Kioxia and Sandisk plan $31bn Japan NAND push as AI storage race accelerates

Kioxia Corporation, owned by Kioxia Holdings Corporation (TSE: 285A), and Sandisk Corporation (NASDAQ: SNDK) are planning more than $31 billion, or approximately ¥5 trillion, of additional investment in Japan through 2032 as artificial intelligence infrastructure drives demand for higher-capacity and more power-efficient flash memory. The programme will support continuing infrastructure buildout at Kioxia’s Yokkaichi and Kitakami manufacturing sites and associated technology investments, while the companies say each intends to support meaningful multi-year bit growth. Importantly, the plan is contingent on Japanese government support, and the companies have not disclosed how the ¥5 trillion will be split between Kioxia, Sandisk and potentially other financing sources. The announcement significantly extends one of the semiconductor industry’s longest-running manufacturing partnerships at a point when NAND is becoming increasingly important to AI data-centre economics.

The scale becomes clearer when compared with the partnership’s history. Kioxia and Sandisk say they have invested more than $50 billion, approximately ¥9 trillion, in Japan during their previous 25 years together. The newly proposed $31 billion programme therefore equals more than 62% of that cumulative dollar investment, despite covering only the years through 2032. In January, the companies separately extended their Yokkaichi joint-venture framework through December 2034, indicating that the latest spending plan sits inside a long-duration manufacturing relationship rather than a temporary response to one favorable quarter of NAND pricing.

Why does AI infrastructure need so much NAND when GPUs and HBM attract most attention?

Artificial intelligence systems need several different forms of memory and storage. High-bandwidth memory sits close to accelerators and supplies data at extremely high speed during computation, while NAND flash provides persistent storage for training datasets, model checkpoints, retrieval databases, inference content and other information that must remain available when power is removed.

As models and datasets grow, the storage tier becomes both larger and more performance sensitive. Hard disk drives remain economical for some cold-data workloads, but enterprise solid-state drives can offer much higher throughput, lower latency and better performance per unit of rack space. Those advantages become increasingly valuable inside data centres where electricity and physical space have become constraints.

Kioxia and Sandisk explicitly connect the new investment to an AI and data-driven economy and say they intend to expand stable supplies of high-capacity, high-performance and power-efficient flash memory. Their BiCS 3D flash architecture is used across data centres, smartphones, personal computers, automobiles and other devices, meaning the investment is not dependent exclusively on AI. AI infrastructure is nevertheless creating a powerful incremental demand source at a time when storage is moving closer to the performance requirements traditionally associated with compute.

The distinction matters for investors because NAND has historically been highly cyclical. If AI simply produces a temporary shortage, aggressive capacity expansion can eventually contribute to oversupply. If AI structurally increases the amount of flash required per data-centre workload, suppliers can add bits while maintaining stronger utilization for longer.

The $31 billion spending plan effectively wagers that the second interpretation is sufficiently durable to justify another enormous manufacturing cycle.

How large is the new investment relative to the Kioxia-Sandisk partnership’s existing footprint?

Kioxia and Sandisk have jointly developed and manufactured flash-memory wafers in Japan for more than 25 years, principally through Kioxia’s Yokkaichi and Kitakami operations. They estimate cumulative Japanese investment over that period at more than $50 billion. The additional programme exceeds $31 billion and runs through 2032, meaning its headline value is roughly 62% of the previous quarter-century total.

That comparison should not be mistaken for a precise increase in manufacturing capacity. Semiconductor construction costs, equipment prices, technology complexity and currency values have changed enormously over 25 years, while part of the new spending will support infrastructure and technology transitions rather than simply adding identical wafer capacity.

It nevertheless indicates how capital intensive the next NAND generation has become. Advanced three-dimensional flash requires increasingly complex deposition, etching and process-control sequences as manufacturers stack more memory layers vertically. Moving to newer generations can demand enormous equipment expenditure even before a fab produces substantially greater output.

The companies have not published annual investment schedules or the proportion directed separately toward Yokkaichi, Kitakami, new equipment and technology development. That makes $31 billion a long-range spending framework rather than a sum that should automatically be inserted into near-term capital-expenditure forecasts.

Government participation is another unresolved variable. The announcement explicitly says the investment is contingent on government support and aligns with Japanese economic-policy goals. Until subsidy structures and project approvals become clearer, the gross headline number should not be treated as capital funded entirely from the two companies’ balance sheets.

Why is Japan willing to support another enormous flash-memory expansion?

Semiconductors have become strategic industrial infrastructure for Japan as governments around the world attempt to reduce dependence on geographically concentrated supply chains. Japan already possesses deep expertise in semiconductor materials, manufacturing equipment, memory production and advanced industrial processes, making expansion of an existing large-scale NAND cluster less speculative than attempting to build an industry from scratch.

