How to read this page. Every maker in our Memory section uses the same 16 sections in the same order, the same shape as our AI Chips brand guides. Each section starts Simple, then goes Deeper, then Expert: stop wherever you have what you need. The small numbers are sources: click one to open the original document. Where a company has not published something, we say so rather than estimate.
To see this company beside the other fifteen — what each makes, where the fabs are, and what each publishes about price — use the comparison table. For how much this company publishes about identifying its own chips once they are on somebody else’s module, see whose DRAM chips are actually inside your memory stick.
1.At a glance
SimpleStart here
Kioxia makes NAND flash and nothing else — no DRAM, no HBM 1. It is the direct descendant of Toshiba's memory business, and NAND flash itself was invented at Toshiba 2.
What it makes"Development, manufacturing and sales of memory and related products" — NAND flash and SSDs
1
How big¥2,337,628 million revenue in FY2025, the year to March 2026
3
Who makes itYokkaichi and Kitakami in Japan, through a joint venture with Sandisk
4
Any HBM?No. Kioxia's AI bet is fast flash, not stacked DRAM
5
DeeperThe detail
The company is two entities. Kioxia Holdings Corporation is the listed parent, established 1 March 2019, whose business is "Group strategy formulation and management oversight" 6. Kioxia Corporation is the operating company, established 1 April 2017, which actually develops and makes the memory 1. Both sit at 3-1-21 Shibaura, Minato-ku, Tokyo, and together employ 15,218 people on a consolidated basis as of 31 March 2026 6 1.
ExpertFor specialists
FY2025, the year ended 31 March 2026, was transformative: revenue of ¥2,337,628 million against ¥1,706,460 million the year before, operating profit of ¥870,369 million against ¥451,748 million, and total profit for the year of ¥554,496 million against ¥272,321 million 3. That last figure is profit including non-controlling interests; the amount attributable to owners of the parent is marginally lower 3. The market re-rated it accordingly — Kioxia joined the Nikkei 225 on 1 April 2026 7 and was upgraded to investment grade, "BBB-", by both S&P and Fitch in May 2026 8.
This page uses the audited FY2025 full-year figures as its financial anchor. Kioxia's quarterly filings are published through a document host this check could not reach, so no quarterly figure is stated here rather than repeating one from secondary coverage.
2.Company card
SimpleStart here
| Listed parent | Kioxia Holdings Corporation, established 1 March 2019 6 |
|---|
| Operating company | Kioxia Corporation, established 1 April 2017 1 |
|---|
| HQ | 3-1-21, Shibaura, Minato-ku, Tokyo 108-0023, Japan 6 1 |
|---|
| CEO | Hiroo Ota, President and Chief Executive Officer, from 1 April 2026 6 9 |
|---|
| Listed | Tokyo Stock Exchange Prime Market, code 285A, from 18 December 2024 10 11 |
|---|
| Fiscal year | 1 April to 31 March 3 |
|---|
| Employees | 15,218 consolidated, as of 31 March 2026 6 |
|---|
DeeperThe detail
Leadership changed in 2026. Kioxia announced in January that "Nobuo Hayasaka will step down as President and Chief Executive Officer on March 31, and from April 1, our current Executive Vice President, Hiroo Ota will assume the role" 9. The listing itself was recent: the Japan Exchange Group's approval notice gives the listing date as 18 December 2024, ticker 285A, at an offering price of "JPY 1,455" 10.
ExpertFor specialists
Ownership is where this page has to be careful about dates. As of 31 March 2026, Kioxia's own stock page lists Toshiba Corporation at 17.59%, BCPE Pangea Cayman2, Ltd. at 14.17%, and several further Bain-consortium vehicles alongside institutional holders 11. That snapshot has since changed: Kioxia's own investor-relations index lists a "Notice Regarding Change in Major Shareholders and Largest Shareholder" dated 10 August 2026, filed through a document host this check could not open. So this page states the ownership Kioxia publishes with its date attached, and records that a change has been formally notified since — rather than describing the new structure from press reports it cannot verify.
