Kioxia expects annual profit to multiply by 31: NAND enters the AI supercycle

In plain terms

Kioxia announces a Q1 net profit multiplied by 46 year-over-year, and an annual profit projection 31 times higher than the previous fiscal year. In other words: NAND enterprise is riding the AI data-center wave just like HBM before it. The memory supercycle has a new numerical proof.

Les chiffres

Q1 fiscal press release picked up by ITmedia AI+ and Nikkei Asia on July 31, 2026. Q1 net profit is up 4500% compared to the same quarter of the previous fiscal year. Kioxia projects a 31-fold increase in profit for the entire fiscal year, explicitly driven by AI data-center demand. Timing: it's the same Kioxia that had fallen 16% intraday on July 17 during the mini-AI deleverage (Nikkei Asia, art « Japanese stocks plunge as AI deleverage sends Kioxia down 16% »). The note and the industrial fundamentals do not coincide.

Under the hood

High-density NAND is the natural complement to HBM in a training/inference cluster: HBM serves the working set, enterprise NAND (DC class) holds the checkpoints, the overflow KV-cache, and the in-flight datasets. The HBM saturation that SK Hynix (57% market share of HBM AI-servers according to Tech in Asia, July 26) has monetized for twelve months is spreading to enterprise NAND with a lag - this Q1 Kioxia is the first quarter where this transfer is reflected in a P&L to this extent.

So what

Three consequences. One: the idea of a "memory pinch" limited to HBM is deteriorating - enterprise NAND is entering the same cycle. Two: analyses on the memory peak (including the CXMT passage in Shanghai at +471% on July 27, or previous publications in the memory-chip-capex thread) must integrate that the supercycle is broader and later than anticipated. Three: on the portfolio side, the restrictive "Samsung/SK/Micron" reading on AI memory must be expanded to Kioxia (Japan) and Nanya (Taiwan). To watch: Kioxia's investment capacity in new fabs (H2 calendar), and the effect of NAND price increases on auto OEMs (cross-reference with the TSMC compute crunch thread).