
This photo taken on July 5, 2025 shows signage for Japanese company Kioxia, the world's third largest manufacturer of NAND flash memory chips, at their Yokohama Technology Campus in the Ofuna area of Yokohama, Kanagawa prefecture, south of Tokyo. AFP via Getty Images/RICHARD A. BROOKS
Japan's Kioxia has formally designated mass production of its 10th-generation BiCS10 NAND flash as a top strategic priority for fiscal year 2026, staking a claim on the most contested segment of the global memory market at the precise moment rivals Samsung Electronics and SK hynix are still weighing when to commit their own capital.
Industry sources confirmed on May 23 that Kioxia has listed the BiCS10 launch — built on a 332-layer stacking architecture — as one of its core strategies for the fiscal year running from April 2026 through March 2027. Capital spending in Kioxia's most recent quarter was concentrated on both its current 8th-generation and the forthcoming 10th-generation NAND lines, signaling that investment is already moving.
The announcement matters because every major NAND buyer — from hyperscale cloud operators to SSD makers supplying AI servers — is competing for supply in a market that TrendForce projects will see overall NAND flash prices surge 85 to 90 percent quarter-over-quarter in Q1 2026. Whoever delivers the highest-density, highest-bandwidth NAND first shapes not just their own market position but the cost and timeline of AI infrastructure builds worldwide.
BiCS10 is the 10th iteration of Kioxia's proprietary Bit Cost Scalable stacking technology. Where the current 8th-generation BiCS chips stack 218 memory cell layers vertically, BiCS10 reaches 332 layers — a 52 percent increase in layer count. The higher stack translates directly to more storage per unit of chip area: TrendForce and Tom's Hardware both confirm a 59 percent improvement in bit density compared with the previous generation.
The speed increase is equally significant. BiCS10 uses the Toggle DDR 6.0 interface standard, pushing data transfer rates to 4.8 gigabits per second — 33 percent faster than the 8th-generation product. That bandwidth matters for enterprise SSDs in AI inference workloads, where NAND must feed compute clusters with sustained, high-throughput reads.
The chip also incorporates Kioxia's CMOS Directly Bonded to Array architecture, which manufactures the logic circuitry and the memory cell array on separate wafers before bonding them. This design improves signal integrity across the tall stack and enables the new PI-LTT (Power Isolated Low-Tapped Termination) power technology, which cuts input power consumption by 10 percent and output consumption by 34 percent compared with BiCS8. Kioxia CTO Hideshi Miyajima stated at the chip's unveiling that the power-efficiency focus responds directly to the demands AI technology places on modern data centers.
Kioxia has confirmed that its BiCS10 product will deliver 2Tb of capacity per die using standard TLC (three bits per cell) technology, without needing the more error-prone PLC (five bits per cell) architecture — a meaningful distinction for enterprise workloads where endurance and reliability matter.
Read more: Samsung Revives Shelved Chip Programs as Worker Walkout Looms
Kioxia will split manufacturing responsibilities across its two main Japanese facilities. The established Yokkaichi complex in Mie Prefecture, which houses the company's research and development center, will continue producing BiCS8 (218-layer) chips for mobile and consumer applications. The newer Kitakami K2 fabrication facility in Iwate Prefecture — which opened in September 2025 and is equipped with tools designed specifically for high-layer NAND — is designated as the production site for BiCS10.
The division makes manufacturing sense: K2's equipment is newer and calibrated for the more demanding deposition control and deep-trench etching that 332-layer stacks require. Kioxia is not building a new facility for BiCS10 — it is repurposing K2, which reduces capital expenditure significantly compared with a greenfield build.
TrendForce reports that the Kioxia and SanDisk consortium — which jointly operates the Yokkaichi and Kitakami plants under a long-standing joint venture — increased capital investment plans by 41 percent year-over-year to $4.5 billion in 2026, focused on expanding 8th-generation production and supporting 10th-generation development. SanDisk, which spun off from Western Digital in 2025 and uses BiCS technology under the joint venture, has confirmed persistent NAND demand outpacing supply through the end of calendar year 2026.
Kioxia's accelerated timeline lands at a strategically vulnerable moment for its South Korean rivals.
Samsung Electronics has designed its own 10th-generation NAND, internally called V10, with a layer count in the 430-layer range — a significantly taller stack that would surpass Kioxia on density. Samsung presented V10 technical details at the International Solid-State Circuits Conference in February 2025, with the chip using a 5.6 GT/s interface and a cell-on-peripheral architecture that applies hybrid bonding for the first time. However, Samsung has delayed full-scale mass production investment. Reports from multiple industry sources indicate that V10 production was originally targeted for the second half of 2025 but has been pushed back, with large-scale capital commitment now unlikely before at least the first half of 2026, citing steep etching challenges for ultra-high-layer stacks and uncertain demand patterns.
Samsung's hesitation reflects a larger organizational reality: the company has spent the past year stabilizing its DRAM and high-bandwidth memory programs after a difficult period, and its capital allocation has been concentrated there. Samsung recently confirmed it has begun mass-producing its sixth-generation HBM4 high-bandwidth memory — a milestone that absorbed extensive engineering and equipment resources.
