Memory Makers Race to Expand Capacity as Kioxia Sticks to Technology-First Strategy
Claire Weston
AI demand is pushing Samsung, Micron, and SK Hynix to spend a combined $1,190 billion on capex in 2026, while Kioxia commits just $3 billion — a stark split between capacity brute-force and a technology-first bet that will reshape the NAND market.
How much is each player spending?
2026 capex: Samsung $51.7 bn, Micron $34.5 bn, SK Hynix $32.8 bn. Kioxia: just $3 bn.
This means → Kioxia's annual spend is less than one-sixteenth of Samsung's — a gap that is no longer in the same league.
Through fiscal 2028, Kioxia plans roughly ¥1.4 trillion (~$8.7 bn) in total investment, still far below the leaders — and shareholders keep pressing for more.
How aggressive are the rivals' expansion plans?
Samsung and SK Hynix announced in late June 2026 a joint ₩800 trillion (~$554 bn) investment to build four fabs in South Korea, aiming to double the country's memory capacity within five years.
Micron said in July it would raise its U.S. memory production and R&D plan from $200 bn to $250 bn, extending the timeline to 2035.
In plain terms = the Korean duo plus Micron are building capacity moats with super-cycle-level capital. If Kioxia cannot keep pace, its share will be physically squeezed.
Can Kioxia's finances hold up?
In Q4 2026, Kioxia's free cash flow reached ¥74.9 bn — roughly 28× year-on-year. The recovery is fast.
Samsung generated about ¥600 bn and SK Hynix about ¥400 bn in the same period — Kioxia's war chest is still an order of magnitude smaller.
This reflects a reality: Kioxia is healing, but it does not have the ammunition for a drawn-out capacity war against the top two.
Why is Kioxia choosing not to follow?
Management stated on the July 31 earnings call: no blind expansion for market share.
The strategy rests on three pillars: deepen the Nvidia partnership + lock in long-term customer contracts + scale advanced NAND products toward broader commercialization.
Management's view: agentic AI — AI systems that execute tasks autonomously — is accelerating, and memory demand is only in its early innings. Better to build a technology moat now than to chase volume.
How does this bet play out?
Kioxia's path = speed of technology iteration > scale of capacity. The wager is that advanced NAND products can command a premium in commercialization.
This means → if advanced products ramp smoothly, Kioxia can lever higher margins from less capital; if commercialization stalls, the capacity giants will widen the share gap further.
In plain terms = Kioxia is running a "small but sharp" playbook — yet in memory's history, "sharp" beating "big" is far from a sure thing.
Content is for reference only, not financial advice.