Kioxia's 29x Profit Jump Sets Up NAND Capex Race With Rivals

Kioxia's turnaround is one of the loudest signals so far that the AI trade is not only a GPU story. According to Bloomberg, the Japanese NAND flash maker that Bain Capital rescued out of the Toshiba wreckage is projecting operating profit to surge roughly 29 times in the first quarter of fiscal 2026, to about ¥1.3 trillion, or around $8.2 billion, on demand from AI data centers.

The scale of the revaluation is easier to see through equity than through earnings. Kioxia's cumulative gains from its late-2024 Tokyo IPO have at points exceeded 4,800%, making it the best-performing constituent in the MSCI World Index. In mid-June its market capitalization briefly peaked at around ¥56 trillion, momentarily passing Toyota to become Japan's most valuable listed company. Bain Capital, which led a consortium that acquired Toshiba Memory for $18 billion in 2018, has now fully exited the position, an eight-year investment that reportedly generated profits exceeding $15 billion, according to Bain Capital managing partner David Gross.

The reason the storage layer has this much open lane right now is competitive as much as it is demand-driven. Samsung and SK hynix are reportedly limiting NAND investment to route wafers into HBM, where pricing is even better. That leaves Kioxia, its partner SanDisk, and China's YMTC, none of them meaningful DRAM players, as the suppliers actually adding NAND capacity into this cycle. The Kioxia/SanDisk alliance is reportedly stepping up capex by about 41% year-on-year to roughly $4.5 billion, focused on BiCS8 production and BiCS9 R&D.

The honest caveat is that a 29x profit quarter is a spot-price story dressed up as a structural one. NAND has historically been the most cyclical of the memory segments, and the reporting does not give you the shape of demand beyond fiscal 2026, nor what happens to pricing if Samsung and SK hynix redirect wafers back once HBM builds normalize. It also does not tell you what Kioxia's post-Bain shareholder register looks like, or whether the current capex plan is actually big enough for hyperscaler order books that are still growing.

What is clear is that the storage tier, long treated as an afterthought behind GPUs and HBM, now has a scarcity story of its own for at least one more cycle. If you are watching where the next round of AI infrastructure margin sits, the NAND makers that stayed in NAND are the ones to track.