SK hynix is pursuing its own 10th-generation NAND at the 300-layer range, with hybrid bonding targeted for its V10 design, and aims to complete development through a pilot line in 2026 before moving to full production in early 2027 — a timeline that puts it behind Kioxia if Kioxia's fiscal 2026 target holds.
According to TrendForce's Q4 2025 market data, Samsung led the global NAND flash market with $6.6 billion in revenue and a 28.0 percent share. SK hynix (including its Solidigm subsidiary) followed with approximately $5.21 billion and 22.1 percent share, after revenue jumped 47.8 percent quarter-over-quarter on explosive enterprise SSD demand. Kioxia ranked third with $3.31 billion in revenue — record quarterly highs for both revenue and bit shipments — at approximately 15.6 percent market share, up two percentage points from the prior quarter.
Read more: Samsung Begins HBM4 Shipments as SK Hynix Lags and a 50,000-Worker Strike in Five Days Threatens the AI Chip Supply
Enterprise data centers building out AI server infrastructure have shifted from hard disk drives to NAND-based solid-state drives at scale. The reason is bandwidth: AI inference workloads require sustained, low-latency access to model weights and activation data, and even the fastest mechanical drives cannot deliver the throughput that GPU and AI accelerator clusters demand.
Higher-layer NAND directly enables larger-capacity SSDs at lower cost per gigabyte. A single 2Tb BiCS10 die in a multi-die package can support drives exceeding 100TB without exotic manufacturing workarounds. SanDisk launched a 256TB enterprise SSD in early 2026. Kioxia is separately developing, in collaboration with Nvidia, an SSD targeting 100 million IOPS for AI servers — roughly 40 to 50 times the performance of current enterprise drives — with samples expected by 2027.
Samsung and SK hynix have simultaneously been diverting wafer capacity from NAND production toward High Bandwidth Memory, which commands higher margins for AI training and inference accelerators. EE Times reporting confirmed that Samsung cut NAND wafer output from approximately 4.9 million wafers in 2024 to 4.68 million in 2025; SK hynix cut from 1.9 million to 1.7 million. This capacity reallocation has directly contributed to the NAND supply tightening that has pushed enterprise SSD prices sharply higher throughout 2025 and into 2026.
Production designation is not the same as production. Industry sources confirmed on May 23 that Kioxia had originally planned to begin 10th-generation NAND investment as early as the second half of calendar year 2025, but confirmed orders have not yet materialized. A clearer picture of committed customer demand is expected to emerge in the second half of 2026.
That gap between strategic intent and confirmed demand is the central uncertainty in Kioxia's plan. Building high-layer NAND at volume without pre-committed customers risks inventory accumulation in a market that already proved brutally cyclical in the 2022–2023 downturn. Kioxia's management appears to be advancing equipment orders and fab preparation while monitoring demand signals in parallel — a calculated risk at a moment when AI infrastructure investment is creating the strongest NAND demand environment in the industry's history.
If Samsung and SK hynix maintain their delayed posture through the remainder of 2026, Kioxia's window to establish BiCS10 as the reference chip for next-generation enterprise SSDs could be significant. Both South Korean companies are expected to move eventually — Samsung's V10 roadmap calls for a layer count that would surpass BiCS10 on density — but the first mover in volume production typically captures the initial design wins with major SSD customers.
What is Kioxia BiCS10 NAND flash?
BiCS10 is Kioxia's 10th-generation 3D NAND flash memory, using a 332-layer cell-stacking architecture that delivers 59 percent higher storage density and 33 percent faster data transfer speeds compared with the 8th-generation product. It uses the Toggle DDR 6.0 interface at 4.8 gigabits per second and is targeted at enterprise solid-state drives for AI data centers.
When will next-generation NAND flash be available in enterprise SSDs?
Kioxia has designated BiCS10 as a fiscal year 2026 production priority, meaning volume output is planned for the April 2026 to March 2027 window. Samsung's competing 430-layer V10 NAND has been delayed from its original 2025 target, with large-scale production investment now expected no earlier than the first half of 2026. SK hynix is targeting full production of its 10th-generation NAND in early 2027.
Why are NAND flash prices rising in 2026?
NAND flash prices are rising because AI infrastructure buildouts are driving explosive demand for enterprise SSDs, while Samsung and SK hynix have simultaneously shifted wafer capacity toward High Bandwidth Memory for AI accelerators — reducing overall NAND supply. TrendForce projected NAND flash prices to surge 85 to 90 percent quarter-over-quarter in Q1 2026. SanDisk has confirmed demand is outpacing supply and expects this imbalance to persist through the end of calendar year 2026.
How many layers does BiCS10 have compared with Samsung's next-generation NAND?
Kioxia's BiCS10 stacks 332 layers. Samsung's 10th-generation V-NAND design, presented at ISSCC 2025, targets a layer count in excess of 400 — making it denser than BiCS10 on paper. However, Samsung has not committed to mass production of V10 at scale, while Kioxia has made BiCS10 a stated fiscal year 2026 production priority.
