The assumption running through every memory desk this year is that a supplier in a shortage will take every yen the market offers. Kioxia has just broken it. On 9 September 2026 the Japanese NAND maker’s president and chief executive, Hiroo Ota, told Bloomberg that memory prices have risen enough, and that the company will work to stop further increases that would damage long-term artificial intelligence demand. He said it after a quarter in which Kioxia’s own NAND average selling prices climbed roughly 70% sequentially. In the same interview he ruled out deeper manufacturing ties or a merger with SK Hynix, the company that holds bonds convertible into a 14.19% stake in his own business. This is not a supplier pleading poverty. It is a supplier with historic pricing power choosing not to use all of it.
The information that matters here is not the rejected SK Hynix tie-up. It is that Kioxia has become the first major memory supplier to publicly cap its own pricing power in the middle of an industry-wide shortage – a supplier-led price ceiling, announced unilaterally, with no cartel, no regulator and no customer revolt forcing it. Every other seller in NAND and DRAM has spent 2026 arguing that memory is no longer a commodity and that scarcity is structural. Kioxia has now put a ceiling on the argument from the inside. For Micron, SanDisk, Western Digital, Samsung and SK Hynix, that changes the read-across from “how high can prices go” to “who blinks on the next negotiation”. For hyperscalers, handset makers and module houses, it is the first evidence that the top of this cycle may be set by supplier restraint rather than by demand destruction.
Key facts
- Kioxia NAND average selling prices rose roughly 70% quarter-on-quarter in the April-June 2026 quarter, on low-single-digit bit shipment growth (TrendForce, 31 July 2026).
- Kioxia’s fiscal Q1 revenue reached ¥1.767 trillion, up 76.2% sequentially and 415.5% year on year, with a gross margin of 80% and an operating margin of 75% (TrendForce, 31 July 2026).
- SK Hynix holds bonds convertible into a 14.19% stake in Kioxia, making its special-purpose vehicle the largest single shareholder since August (The Korea Herald, 2 September 2026).
- Industry NAND flash contract prices rose 55-60% in Q1 2026 and 70-75% in Q2 2026, but TrendForce forecasts only 10-15% for Q3 2026 (TrendForce, 3 July 2026).
- Gartner data cited in SK Hynix’s US listing filing forecasts NAND revenue growing from $68 billion in 2025 to $341 billion in 2027, a 123.7% compound annual rate (Blocks & Files, 9 July 2026).
- Kioxia and SanDisk committed on 27 August 2026 to invest more than $31 billion (about ¥5 trillion) in the Yokkaichi and Kitakami plants through 2032 (Kioxia, 27 August 2026).
- Kioxia shares closed at ¥57,000 on 9 September 2026, down 0.71%, against a 52-week range of ¥3,130 to ¥112,700 (StockAnalysis, 9 September 2026).
What actually happened, and why a supplier would do this
Ota, who took the chief executive role in April 2026, made two separate statements that markets have conflated. The first was structural: no deeper manufacturing alliance with SK Hynix, no merger, no three-way arrangement. The second was commercial: NAND prices have gone far enough, and Kioxia will manage the next round of negotiations accordingly.
On the corporate question, Ota was blunt about the practical obstacles. Kioxia’s fabs are not wholly its own: the Yokkaichi and Kitakami sites operate inside a joint venture framework with SanDisk that dates back more than 25 years and was extended in January 2026 through December 2034. Antitrust is the second obstacle, since any production-level combination would face review in multiple jurisdictions at a moment when memory pricing is politically sensitive. Ota also disclaimed knowledge of what prompted SK Group chairman Chey Tae-won to raise the idea publicly a week earlier.
The pricing statement is the more consequential one. Kioxia’s fiscal first quarter was among the most profitable in NAND history: an 80% gross margin, a 75% operating margin and ¥1.1747 trillion of data centre revenue in a single quarter. Bit shipments grew in the low single digits, so almost all of the revenue growth was price.
