Kioxia, the Japanese NAND flash maker, reported another record quarter on October 30, 2026, and the market still marked it down. Operating profit for July to September was about ¥1.995 trillion on AI data center demand, but the company's forecast for the next quarter came in roughly 15 percent below analyst expectations. A headline circulating alongside the results says Kioxia has sold out its entire 2027 NAND production capacity. We could not confirm that wording.
This post is a claimed-versus-verified read. Figures below come from outlets reporting on the results and from Kioxia's July disclosures as summarized by TrendForce and Reuters; we could not open the October results filing itself, which Kioxia's news page lists as "Consolidated Financial Results for the Six Months Ended September 30, 2026" and a separate notice on shareholder return policy and first dividends. Treat the numbers as reported, not audited by us.
TL;DR: Kioxia's quarter
| Item | Reported figure | Status |
|---|---|---|
| July to September revenue | About ¥2.48 trillion, up 40.3% on the prior quarter | Reported by outlets citing the results |
| July to September operating profit | About ¥1.995 trillion, up 57.1%, vs. ¥85.9 billion a year earlier | Reported; slightly above a roughly ¥1.95T estimate |
| Prior guidance (July) | ¥1.89 trillion operating profit | Confirmed by Reuters |
| October to December forecast | About ¥2 trillion operating profit, revenue about ¥2.5 trillion | Reported; about 15% below a ¥2.35T estimate |
| Interim dividend | ¥90 per share, described as the first | Reported; notice listed on Kioxia's site |
| "2027 capacity sold out" | Not found | Unverified; Kioxia said demand should outpace supply |
What did Kioxia report?
According to reports on the results, revenue for the fiscal second quarter rose 40.3 percent from the previous quarter to about ¥2.48 trillion, while operating profit rose 57.1 percent to about ¥1.995 trillion. That is roughly 23 times the ¥85.9 billion a year earlier, which fits the headline that profit is up more than twenty-fold. It was a little above the ¥1.95 trillion average estimate that Reuters cited for the July guidance.
Kioxia had previewed the quarter on July 31, when it said it expected ¥1.89 trillion, versus an LSEG average estimate of ¥1.95 trillion, after April to June operating profit of ¥1.27 trillion. At the same time it announced a buyback of up to ¥800 billion for up to 30 million shares and a three-for-one stock split effective October 1, per Reuters via MarketScreener. The October results add an interim dividend of ¥90 per share, which the company's notice on first dividends appears to cover.
Why is the forecast called a miss?
For October to December, reports put Kioxia's guidance at about ¥2 trillion in operating profit, which is flat on the quarter just reported and about ¥350 billion below an estimated ¥2.35 trillion. One summary says the company expects strong data center demand to be offset by a negative exchange-rate effect, leaving results similar to Q2.
A miss against consensus does not mean demand fell. It means the pace of growth the market priced in has slowed from the 40 to 76 percent sequential jumps of recent quarters. In July, TrendForce reported that Kioxia's April to June revenue of ¥1.767 trillion was up 76.2 percent on the quarter, with an 80 percent gross margin, and noted Nikkei's point that net profit fell short of consensus then, too.
A stacked memory chip beside a gauge pointing high, representing Kioxia NAND memory profit margins
Is 2027 really "sold out"?
This is the part we could not verify. The phrasing "sold out its entire 2027 NAND production capacity" implies every wafer is under contract for the year. What Kioxia has actually said, as summarized by TrendForce after the July results, is narrower: NAND demand continued to outpace supply, average selling prices rose about 70 percent quarter on quarter in April to June, bit shipments grew only in the low single digits, and the company "expects demand to continue outpacing supply in calendar 2027".
Kioxia also said it is on track to secure 50 percent long-term-agreement (LTA) volume coverage for calendar 2028 with key customers. TrendForce, citing Newspim, adds that Kioxia discloses less about its LTAs than Samsung and SK hynix, which give details such as customer counts, terms and advance payments. So we know demand exceeds supply and that LTAs are growing; we do not have a disclosed share of 2027 output that is contracted.
| Possible meaning of "sold out" | Evidence we found |
|---|---|
| Demand exceeds supply in 2027 | Yes: Kioxia said so in July (TrendForce) |
| Customers have contracted most 2027 volume | Not disclosed |
| 100% of 2027 capacity is committed | Not found |
| LTA coverage of 50% for 2028 is the target | Yes: reaffirmed in July |
If you see the sold-out line, check whether it quotes the filing or a presentation slide. We found neither.
