2026) | 13.4% | 27.3% | Up |
| Micron (quarter ended May 28, 2026) | 30.8% | 25.3% | Down |
| Kioxia (Apr–Jun 2026) | 5.9% | 5.6% | Down |
The underlying figures come from the previous-quarter columns of the same documents used for Table 1.
Quarter-end dates and calculation methods differ by company. Micron's previous quarter ended May 28, and Kioxia's current-quarter figure is derived by subtraction from cumulative data. So the 2–2 split shouldn't be read as a Korea-versus-US/Japan divide. That's why this piece limits "lower" to "lower than the same quarter a year earlier." "Below half" holds on both baselines: even the highest previous-quarter ratio is Micron's 30.8%.
The year-on-year direction holds under other readings as well. On top of the two readings in the Table 1 notes, here's a third. Subtract government grant receipts (政府補助金による収入) from Kioxia's April–September cumulative cash paid for equipment, and the ratio is 4.8%, against 62.7% for the same period a year earlier (PRISM calculation). On every reading, this quarter's ratio is below the year-ago level and under 50%.
The answers, one by one
Question 1. What share of operating cash went back into equipment, and is that up from a year ago? All four companies stayed below half, and all four were lower than in the same quarter a year earlier (Table 1). Against the previous quarter, SK Hynix and Samsung are actually higher (Table 2). Whether the level counts as high or low can be answered only against these three yardsticks (the 50% line, the year-ago quarter, the previous quarter). Industry averages and multi-year trends are beyond this piece's data. As a point of contrast, the five Big Tech chip buyers together spent more cash on capex than they brought in from operations in the same third quarter.
Question 2. Did the ratio fall because equipment spending shrank, or because operating cash grew faster? Kioxia is the only one whose cash paid for equipment fell. At the other three, equipment spending grew too, and operating cash grew by a bigger multiple (Part A). That's arithmetic. The cash flow figures alone can't explain why operating cash grew that much (Kioxia named higher pre-tax profit as the main driver of its April–September increase). Among large outflows other than equipment, PRISM confirmed a same-quarter figure only at SK Hynix: cash that left on its 장기운용상품의 증감 (change in long-term investment products) line that quarter exceeded its equipment purchases (Part C).
Question 3. Which lines should I look at in the next earnings report? Two lines on the cash flow statement: operating cash flow and the equipment purchase line. Then lay the companies' own plan statements over them. Those are SK Hynix's "substantial increase expected" for 2027 investment (PRISM translation) and Micron's estimate of "approximately $11.5 billion" in net capex for fiscal Q1 2027. SK Hynix, Samsung and Micron haven't announced their next earnings dates yet, and PRISM couldn't confirm Kioxia's. The line names and how to measure them are laid out in the PRISM Insight below.
What this quarter's filings support
In July–September 2026 (May 29–September 3 for Micron), the four memory makers paid 5.6–27.3% of their operating cash flow for equipment (the 27.3% is Samsung's consolidated figure, which includes smartphones and home appliances), a smaller share than in the same quarter a year earlier (35.0–98.7%). Over the same period, cash paid for equipment at SK Hynix, Samsung and Micron grew to 1.96–2.52x year-ago levels (14 weeks against 13 for Micron), and fell 7.2% at Kioxia alone (PRISM calculation). Against the previous quarter, the ratio rose at SK Hynix and Samsung (consolidated) and fell at Micron and Kioxia (Table 2). "Lower" here means lower than the same quarter a year earlier.
The strength of the evidence varies by layer. PRISM checked the underlying figures directly in the four companies' original filings. The ratios and multiples, and the breakdown of each ratio's change into denominator and numerator multiples, are PRISM calculations from those figures. The cause of the lower ratios isn't confirmed. SK Hynix's and Micron's plans for next year are confirmed in company statements; whether they get carried out isn't known yet. These numbers can't support any implication for memory supply or prices, so this piece doesn't draw one.
Some findings would change the conclusion. The SK Hynix and Samsung rows rely on IR summary tables. The DART quarterly reports are due by the November 14 statutory deadline. If their cash flow statements differ from the summaries and either company's ratio comes out higher than a year earlier, PRISM drops the sentence "all four were lower." If Samsung's quarterly report shows its 유형자산의 취득 (acquisition of PP&E) line differing from the IR's 유형자산의 증가 (increase in PP&E) such that the figure didn't grow from a year earlier, "grew at three companies" becomes "two."
What to watch next
- DART quarterly reports (statutory deadline November 14): SK Hynix's and Samsung's third-quarter cash flow statements. Check whether the values match the IR summaries, and how Samsung's 유형자산의 취득 (acquisition of PP&E) differs from the IR's 유형자산의 증가 (increase in PP&E). If the values differ, PRISM corrects Table 1.
- Micron's fiscal Q1 2027 results: The company hasn't announced the date. It's the first quarter in which actual net capex can be compared net-to-net with the 10-K estimate.
- SK Hynix's and Samsung's fourth-quarter results: Neither has announced a date. Samsung's earnings presentation flagged that it would present its next shareholder return policy and the final shareholder return amount from its remaining 2026 funds at "a board meeting in late January 2027" (PRISM translation). That's separate from any results-date announcement.
- Kioxia's next kessan tanshin: Subtracting the April–September cumulative figures from the April–December cumulative figures gives October–December. PRISM couldn't confirm the release date.
Related PRISM coverage: SK Hynix's operating margin in our October 28 story, and how HBM (high-bandwidth memory) is built in our November 7 explainer.
Figures are based on each company's third-quarter filings (September–October 2026), checked by PRISM on November 9. English renderings of Korean- and Japanese-language quotes and line names are PRISM translations (originals at each link); Micron quotes and line names are in the original English.
This content is AI-generated based on source articles. While we strive for accuracy, errors may occur. We recommend verifying with the original source.
Related Articles
Meta has increased its El Paso AI data center investment more than sixfold, from $1.5B to $10B, targeting 1GW capacity by 2028. What this means for investors, competitors, and the AI infrastructure race.
SK Hynix has filed confidentially with the SEC for a potential U.S. ADR listing in 2026, targeting up to $10 billion. Here's what it means for investors, competitors, and the global memory race.
CoreWeave's aggressive $30-35B capex plan for 2026 spooked investors, raising questions about the AI infrastructure company's debt-heavy business model and customer concentration risks.
Nintendo shares plunge 10% as memory chip shortages driven by AI boom threaten gaming console margins. Switch 2 faces critical test in 2026.
Thoughts
Share your thoughts on this article
Sign in to join the conversation