That configuration invites a customer backlash. When margin expansion comes from price rather than volume, the buyer sees a transfer of value rather than a shared benefit, and starts engineering the supplier out. Kioxia’s reasoning, as reported, is that hyperscaler budgets are finite and that pricing past that point destroys the demand curve it wants to sell into for the rest of the decade.
“We can’t just say, ‘Well then, let’s make it three companies,'” Ota said of the hypothetical Kioxia-SK Hynix-SanDisk arrangement, adding on Chey’s remarks: “We have no idea what prompted (Chey) to say what he did.” Hiroo Ota, President and Chief Executive Officer at Kioxia Holdings Corp, had used the same long-horizon framing two weeks earlier when announcing the SanDisk capex plan, calling it a commitment “to contributing to the advancement of an AI-driven society”.
Supplier and buyer response: who has said what
The suppliers have not followed
As of 9 September 2026, no other major memory supplier has publicly matched Kioxia’s pricing language. Samsung Electronics, SK Hynix and Micron Technology have issued no statement capping their own price increases. That silence is itself the story: a unilateral ceiling only functions as a ceiling if rivals decline to sell into the gap it creates.
The public positioning of the others runs the other way. Sanjay Mehrotra, Chief Executive Officer at Micron Technology, has spent 2026 arguing that memory has stopped behaving like a commodity and that tight conditions persist beyond calendar 2027, and the company has signed five-year strategic customer agreements with 16 customers. That is a supplier locking in duration, not moderating price. Our analysis of Micron’s DRAM share position against SK Hynix as prices cool set out how much of the current multiple depends on that view holding.
SanDisk, Kioxia’s own joint venture partner, has converted price into contract instead. It disclosed on 18 August 2026 ten new business model agreements with eight data centre and edge customers, covering more than 50% of fiscal 2027 bits and roughly two-thirds of fiscal 2028 bits, with a weighted average duration above four years and $16.5 billion of guarantees against customer default.
“The total expected revenue from all our NBMs we have signed is a minimum of $93.9 billion, assuming floor pricing,” said Luis Visoso, Executive Vice President and Chief Financial Officer at SanDisk. That structure is the mirror image of a ceiling: it establishes a floor. The detail sits in our SanDisk investor day preview and our longer SanDisk scenario work.
The buyers have said nothing, and that is rational
Apple, the hyperscalers and the large Chinese handset makers have not publicly welcomed Ota’s remarks, and would be unwise to. Praising a supplier for restraint converts an informal signal into a negotiating expectation, which is the fastest way to make the supplier withdraw it.
The buyer-side pressure is nonetheless documented. Gartner estimates memory prices will rise about 130% by the end of 2026 versus 2025 levels, pushing smartphone prices up roughly 13% and cutting smartphone shipments by 8.4%. IDC’s forecast is harsher, at a 12.9% decline to 1.12 billion units. On an entry-level handset with 6GB of LPDDR4X and 128GB of storage, memory reached 43% of the total bill of materials in Q1 2026, against 15-20% on a mid-range device and 10-15% on a flagship.
Module houses and second-tier OEMs are the exposed link. They buy on spot or short contracts and cannot sign four-year floor-price agreements. A supplier-led ceiling helps them more than it helps Apple, which has the balance sheet and volume to absorb the increase.
“SK hynix appears ready to use the profits and confidence gained from its HBM leadership to pursue the top position in the broader memory market,” said Lee Jong-hwan, professor of system semiconductor engineering at Sangmyung University, quoted by The Korea Herald. That is the competitive logic Kioxia is declining to join.
Market impact: what the data actually shows
The most important point about Ota’s ceiling is that the market had already started building one. TrendForce’s contract series shows NAND rising 55-60% in Q1 2026 and 70-75% in Q2, then decelerating to a forecast 10-15% in Q3 on base effects and weak consumer demand. Kioxia is not capping a still-accelerating market. It is putting a public label on a deceleration already visible in the contract data.
