Kioxia Holdings Corp, in partnership with SanDisk Corp, recently unveiled a 5 trillion yen capacity expansion roadmap spanning through fiscal 2033. According to a research note published by UBS Group AG on August 28, the announcement carries a crucial strategic signal: while Kioxia possesses the technical and financial capacity to expand production, it has no intention of doing so immediately. This positioning, UBS argues, is designed to serve as a powerful bargaining chip in long-term agreement (LTA) negotiations with hyperscale cloud providers.

Currently, the continuous expansion of context windows in AI models is driving robust growth in inference-side NAND demand, with no signs of deceleration on the horizon.

UBS projects that NAND supply and demand dynamics will tighten further in 2027 compared to the prior year. This suggests hyperscale customers will likely be the more concessionary party at the negotiating table, enabling Kioxia to secure favorable terms in its LTA discussions.

How the 5 Trillion Yen Will Be Allocated

Of the total 5 trillion yen investment, Kioxia will contribute 60%, with SanDisk funding the remaining 40%. A substantial 1.8 trillion yen has been explicitly earmarked for the new K3 fabrication facility, where land preparation has already commenced and production is scheduled to begin in fiscal 2030. The designated site for the remaining 3.2 trillion yen in capital expenditures has yet to be disclosed.

The investment timeline is structured in two distinct phases. During fiscal years 2027 through 2029, the combined annual spending will remain consistent with the previously announced 1.4 trillion yen figure, averaging approximately 450-500 billion yen per year. However, some expenditures originally planned for fiscal 2029 will be pulled forward into fiscal 2027 and 2028. From fiscal 2030 through 2033, annual investment levels are expected to rise substantially, reaching approximately 1.25 trillion yen per year.

UBS estimates that these investments will add 200,000 to 220,000 wafers of new monthly capacity for the 332-layer BiCS10 generation, and between 180,000 and 200,000 wafers for the 4xx-layer generation. Based on a monthly capacity baseline of 350,000 wafers at the end of fiscal 2028, full deployment of these investments would allow capacity to recover to approximately 550,000 wafers, keeping the risk of market share erosion relatively contained. Kioxia has also modestly revised its bit growth outlook upward, from roughly 20% to slightly above that threshold.

Why This Serves as a Negotiating Advantage

Hyperscale customers historically prefer that Kioxia accelerate its expansion timeline to secure supply certainty in their long-term agreements. However, the tone of Kioxia's latest announcement suggests the opposite approach: the company has publicly demonstrated both its expansion capability and its detailed plans, while deliberately adopting a patient timeline for actual production commencement.

Under current conditions of constrained supply, keeping production capacity on hold shifts the balance of power toward the supply side. UBS therefore concludes that Kioxia is well-positioned to finalize long-term agreements on favorable terms, with hyperscale customers likely needing to offer concessions on both pricing and contract duration to lock in future NAND supply.