A second live story: on 15 September 2026 Kioxia confirmed it "is preparing to list American Depositary Shares representing its common shares on a U.S. stock exchange", while stating that "details of the listing such as the schedule, listing market, and listing method have not yet been decided, and the Company may decide not to continue pursuing the listing" 12.
3.The story
SimpleStart here
- 1987 NAND flash is invented at Toshiba by Dr. Fujio Masuoka 2.
- 2017 Toshiba's memory business becomes a separate company, 1 April 1.
- 2018 Sale to a Bain Capital-led consortium completes, 1 June 13.
- 2019 Renamed Kioxia, effective 1 October 14.
- 2024 Lists on the Tokyo Stock Exchange, 18 December 10.
- 2026 Joins the Nikkei 225; upgraded to investment grade 7 8.
DeeperThe detail
Kioxia's claim to the technology is direct and documented: "Toshiba created the world's first NOR flash memory in 1984, and in 1987, it invented NAND flash memory, which is widely used still today", crediting "Dr. Fujio Masuoka (formerly with Toshiba, now an honorary professor at Tohoku University)" 2. Kioxia labels its own lineage "Toshiba Memory (Current KIOXIA)" 2.
The name was chosen to carry that meaning: Kioxia combines "kioku", Japanese for memory, with "axia", Greek for value — "Merging 'memory' with 'value,' Kioxia represents the company's mission to uplift the world with 'memory'" 14.
ExpertFor specialists
The 2018 sale is the pivot, and its structure explains a lot about Kioxia today. Toshiba's own filing records that "the closing of the sale has been completed today as scheduled" on 1 June 2018, to K.K. Pangea, "a special purpose company controlled by a consortium led by Bain Capital Private Equity, LP" 13. Toshiba did not simply exit: it reinvested and retained roughly 40.2% of Pangea's voting rights, making the memory business an equity-method affiliate rather than a consolidated subsidiary 13. Hoya took 9.9%, with INCJ and the Development Bank of Japan holding instruction rights over a further 16.7% 13.
Kioxia's sale price is widely quoted in dollars. The figure in Toshiba's own filing did not extract cleanly in this check, so no price is stated here rather than repeating a secondary number 13.
4.The lineup
SimpleStart here
| Product | What it is | Status |
|---|
| BiCS FLASH generation 10 | 332-layer 1Tb TLC NAND | Sample shipments from July 2026 15 |
| BiCS FLASH generation 9 | 230-layer 1Tb TLC, cost-optimised | Sample shipments from July 2026 16 |
| BiCS FLASH generation 8 | 218-layer, the volume generation | Targeted at "Over 80%" of output by end FY26 17 |
| CM10 Series | Kioxia's first PCIe 6.0 enterprise SSD | Sampling; 1.60TB to 61.44TB 18 |
| GP1 Series | Super-high-IOPS SSD for AI | Samples to select customers by end 2026 5 |
| XL1 Series | CXL memory expansion module | Evaluation samples from August 2026 19 |
| EXCERIA | The consumer SSD brand | On sale; Kioxia publishes no price 20 |
DeeperThe detail
The generation numbering is genuinely confusing, and worth getting right. Kioxia shipped generation 10 samples first, on 3 July 2026, at 332 layers 15 — then generation 9 samples on 30 July 2026, at 230 layers 16. Generation 9 is not a predecessor that arrived late; it is a lower-layer, cost-optimised part aimed at "AI-enabled PCs and smartphones with low- to mid-level storage capacities" 16. Both run a 4.8 Gb/s interface, "a 33% improvement over the 8th generation" 15 16.
ExpertFor specialists
At the top of the range Kioxia is pushing density hard, jointly with Sandisk. Its 332-layer QLC part is claimed to exceed "37 Gb/mm²" of bit density, an "up to 60% increase in bit density compared to their 8th-generation" 21, and a 2Tb QLC device with a "6-plane design" followed in August 2026 22. Hideshi Miyajima, Chief Technology Officer, frames the purpose in AI terms: "QLC technology enables the efficient storage of rapidly expanding data volumes, helping deliver greater performance and scalability for AI systems" 21.
