Kioxia's Conservative Capex Strategy: A Restrained Bet Amid the AI Storage Boom
Claire Weston
Kioxia is capping annual capex at roughly ¥470 billion over three years — about 10% below its historical peak — even as AI-driven NAND demand grows at 22% annually. The company is betting that cycle discipline is worth more than capacity leadership.
How hot is the NAND market right now?
TechInsights forecasts the NAND market will grow at 22% annually from 2025 to 2028. Data-center NAND leads at 46%, dwarfing the 8% growth for PC and smartphone NAND.
TrendForce data shows NAND prices rising over 70% quarter-on-quarter in the April–June quarter — outpacing DRAM's roughly 60% gain.
This means → AI's pull on storage is no longer a forecast. It is arriving now, and data centers are the overwhelming driver.
Why is Kioxia hitting the brakes during an industry upcycle?
Kioxia plans average annual capex of roughly ¥470 billion through March 2029 — about 10% below the ¥510.4 billion record set in the fiscal year ending March 2023.
That translates to a capex-to-revenue ratio of just 3%–5%, well under the company's own 20% ceiling and below last fiscal year's 12% actual level.
Chief Strategy Officer Junichiro Yaguchi stated plainly: "Even if we expect strong cash flow, that does not mean we step on the gas."
In plain terms = Kioxia has the cash but is choosing not to spend it — prioritizing shareholder returns over new capacity buildouts.
Where is near-term investment going?
Spending is focused on unused space inside existing cleanrooms at the Kitakami fab in northern Japan — utilizing what is already built, not breaking ground on new facilities.
This means → Kioxia is squeezing more output without raising the capacity ceiling. The risk profile is far lower than a greenfield expansion.
Are long-term agreements a bullish sign — or a cycle-top warning?
President Hiroo Ota disclosed that long-term purchase agreements (LTAs) — contracts where customers lock in volumes years ahead — will account for roughly 50% of shipments by 2028. Several hyperscale cloud customers want contracts extending into 2029 and beyond.
However, a veteran semiconductor market observer warned: "Historically, a surge in LTA negotiations often signals the market is near a peak."
This reflects a tension: customers scrambling to lock in supply confirms strong demand, yet every past wave of aggressive long-term contracting has preceded a downturn.
How painful was the last expansion cycle?
In October 2022, Kioxia committed roughly ¥1 trillion to expand its Yokkaichi fab. The market then turned down, and Kioxia posted net losses for five consecutive quarters — only returning to profit in December 2023.
Samsung Electronics and Micron Technology also fell into losses during the same period — this was an industry-wide oversupply cycle, not a single-company misjudgment.
In plain terms = the last expansion hit the cycle peak dead-on. A trillion yen bought more than a year of losses — and that memory is the direct reason Kioxia is choosing restraint this time.
Is the market pricing Kioxia fairly?
QUICK consensus forecasts show Kioxia's operating profit could exceed ¥10 trillion (roughly $62.3 billion) by the fiscal year ending March 2029 — approximately 12 times the level just reported.
Yet Kioxia currently trades at roughly 10x earnings, far below the ~40x multiples of chip-equipment peers like Tokyo Electron and Advantest.
This means → the market has not fully priced in Kioxia's profit trajectory. The discount reflects either distrust of NAND cycle durability or doubt that Kioxia can hold margins.
Capex restraint — smart discipline or a missed window?
BULL
Cycle scars run deep
The last trillion-yen expansion bought five quarters of losses. Discipline is itself a competitive edge.
Valuation gap persists
10x P/E against a 12x profit growth forecast — the market hasn't caught up.
Cash returned to shareholders
Dividends over new fabs means higher capital efficiency.
BEAR
Rivals may not hold back
If Samsung or Micron expand first, Kioxia risks losing share.
LTA buildup cuts both ways
Locked-in volumes become a burden if prices collapse in a downturn.
AI demand pace is uncertain
A 46% data-center growth rate sustained for three years — no one can guarantee that.
In plain terms = Kioxia is betting that the safety of not expanding outweighs the share gains from expanding. Whether that bet pays off depends on Samsung and Micron exercising the same restraint — and in a prisoner's dilemma, history says few players hold back for long.
Content is for reference only, not financial advice.