Equity markets have priced the boom far more aggressively than the price series alone justifies. The dispersion between the names is the tell.
| Company | NAND share, Q2 2026 (shipments) | Stated pricing posture | Share price, 8-9 Sep 2026 |
|---|---|---|---|
| Samsung Electronics | 25% | No public cap; roughly 50% capacity expansion planned for 2026 | n/a |
| SK Hynix | 22% | No public cap; chairman open to Kioxia joint production; Nasdaq ADS listing 10 July 2026 | n/a |
| Kioxia (285A.T) | 14% | Prices “have risen enough”; ASPs +70% QoQ in Apr-Jun 2026 | ¥57,000, -0.71% |
| SanDisk (SNDK) | JV partner at Yokkaichi and Kitakami | Floor pricing via 10 NBM deals worth $93.9bn minimum | $1,737.99, -0.12% (+3,192% YTD) |
| Micron (MU) | DRAM and NAND | No public cap; 16 five-year strategic customer agreements | $1,000.26, -1.61% (+750% YTD) |
| Western Digital (WDC) | HDD only after the SanDisk separation | Not a NAND price setter | $477.30, +2.14% |
Share prices are last close per StockAnalysis, 8 September 2026 for the US names and 9 September for Kioxia. Q2 2026 shipment shares are Asahi Shimbun data cited by The Korea Herald.
Three readings follow. Western Digital is now structurally insulated: after the SanDisk separation it is a hard disk drive business, so a NAND ceiling reaches it only through data centre storage budgets, which is why our Western Digital scenario analysis treats it as a different cycle.
SanDisk is the most exposed to a credible ceiling in sentiment terms and the least exposed in cash terms. Floor-priced contracts protect the downside, but a market that stops believing in open-ended price escalation will not sustain a +3,192% year-to-date move on multiple expansion alone.
Micron sits at the intersection, the only large supplier with material DRAM, HBM and NAND exposure. DRAM contract prices are still forecast at 13-18% sequential growth in Q3 2026, well above NAND, so a NAND-specific ceiling does not automatically travel to DRAM. Our note on Samsung and SK Hynix rallying on GPT-6 demand shows how quickly the two product lines decouple in sentiment.
Antitrust and the regulatory overlay
Ota’s antitrust objection is real, and the arithmetic explains why. On Q2 2026 shipment shares, SK Hynix at 22% plus Kioxia at 14% would exceed Samsung’s 25%. A manufacturing-level combination would need clearance in Japan, South Korea, the United States, the European Union and, most awkwardly, China, which has previously used merger review as leverage in semiconductor matters.
The shareholding is already fenced. The vehicle through which SK Hynix holds its convertible position became Kioxia’s largest shareholder at 14.19% in August 2026, and a standstill agreement bars it from 15% or more of Kioxia’s voting rights without consent until 2028. That is a financial position deliberately capped below a control threshold, not a merger in waiting.
The wider legal weather is worse. On 25 June 2026, Samsung, SK Hynix and Micron were sued in the Northern District of California by 17 plaintiffs alleging they conspired to keep conventional DRAM artificially scarce while shifting capacity to HBM, with the complaint citing DRAM price increases of as much as 700% over four years. The industry has lost this argument before: Samsung paid a $300 million US fine in 2005 and SK Hynix $185 million.
That history creates an uncomfortable second reading of Ota’s statement. One supplier announcing that prices should stop rising is unilateral conduct and lawful; the same statement becomes evidence if rivals visibly follow it. Kioxia’s competitors therefore have a legal incentive, not merely a commercial one, to say nothing and price independently.
Export controls add a third layer. US restrictions target advanced NAND at 128 layers and above, fragmenting the market and raising compliance cost for multinational procurement teams. Kioxia began shipping 332-layer 3D NAND on its tenth-generation process this year, placing its leading product squarely inside the controlled band.
SK Hynix’s own US filing frames the constraint plainly: “Our future long-term growth depends to a significant extent on our ability to increase production capacity.” Capacity, not price restraint, is the stated strategy of the Korean incumbent.