5.How it’s made
SimpleStart here
Kioxia's distinguishing manufacturing technique is CBA, short for "CMOS directly Bonded to Array".
- The control circuitry (CMOS) and the memory cell array are built on two separate wafers 23.
- Each is made under conditions optimised for that job alone 23.
- The two wafers are then bonded together, copper pad to copper pad 23.
DeeperThe detail
Kioxia's own definition: CBA is "Technology wherein each CMOS wafer and cell array wafer is manufactured separately in their optimized conditions and then bonded together", with "copper (Cu) bonding pads formed on their surfaces" 23. It contrasts this with the two older approaches by name: CNA, where "CMOS is formed beside the cell array, so the chip size is approximately the sum of the CMOS and cell array sizes", and CUA, where "CMOS is formed beneath the cell array", which creates "a trade-off between improving cell and CMOS performance" 23.
The payoff Kioxia claims for generation 8: "cell current improved by 35% compared to the previous generation, and interface speed reached 3.6 Gbps" 23.
ExpertFor specialists
Generation 9 adds a second technique on top of CBA, which Kioxia names OPS (On-Pitch Select Gate Drain) 16 — the mechanism by which a 230-layer part achieves the same 4.8 Gb/s interface speed as the 332-layer generation 10 16 15. That pairing is the clearest signal of Kioxia's strategy: layer count is not the only lever, and the company is deliberately building two products at different layer counts for different markets rather than pushing every customer onto the tallest stack.
6.Where it’s made: fabs and addresses
SimpleStart here
Everything Kioxia makes, it makes in Japan, at two plants.
| Plant | What Kioxia publishes |
|---|
| Yokkaichi, Mie Prefecture | 800 Yamanoisshiki-cho, Yokkaichi-shi, Mie 512-8550; founded January 1992; 694,000 m²; 6,788 employees at 31 March 2026 24 |
| Kitakami, Iwate Prefecture | 5-29, Kitakogyodanchi, Kitakami City; Fab1 from 2020, Fab2 from September 2025 25 26 |
| Kitakami Fab3 | Announced August 2026, "targeting operations to commence in fiscal 2029" 27 |
DeeperThe detail
The fabs are not wholly Kioxia's, and this is the single most misunderstood thing about the company. Kioxia and Sandisk jointly operate Flash Ventures, described in Sandisk's SEC filing as "three business ventures, Flash Partners Ltd., Flash Alliance Ltd. and Flash Forward Ltd." in which Sandisk holds "a 49.9% ownership position" — leaving Kioxia with 50.1% 4. Output is shared: "We and Kioxia are entitled to purchase a share of Flash Ventures' output, which generally equals 50% each", at "cost plus a small markup" 4.
The division of property is unusual and worth stating precisely: "Kioxia owns the facilities and provides wafer manufacturing services to Flash Ventures at cost using manufacturing equipment owned or leased by Flash Ventures", and "Flash Ventures accounts for approximately 80% of the total manufacturing capacity in the facilities owned by Kioxia" 4. Kioxia owns the buildings; the joint venture owns the tools.
ExpertFor specialists
Individual fabs are documented in their own announcements rather than on the plant pages. Yokkaichi's Fab7 opened in October 2022 with "Total investment in phase one of Fab7…expected to be approximately one trillion yen", partly government-subsidised, built "capable of absorbing earthquake shocks" and able to produce the then-current 162-layer flash 28.
The wider investment numbers are large and recent. In January 2026 the two companies extended the joint venture: "Originally set to expire on December 31, 2029, the agreement will now run through December 31, 2034", with Sandisk paying "USD 1.165 billion for manufacturing services and continued availability of supply" in instalments from 2026 to 2029 29. In August 2026 they committed "over $31 billion (approximately 5 trillion yen)" through 2032 across both plants, "contingent upon government support", noting that the partnership has "invested over $50 billion (approximately 9 trillion yen) in Japan over the past 25 years" 30. No per-company split of the $31 billion is disclosed 30.
Japanese state support is officially confirmed but modestly described: in February 2024 the joint venture facilities at both plants "have been approved to receive an up to 150 billion yen subsidy" under "a designated government program aimed at facilitating corporate investment in cutting-edge semiconductor production facilities" 31. Fab3 at Kitakami is explicitly conditional: "Investment plans for Fab3 are contingent upon government support" 27.
