Kioxia Begins Mass Production of 10th-Gen Flash Memory as AI Inference Demand Drives NAND Recovery
Taylor Wilson
Kioxia will formally mass-produce its 10th-generation BiCS flash on July 4, capping a seven-fold stock surge that pushed its market cap past $250 billion — above Toyota. This means → the shift from AI training to inference is turning once-neglected NAND into the tightest bottleneck in memory.
How did Kioxia go from "Japan's chip-industry poster child for trouble" to the hottest name in memory?
Kioxia will hold a mass-production ceremony on July 4 at its Kitakami plant in Iwate, Japan, for the 10th-generation BiCS flash co-developed with SanDisk.
The company, formerly Toshiba Memory, invented NAND flash in the 1980s but spent years under pressure from falling memory prices. Bain Capital, which led a ¥2 trillion (~$12.3 billion) buyout in 2018, repeatedly delayed the IPO until late 2024.
This means → Kioxia's turnaround is not a company-level reinvention. The entire demand structure shifted — AI inference promoted NAND from a supporting role to the lead.
What does AI inference actually have to do with NAND flash?
AI applications are expanding from model training into inference — in plain terms = training teaches the AI; inference is the AI answering your questions.
Inference workloads require large-capacity NAND storage to handle massive query volumes, yet industry capex had been concentrated on DRAM and HBM (high-bandwidth memory), leaving NAND investment well behind.
This reflects a mismatch: everyone raced to build "fast memory," but nobody stocked enough "big warehouses" — and inference needs the warehouse most.
How far ahead is Kioxia's technology?
Iwai Cosmo Securities analyst Kazuyoshi Saito says Kioxia's lead in NAND performance and power efficiency is two to four years ahead of rivals, driven by techniques such as wafer bonding — joining two wafers directly, face-to-face.
Consultant Satoru Oyama notes that other chipmakers over-prioritized DRAM and pushed NAND investment and R&D to the back seat.
Put simply = rivals haven't even started catching up, while Kioxia has already finished the exam — "They simply cannot cope with the current NAND demand boom; that is why demand is concentrated in Kioxia alone."
The stock is up seven-fold — where is the money coming from, and can it keep going?
Kioxia's share price has risen more than seven-fold year-to-date, pushing its market cap above $250 billion — surpassing Toyota.
The company says it is considering a stock split and plans to list ADRs on a U.S. exchange starting in the fiscal year beginning April 2027. South Korea's SK Hynix is also planning a U.S. listing, targeting up to $29.4 billion in proceeds.
This means → memory chipmakers are collectively gravitating toward U.S. capital markets, competing for AI-driven valuation premiums.
How long can this NAND rally last?
The key variable is a race between two speeds: how fast AI inference demand expands vs. how fast rivals ramp up NAND capacity.
If competitors close the capacity gap faster than inference demand grows, Kioxia's supply-demand window shuts — and the valuation premium with it.
In plain terms = Kioxia is feeding on a "nobody else was ready" dividend. How long that dividend lasts depends on how quickly everyone else wakes up.
Content is for reference only, not financial advice.