Kioxia plays its “negotiation trump card”: to expand production, sign long-term agreements!

Recently, Kioxia, together with SanDisk, announced a 5 trillion yen expansion plan through fiscal year 2033. In its latest research report released on August 28, UBS believes that the key signal from this move is: Kioxia has the capability to expand production, but has no intention of executing immediately—this will become its main bargaining chip when negotiating Long-Term Agreements (LTA) on NAND with hyperscale cloud providers.

Currently, the ongoing extension of AI model context length is driving rapid growth in NAND demand for inference, and there are no signs of slowing down.

UBS judges that NAND supply and demand will tighten further in 2027 compared to 2026. Hyperscale customers are more likely to make concessions during LTA negotiations, and Kioxia is expected to be able to secure favorable terms.

How will the 5 trillion yen be spent?

Of the total 5 trillion yen investment, Kioxia will contribute 60% and SanDisk 40%. 1.8 trillion yen is designated for the new K3 plant, with land preparation already underway and plans for production to begin in fiscal year 2030. The target location for the remaining 3.2 trillion yen has not yet been disclosed.

The investment will be rolled out in two phases. For the 2027-2029 fiscal years, the three-year total will remain at the previously announced 1.4 trillion yen, averaging about 450-500 billion yen per year, but some expenses originally planned for 2029 will be brought forward to 2027 and 2028. For the 2030-2033 fiscal years, annual investment could rise sharply to about 1.25 trillion yen.

According to UBS estimates, this investment can add monthly capacity of 200,000 to 220,000 wafers for the 332-layer BiCS10 generation, and 180,000 to 200,000 wafers for the 4xx-layer generation. Based on a projected monthly capacity of 350,000 wafers at the end of fiscal 2028, after the investment is complete, capacity is expected to recover to 550,000 wafers, with a low risk of losing market share. Kioxia has also slightly raised its bit growth expectation from around 20% to just above 20%.

Why is this considered a "negotiation trump card"?

In LTA negotiations, hyperscale customers naturally hope Kioxia will expand production in advance to secure supply. But Kioxia's latest announcement shows the opposite attitude: expansion capability and plans are made clear, yet the production schedule is not rushed.

In a tight supply environment, keeping production capacity "on hold" shifts the initiative to the supply side. Accordingly, UBS believes that Kioxia is likely to reach LTAs on favorable terms—if hyperscale customers want to secure future NAND capacity, they will need to make concessions on price and contract duration.