7.How the memory works
SimpleStart here
NAND flash traps electric charge inside a cell, where it stays without power — which is why your phone remembers its photos after the battery dies. Kioxia's BiCS FLASH builds those cells in vertical stacks rather than spreading them sideways, which is how capacity keeps rising 15.
DeeperThe detail
Cells can hold more than one bit each, by storing several distinguishable charge levels. Kioxia ships TLC, three bits per cell, as its mainstream density 15, and QLC, four bits per cell, where capacity matters more than endurance 21. More bits per cell means more capacity from the same silicon and less tolerance for error — which is why the QLC parts are pitched at storing "rapidly expanding data volumes" rather than at write-heavy work 21.
ExpertFor specialists
Kioxia's most technically distinctive product is XL-FLASH, which it describes as "an extremely low-latency, high-performance flash memory" built to "address the performance gap that currently exists between volatile memories such as DRAM and current flash memory" 32. Read latency is under 5 microseconds, using a "16-plane architecture for improved latency" and MLC cells operable in SLC mode 32.
That gap-filling idea is the whole of Kioxia's AI strategy. It has no DRAM to sell, so instead of stacking DRAM higher it is making flash fast enough to take work DRAM would otherwise have to hold — see The HBM position below.
8.Specs explained
SimpleStart here
| Product | Headline numbers Kioxia publishes |
|---|
| BiCS FLASH gen 10 | 332 layers; 1Tb TLC; 4.8 Gb/s interface; bit density +59% vs gen 8 15 |
| BiCS FLASH gen 9 | 230 layers; 1Tb TLC; 4.8 Gb/s interface 16 |
| 332-layer QLC | Bit density "exceeds 37 Gb/mm²", up to +60% vs gen 8; Toggle DDR6.0 21 |
| CM10 Series SSD | PCIe 6.0, NVMe 2.1; 1.60TB–61.44TB; ~92% higher sequential read than CM9 18 |
| GP1 Series SSD | "Up to 10 million random read IOPS at a 512-byte block size"; up to 50 DWPD 5 |
| XL-FLASH | Read latency under 5 microseconds; 16-plane architecture 32 |
| EXCERIA PRO | Up to 7,300 MB/s read, 6,400 MB/s write; PCIe 4.0 20 |
DeeperThe detail
The generation-10 gains are stated against generation 8, not generation 9, which is the correct comparison given how the two were released: bit density up 59%, write power efficiency up 18%, read power efficiency up 30% 15. The CM10 is "Kioxia's first PCIe® 6.0 enterprise SSD" — its own first, not an industry one, though the headline invites the wider reading — offering "up to approx. 92 % higher sequential read performance and up to approx. 85 % higher random read performance" over the CM9, with liquid cooling available on the E3.S and E1.S 9.5mm form factors 18.
ExpertFor specialists
GP1 is the most aggressive number Kioxia publishes and the clearest statement of where it thinks flash is going: "Up to 10 million random read IOPS at a 512-byte block size", PCIe 6.0, up to 50 drive writes per day, with an architecture Kioxia says is designed to scale toward "future generations targeting up to 100 million IOPS" 5. Availability is honest about its stage: "Evaluation samples will be available to select customers by the end of 2026" 5.
9.Official pricing
SimpleStart here
Kioxia publishes no prices at all — not for NAND chips, and not even for its own consumer SSDs. The EXCERIA PRO product page carries full specifications and a "Where to Buy" link, and no figure 20.
DeeperThe detail
That makes Kioxia the most reticent of the makers covered so far: Samsung and SK hynix at least publish consumer drive prices, and Micron did until it left the consumer market. Kioxia's consumer line — EXCERIA, EXCERIA PRO, EXCERIA SATA, EXCERIA BASIC — is specified in public and priced only by retailers 20. Any price for a Kioxia product must therefore be sourced to a shop and dated, not to Kioxia.
