Key Takeaways:
Kioxia's 332-layer NAND chip delivers 59% higher bit density and 30% faster data transfer, challenging Samsung and SK Hynix in the AI storage market.
Kioxia Holdings Corp. began sampling its 10th-generation BiCS FLASH memory for AI data centers, a 332-layer chip with 59% higher bit density and 4.8 Gb/s interface speed that challenges Samsung Electronics Co. and SK Hynix Inc. in the NAND market.
"We see no signs of weakening demand from data center customers," Chief Executive Officer Hiroo Ota said at a media event, adding that Kioxia may increase capital expenditure to meet growing AI storage needs.
The 1Tb TLC device, manufactured at Kioxia's Kitakami Plant Fab2 in Iwate Prefecture, uses CMOS directly Bonded to Array technology to achieve a 33% interface speed improvement over the 8th generation. Write power efficiency improved 18 percent and read power efficiency 30 percent, according to the company. Mass production is scheduled for 2027.
Kioxia's market value has surged more than sevenfold this year to exceed $250 billion, surpassing Toyota Motor Corp., as the shift from AI training to inference drives demand for high-capacity NAND storage. The company lacks a high-bandwidth memory product line — a strategic weakness versus Korean rivals who bundle NAND with HBM — but analyst Kazuyoshi Saito at IwaiCosmo Securities estimates Kioxia holds a two- to four-year lead in NAND performance and power efficiency.
Why 332 Layers Instead of 400+
Kioxia's decision to stop at 332 layers — while competitors push toward 400-plus — reflects a deliberate trade-off. Atsushi Inoue, general manager of Kioxia's memory business unit, said stacking beyond 400 layers increases power consumption as more storage layers activate during read-write operations, and thinner cell layers can reduce charge retention. The 332-layer design delivers about 10 percent lower cost per gigabyte, 10 percent better power efficiency and 35 percent higher reliability compared with 400-plus layer alternatives, Inoue said.
The CBA technology, introduced with the 8th generation, has pushed interface speeds from 3.6 Gb/s to 4.8 Gb/s, a lead that EE Times Japan estimates gives Kioxia roughly a one-year advantage over competitors on data transfer performance.
Korean Rivals Expand Capacity
While Kioxia advances its technology, SK Hynix and Samsung are moving to expand NAND production capacity. SK Hynix on July 2 announced plans to invest 100 trillion won ($615 billion) in a new complex in Cheongju, with 80 trillion won allocated to an M17 NAND fabrication facility. Chief Executive Officer Kwak Noh-jung said NAND demand is growing rapidly while supply remains tight, with construction starting next year and operations targeted for the first half of 2029.
Samsung is also planning a new NAND production line at its P5 campus in Pyeongtaek, with clean rooms expected to be completed next year, according to a report by The Bell in April. The simultaneous expansion by the top three NAND manufacturers has raised concerns about potential oversupply and pricing pressure — a factor that weighed on Kioxia's stock before the latest announcement.
Kioxia's market share in data center NAND stands at about 10 percent, compared with Samsung's roughly 40 percent and SK Hynix's 30 percent, according to Omdia analyst Akira Minamikawa. However, Minamikawa said Kioxia's chips process data faster than competitors, a metric that US hyperscale data center operators prioritize above all else. "The 10th-generation chip achieves a major breakthrough in this area," he said. "It is highly competitive."
Kioxia shares fell as much as 12 percent intraday before reversing to close more than 10 percent higher, reflecting both the product milestone and Ota's demand reassurance. The stock has gained more than 680 percent year to date. The company is considering a stock split and plans to list American depositary shares on a US exchange early in the fiscal year starting April 2027, aiming to broaden its investor base. SK Hynix is also pursuing a US listing, targeting up to $29.4 billion, as Asian semiconductor firms tap deeper capital pools to fund AI-driven expansion.
This article is for informational purposes only and does not constitute investment advice.
