In April-June 2026, Japanese Kioxia Holdings expects net profit of 869 billion yen — approximately $5.6 billion. A year earlier during the same quarter, the company earned 18.27 billion yen ($118 million). Operating profit, according to forecasts, will grow 29 times — to 1.29 trillion yen ($8.3 billion).
But to understand what is really happening, it's worth looking not forward, but back — to autumn 2025.
A company that was just emerging from crisis
In November 2025, Kioxia reported a 62% year-over-year profit decline — and analysts recorded this as the bottom of the cycle. The quarterly result then fell short of even the consensus forecast of 47.4 billion yen. The reason — a failed product mix: seasonal smartphone demand was pulling the company into low-margin segments, while data centers were already transitioning to enterprise storage solutions.
The turnaround came not through luck, but through a technological bet. According to TrendForce, in the third quarter of 2025, Kioxia showed the strongest quarterly growth among all NAND manufacturers — plus 33.1% — and for the first time surpassed Micron in market share, securing third place with a 15.3% share.
What changed: BiCS8 and the bet on AI servers
The key driver — a large-scale transition to BiCS8 technology (218-layer 3D NAND) and growing supplies of enterprise SSDs for AI servers. This segment, not smartphones, determines margins in the current cycle.
"NAND flash prices in the first quarter of 2026 will increase by 85–90% on a quarterly basis"
TrendForce, December 2025
The reason is structural: while manufacturers redirect production lines to QLC storage devices with capacities of 122–245 TB for AI infrastructure, the supply of consumer memory is shrinking. Demand exceeds supply — and it is in this window that Kioxia is fixing its record forecasts.
The next bet — to replace HBM
In parallel, the company is preparing a technological leap that could change the very architecture of AI servers. According to TrendForce, Kioxia in partnership with NVIDIA is developing SSD storage that will connect directly to GPUs and partially replace HBM (High Bandwidth Memory). The launch is scheduled for 2027; read speeds — 100 times faster than current models.
This is not just a product novelty: if the technology works, Kioxia will enter a field currently monopolized by SK Hynix with its HBM3E. At the same time, SanDisk — a former partner and current competitor — has already reached an agreement with SK Hynix on joint development of an alternative HBF (High Bandwidth Flash). The race is happening on several fronts simultaneously.
TrendForce predicts that by 2026, the NAND shortage will deepen: demand capacity growth will reach high-teens percent against strict supply constraints.
The question is not whether Kioxia will meet its forecast for April-June 2026 — market conditions are currently in its favor. The question is whether the company will manage to monetize the technology cycle before Samsung and SK Hynix scale their own QLC lines and restore price competition — estimated for the second half of 2026.