ExpertFor specialists
One official price mechanism does exist, and it is internal. Sandisk's SEC filing describes how wafers move inside Flash Ventures: "The price for which we and Kioxia pay Flash Ventures for flash memory wafers is cost plus a small markup" 4. That is the only published statement of what Kioxia's own output costs to acquire — a transfer-pricing basis between joint-venture partners, not a market price, but a genuinely official one.
10.The HBM position
SimpleStart here
Kioxia has no HBM and makes no DRAM. That is not an oversight on this page — it is the defining fact about the company's position in the AI build-out. Its answer to the same problem is to make flash fast enough to do some of what DRAM does 5.
DeeperThe detail
Kioxia's own framing, from its June 2026 Investor Day: "Flash memory and SSDs are poised to become core structural components of next-generation AI architectures" 33. The concrete products behind that are the GP1 super-high-IOPS drive, the XL1 CXL memory expansion module, and XL-FLASH underneath both 5 19 32. The GP1 release describes the target explicitly as "emerging AI storage architectures, which extend High Bandwidth Memory (HBM) as a fast flash memory media-based tier", letting systems reach larger datasets at lower cost than expanding HBM 5.
The financial commitment is stated too: capital expenditure of roughly ¥470 billion a year and R&D of roughly ¥230 billion a year over three years, aiming to lift the "sales ratio in the datacenter and enterprise markets to over 60%" 33 17.
ExpertFor specialists
One widely-repeated claim needs correcting, because it appears even in careful coverage. High Bandwidth Flash (HBF) is not a Kioxia project. The standardisation effort is Sandisk and SK hynix, who announced it jointly in February 2026 and are running it "under the Open Compute Project (OCP)" 34. Kioxia appears in that release only inside Sandisk's forward-looking-statements boilerplate, as a party Sandisk depends on commercially — not as a participant 34. This check found no Kioxia release mentioning HBF at all. Kioxia's flash-for-AI bet is GP1, XL1 and XL-FLASH, which is a different architecture aimed at a similar problem 5 19.
Kioxia does have DRAM in its laboratories, which is worth stating carefully so it is not mistaken for a product. In December 2025 it presented "highly stackable oxide-semiconductor channel transistors" for 3D DRAM at IEDM, having fabricated and verified an eight-layer stack 35. Kioxia describes this as research: "we will continue our research and development of this technology in order to realize the deployment of 3D DRAM in real-world applications" 35. It is not HBM, not a product, and not evidence that Kioxia is entering the DRAM market.
11.Who buys it and why
SimpleStart here
Kioxia sells NAND and SSDs to data-centre operators, PC and phone makers, and carmakers. Its stated priority is the data centre: it wants that market above 60% of sales 33.
DeeperThe detail
Kioxia organises its AI-era enterprise range around three jobs, named at its Investor Day: the CM Series for KV cache storage, the GP Series for GPU memory expansion, and the LC Series for high-capacity data storage, including a 245TB model — with the CM Series stated to be compatible with "NVIDIA's CMX Platform" 33. That is a portfolio designed around the shape of an AI inference system rather than around drive form factors.
ExpertFor specialists
Kioxia's stated commercial strategy is to trade some price upside for predictability: it aims to secure multi-year long-term agreements "to improve revenue visibility and profit quality" 33 17. Its market framing is aggressive — Investor Day slides put data-centre flash inference demand growing at "86% CAGR" and the flash market rising from 295 exabytes in calendar 2025 to 1,807 exabytes by 2028 17. Those are Kioxia's own projections, presented to investors, and this page reports them as such rather than as market fact.
12.Hidden in plain sight
SimpleStart here
The company that invented NAND flash does not own its own factories outright. The tools inside Kioxia's fabs belong to a joint venture with Sandisk; Kioxia owns the buildings 4.
DeeperThe detail
Sandisk's filing states it plainly: "Kioxia owns the facilities and provides wafer manufacturing services to Flash Ventures at cost using manufacturing equipment owned or leased by Flash Ventures", and "Flash Ventures accounts for approximately 80% of the total manufacturing capacity in the facilities owned by Kioxia" 4. Half the output of Japan's flagship memory plants is contractually Sandisk's 4. A reader looking at Kioxia's capacity figures and assuming Kioxia can sell all of it would be wrong by roughly half.