What does this mean for AI builders?
NAND is the storage behind training data lakes, checkpoints, vector databases and the SSD tiers that feed GPU clusters. Kioxia said in July that agentic AI is driving strong NAND demand and that data center and related sales reached ¥1.1747 trillion in April to June, more than five times a year earlier. For teams buying hardware, the practical consequence is price and lead time, not model quality.
- Expect high SSD prices through 2027. If demand outpaces supply, quoted prices for enterprise SSDs and the servers that include them are unlikely to ease soon.
- Watch the second half of 2027. TrendForce projects NAND to enter a looser supply environment in the second half of 2027 as new capacity comes online while consumer demand stays weak. That is a forecast, not a Kioxia statement, and it points the other way from the "sold out" framing.
- Plan contracts, not spot buys. Samsung and SK hynix have been pushing long-term deals; our posts on Samsung's Q3 record and 70 percent long-term contract claim and SK hynix's record quarter and shortage through 2030 show how those claims were framed.
A wafer disc in front of a tall chip factory block, representing NAND flash supply capacity
How does this fit the memory cycle?
Memory profits have swung wildly for decades, which is why the question analysts ask is always when the cycle turns. Our explainer on 65 years of DRAM, HBM and NAND prices shows how prices have repeatedly overshot in both directions. The current upswing is unusual because AI infrastructure spending, rather than phones and PCs, sets demand at the margin, a point we also made when covering the Anthropic S-1 risk factors on HBM suppliers.
Supply is responding. Samsung and SK hynix are upgrading their China NAND plants, as covered in our Reuters-based post on Xi'an and Dalian, and the wider AI chip supply chain picture in October shows TSMC and others at record levels. More supply in 2027 would explain why the market is unwilling to extrapolate Kioxia's current growth.
What are readers asking?
Is a miss on guidance bad news for AI storage demand? Not by itself. Kioxia's forecast is flat on a record quarter, and the stated drag is currency, not orders. The signal to watch is average selling prices: after a roughly 70 percent quarter-on-quarter jump in April to June, further large increases are what the market was pricing in, and flat guidance suggests that pace is cooling.
Why compare Kioxia with Samsung and SK hynix? Both larger rivals report HBM and DRAM alongside NAND, so their results mix several products. Kioxia is a near pure-play on NAND, which makes its numbers a cleaner read on SSD demand from AI data centers, though it also means it has no HBM upside to cushion a NAND price correction.
Does this change what I pay for storage? Possibly, with a lag. Contract prices usually move a quarter or two after the memory makers' own results, so enterprise buyers should expect firm pricing into early 2027 unless the looser supply TrendForce projects for the second half arrives sooner.
Where can I check the primary numbers? Kioxia's news page lists the filing; Reuters and TrendForce summarize the July disclosures used above. Compare any headline figure against the filing's own table before repeating it.
What we checked, and what we could not
- Checked: Kioxia's news page for the existence of the results and dividend notices; Reuters' July guidance; TrendForce's summary of Kioxia's July statements on demand, ASPs and LTAs; outlets summarizing the October 30 numbers.
- Not checked: the October results filing and presentation themselves, any statement on 2027 contracted volume, and the currency assumptions behind the Q3 forecast.
What to watch next
Watch for the earnings presentation's LTA slide, which would show whether the 2027 coverage is disclosed, and for Micron, Samsung and SK hynix commentary on whether NAND capacity additions land in 2027. A clear contracted-volume figure would turn the sold-out claim into a verified fact; so far, the verified claim is that demand should outpace supply.
Bottom line
Kioxia's quarter is real and large, driven by AI data centers, and its guidance disappointed analysts rather than showing weakness. The "2027 sold out" line goes further than anything we found Kioxia say. Read it as shorthand for tight supply, not as a disclosed contracted share.
Related reading
-
Samsung Q3 2026: record profit and the long-term contract claim
-
SK hynix Q3 2026: record profit and memory shortage through 2030
Figures reflect public reporting as of October 30, 2026 and may be revised when the filing is reviewed.