What happens next
Prediction 1: the ceiling holds for NAND and fails for DRAM, by Q4 2026
Kioxia has no DRAM business, so its ceiling covers only the product where it sets prices. TrendForce already forecasts NAND at 10-15% sequential growth in Q3 2026 against 13-18% for conventional DRAM, and expects Q4 NAND increases to converge further under weak end demand and elevated inventory. Expect the Q4 2026 contract round to show NAND rising in single digits while DRAM continues in the teens. The gap, not the level, is the evidence that a supplier-led ceiling is operative.
Prediction 2: rivals answer with contract structure, not with price cuts, through H1 2027
Neither Samsung, SK Hynix nor Micron can match Kioxia’s statement without inviting the plaintiffs in the California case to characterise it as coordination. The available response is the SanDisk template: multi-year agreements with floor pricing, prepayments and default guarantees, which lock in revenue without any public statement about price direction. SanDisk has already covered more than half of fiscal 2027 bits this way. Micron’s 16 five-year agreements point the same direction. Watch for the number of such agreements to rise through the December 2026 and March 2027 reporting rounds.
Prediction 3: the SK Hynix stake gets resolved before the 2028 standstill expires
Chey Tae-won has already said publicly that SK would end its Kioxia investment if no cooperative relationship can be built. Ota has now said no such relationship is coming at the manufacturing level. A 14.19% position that generates no strategic return, is capped below 15% until 2028, and sits in a stock trading around half its 52-week high is a candidate for monetisation. A sale or partial unwind of the convertible position is the most likely resolution, and it would be a meaningful overhang event for Kioxia’s float.
“If SK hynix is to be included in Kioxia’s future strategy, we are always ready to become a partner,” Chey Tae-won, Chairman of SK Group, told the Asahi Shimbun in an interview published on 2 September 2026, describing joint production with Kioxia as “one option”. Ota’s answer a week later closed that door, which leaves the shareholding without a strategic rationale.
FAQ
What exactly did Kioxia’s CEO say about NAND prices?
Hiroo Ota, President and Chief Executive Officer at Kioxia Holdings, said on 9 September 2026 that memory prices have risen enough and that Kioxia aims to prevent further increases that would damage long-term AI demand. He noted that even hyperscalers have limited budgets, and has directed sales teams not to push aggressively for higher prices in data centre negotiations, according to Bloomberg’s report of the interview.
Why did Kioxia rule out a merger with SK Hynix?
Ota cited two obstacles. Antitrust review would be difficult given the combined NAND share of the two companies, which on Q2 2026 shipment data would exceed Samsung’s 25%. Second, Kioxia’s Yokkaichi and Kitakami fabs operate inside a joint venture framework with SanDisk extended to December 2034, so a third manufacturing partner cannot simply be added to the existing structure.
How large is SK Hynix’s stake in Kioxia?
SK Hynix holds bonds convertible into approximately 14.19% of Kioxia, held through a special-purpose vehicle that became Kioxia’s largest shareholder in August 2026. A standstill agreement prevents SK Hynix from holding 15% or more of Kioxia’s voting rights without consent until 2028, which keeps the position below any control threshold.
Did NAND prices really rise 70% in one quarter?
Yes, on both company and industry measures. Kioxia’s own NAND average selling prices rose roughly 70% quarter-on-quarter in the April-June 2026 quarter on low-single-digit bit growth. Industry-wide, TrendForce recorded NAND flash contract prices rising 70-75% sequentially in Q2 2026, following a 55-60% rise in Q1 2026.
What does a supplier-led price ceiling mean for buyers?
It helps the buyers with the least negotiating leverage most. Hyperscalers and Apple already secure supply through multi-year agreements at negotiated terms. Module houses, second-tier handset makers and channel assemblers buy closer to spot and carry memory at up to 43% of the bill of materials on entry-level devices, so any moderation in contract escalation lands directly on their gross margin.
Have Samsung, Micron or SK Hynix responded?
Not publicly, as of 9 September 2026. None has issued a statement matching or rejecting Kioxia’s pricing language. Their disclosed strategies point the other way: Samsung is expanding capacity, SK Hynix’s filings emphasise production growth, and Micron has signed 16 five-year strategic customer agreements. With a US price-fixing class action pending, visibly echoing a rival’s pricing statement carries legal risk.