ExpertFor specialists
A second: the most detailed public description of Kioxia's own manufacturing arrangements is not published by Kioxia. The ownership percentages, the output split, the cost-plus transfer price and the 80% capacity figure all come from Sandisk's SEC filing 4, because Sandisk is US-listed and must describe its supply arrangements in detail. Kioxia's own site describes the Yokkaichi plant without enumerating its fabs or naming the joint venture at all 24. Where a company is listed shapes what the public can learn about it — the same reason this site can say more about Micron than about its Asian peers.
13.Weak spots
SimpleStart here
- One product, one country. NAND only, made only in Japan 1 4.
- No DRAM and no HBM in the generation where those are where the money is 5.
- Half the fab output is contractually Sandisk's 4.
DeeperThe detail
Concentration is the structural weakness. Every wafer Kioxia makes comes from two sites in Japan 24 25, through a joint venture it controls by a 0.2-point margin 4, and its expansion is explicitly conditional on state support: "Investment plans for Fab3 are contingent upon government support" 27, and the $31 billion programme likewise "contingent upon government support" 30. A memory maker with no DRAM also has no hedge when NAND prices fall and DRAM prices do not.
ExpertFor specialists
Disclosure is the other weakness, and it is partly structural. Kioxia publishes its quarterly results through a document host this check could not reach, so quarterly figures could not be verified here at all 3. Its ownership page is dated 31 March 2026 while a change of largest shareholder was formally notified in August 2026 11. Its plant pages do not enumerate fabs or describe the joint venture 24. None of this is concealment — the filings exist — but a reader relying on Kioxia's English-language public pages alone will have an out-of-date picture of who owns the company and a materially incomplete one of who owns its output.
14.India angle
SimpleStart here
Kioxia has no presence in India at all — no plant, no research centre, and not even a sales office 36 37.
DeeperThe detail
Kioxia's global network page lists every group company: six in Japan, one in the United States, two in Europe, four across Asia-Pacific, one in China, one in Israel, plus Solid State Storage Technology Corporation. India does not appear 36. Its sales-office list covers the US, Canada, Germany, France, Ireland, Spain, Sweden, the UK, China, Singapore, South Korea and Taiwan; India is handled through a web enquiry form covering "ASEAN countries, India, Australia and New Zealand" 37.
ExpertFor specialists
This is a clean confirmed negative, and it is a useful contrast for the section as a whole. Of the four memory makers documented so far, only Micron has manufacturing in India, and even that is assembly and test rather than wafer fabrication. Samsung has research in Bengaluru; SK hynix has a single Delhi sales office; Kioxia has neither. India's semiconductor ambitions are real and funded, but as of this check the memory industry's physical presence there consists of one packaging plant in Gujarat.
15.What’s next
SimpleStart here
Only what Kioxia has announced:
- Kitakami Fab3: "targeting operations to commence in fiscal 2029", contingent on government support 27.
- $31 billion with Sandisk across both plants through 2032 30.
- GP1 samples to select customers by the end of 2026 5.
- US listing: in preparation, with no schedule decided and no commitment to proceed 12.
DeeperThe detail
The joint venture now runs to the end of 2034 29, and the capital plan runs to 2032 30, so Kioxia's manufacturing arrangements are settled further out than most of this industry's. Its own three-year financial targets are equally explicit: capex around ¥470 billion a year, R&D around ¥230 billion a year, a data-centre and enterprise sales ratio above 60%, an equity ratio rising from 38% to the "High 50s", and a progressive dividend 33 17.
ExpertFor specialists
Two open questions sit outside those plans. The first is ownership: a change of largest shareholder was notified in August 2026 and this page could not read the notice 11, so who controls Kioxia as of this check is not something this page can state. The second is the US listing, which Kioxia describes in conditional terms it chose carefully — "the Company may decide not to continue pursuing the listing depending on the circumstances over the course of the preparation" 12. Both are the kind of thing that will be settled by a filing rather than a product launch, and this page will follow the filing.