Kioxia Investor Day: 10th-Gen Flash Samples to Ship This Summer, Average Annual Capex of ¥470 Billion
Claire Weston
Kioxia announced its first-ever progressive dividend policy, disclosed that 332-layer Gen.10 flash samples ship this summer, and laid out roughly ¥470 billion in annual capex over the next three years — a decisive pivot from consumer-cycle dependence to AI data-center growth.
Why is Kioxia only now starting to pay dividends?
Kioxia framed its policy as a "progressive dividend" — payouts that only rise, never fall — but disclosed no specific amount.
This means → the company is sending a signal, not a check: telling institutional investors it has the will and capacity to return cash steadily.
The logic underneath is a structural demand shift. Tech Insights data projects data-center flash demand jumping from 295 EB in 2025 to 1,807 EB in 2028 — a 46% CAGR — with AI inference driving 86% of the incremental volume.
In plain terms = the flash market is moving from "sell more phones, earn more; sell fewer, lose money" to "AI servers keep buying." Stable revenue is what gives Kioxia the confidence to promise rising dividends.
332 layers instead of 400+ — why stack less?
Gen.10 BiCS FLASH uses a 332-layer, 1Tb TLC design. Samples ship this summer; volume ramp follows.
Versus the prior generation: bit density up 59%, read power efficiency improved over 40%, write throughput up over 30%.
Some rivals are pushing past 400 layers. Kioxia's own analysis shows 332 layers cost roughly 23% less per GB, use about 10% less power, and deliver roughly 35% higher cell reliability.
This reflects a strategy that is not about stacking the highest — it is about finding the best trade-off across cost, power, and reliability. Layer count is the means; total cost per GB is the goal.
CBA technology leads by four years — what does that mean?
CBA — CMOS Bonded Array, a process that fabricates logic and memory arrays separately, then bonds them face-to-face — is already in commercial use. Kioxia claims a roughly four-year lead over U.S. and Korean peers.
By 2029, Kioxia expects to reach 4.8 Gbps interface speed, while other major players remain at 3.6 Gbps.
This means → a 33% speed gap at the interface level translates directly into how fast data moves during AI inference. At scale, that gap compounds into system-level advantages in both performance and power.
Where does ¥470 billion a year go?
Planned capex averages roughly ¥470 billion per year from FY26 to FY28, a sharp step-up from FY25. Annual R&D budget rises to about ¥23 billion, up roughly 63% from FY25.
Capex priorities: Gen.10 volume ramp, next-generation process infrastructure, and back-end process upgrades. R&D focuses on Gen.10/Gen.11 development, ultra-high-IOPS SSDs, and OCTRAM — a new memory device class.
The company set an investment discipline: every capex decision must clear a hurdle rate above its weighted-average cost of capital (WACC). Trailing-twelve-month ROIC has already risen to 31%, up from 18% in FY25.
In plain terms = spending is large but gated — every major investment must prove it can earn more than the cost of the capital behind it. This is not "spend because we can."
What does Kioxia's AI product lineup look like?
For AI inference workloads, Kioxia has built three SSD product lines: CM series (high bandwidth, targeting KV-cache), GP series (ultra-high IOPS with XL-Flash, compatible with NVIDIA Storage-Next), and LC series (high capacity, up to 245 TB, already in mass production).
The CEO said Kioxia is shifting from consumer-cycle dependence to stable growth led by AI infrastructure. Multi-year long-term agreements (LTAs) will improve revenue visibility.
This reflects a clear transformation target: push data-center revenue above 60% of total by FY28, lock in income with long-term contracts, and dampen the boom-bust earnings swings that have defined the flash industry.
Can the balance sheet support all of this?
The CFO disclosed a target to raise the equity ratio from the current ~38% to above 50%, while maintaining ample liquidity.
Kioxia reached a net cash position in FY26 Q1 — meaning cash exceeds debt — and guided Q1 operating margin at 74%.
Per Bloomberg, Kioxia's stock price has risen more than 30× over the past year.
This means → cash on hand, an extremely high margin, and a stock that has already re-rated sharply — that is the financial foundation allowing Kioxia to launch dividends, expand capacity, and boost R&D all at once.
Content is for reference only, not financial advice.