Kioxia and Sandisk describe the planned investment as aligned with Japanese government policy and broader U.S.-Japan economic cooperation. Sandisk is headquartered in the United States, while manufacturing is conducted through the long-standing Japanese joint venture, giving the arrangement geopolitical importance alongside its commercial logic.

Government support can improve investment economics by sharing some of the cost of factories and technology transitions that might otherwise offer insufficient returns during weaker memory cycles. It can also accelerate construction where policymakers place strategic value on domestic production capacity beyond the immediate return available to shareholders.

The trade-off is potential overbuilding. Subsidies can encourage several manufacturers or countries to add semiconductor capacity simultaneously, and memory markets have repeatedly experienced severe downturns when supply growth runs ahead of final demand.

That makes capital discipline crucial even during an AI boom. Kioxia and Sandisk need to expand capacity quickly enough to serve customers without recreating the oversupply conditions that have historically compressed NAND pricing.

Does the $31 billion announcement mean Kioxia and Sandisk are each committing $15.5 billion?

No. The companies disclosed an aggregate expected investment exceeding $31 billion but did not provide a 50-50 allocation of spending. Their manufacturing partnership involves shared economics around wafer production, while individual capital contributions, government support and other funding arrangements can vary according to project structure.

The release also uses forward-looking language and makes the investment contingent on government support. It is therefore more accurate to describe the figure as an anticipated joint investment programme through 2032 rather than an immediately funded $31 billion contractual obligation.

This distinction matters because semiconductor headlines can create the impression that the entire amount will be spent immediately. In reality, a programme extending through 2032 can be adjusted around market conditions, technology transitions, construction schedules and government decisions.

Investors should watch subsequent annual capital-expenditure guidance from both listed companies to understand when the long-range plan begins materially affecting cash flow. Equipment orders, cleanroom expansion, production starts and government subsidy approvals will provide more actionable milestones than the aggregate number alone.

How has the market treated Kioxia and Sandisk around the new spending announcement?

Kioxia Holdings shares closed at ¥47,900 on August 28, falling 8.76% in the latest session. The stock was approximately 11.8% below its August 21 close of ¥54,320, although it remained about 24.8% above the July 29 close of ¥38,380. The share price has been extraordinarily volatile, trading within an approximately ¥2,541 to ¥112,700 52-week range according to recent market data.

Sandisk produced a different reaction. Its shares finished August 28 at $1,484.98, essentially unchanged for the session, but approximately 7% below their August 21 close of $1,596.08. Compared with the July 29 close of $1,015.89, the shares were still about 46% higher. Sandisk’s recent 52-week range stretches from approximately $50.07 to $2,354.39, reflecting similarly extreme investor sensitivity to changes in NAND pricing and AI storage expectations.

Neither company’s short-term stock move should be attributed entirely to the investment announcement. Memory equities have been exceptionally volatile as investors attempt to determine whether current pricing and AI demand represent a structural change or another unusually strong cyclical peak.

The difference between operating demand and capital returns will become increasingly important. Expanding production can capture more AI storage demand, but every new fab and technology transition also raises depreciation and increases future industry supply. The ideal outcome is sustained bit growth combined with enough pricing discipline to earn attractive returns on the new capacity.

The market is already pricing substantial optimism into both companies compared with their earlier 52-week lows. That makes execution, rather than the existence of an AI storage opportunity, the more important debate from here.

What must happen before the $31 billion NAND plan becomes financially convincing?

First, government support has to become sufficiently concrete for the investment programme to proceed on attractive terms. The companies have made support an explicit condition, so subsidy approvals and policy structures become part of the financial model rather than peripheral political developments.

Second, customer demand needs to justify multi-year bit growth rather than merely higher near-term prices. NAND suppliers ultimately earn returns from increasing the amount of storage shipped while maintaining disciplined pricing and manufacturing costs.

Third, technology transitions at Yokkaichi and Kitakami must produce competitive cost per bit. A manufacturer can add capacity and still lose economic ground if competitors achieve denser flash structures, better yields or lower production costs.

Fourth, AI-driven enterprise SSD demand needs to remain durable enough to absorb growing supply. The strongest scenario for the partnership is one in which more AI data-centre spending permanently increases storage requirements while new flash technology lowers power consumption and cost per terabyte.

The investment plan is consequently much larger than a conventional capacity announcement. Kioxia and Sandisk are effectively wagering that storage will remain a critical component of AI infrastructure well into the next decade and that their 25-year partnership can keep pace technologically while adding enormous amounts of production capital.

If those assumptions hold, another $31 billion of Japanese investment could secure a central position in the AI storage supply chain. If memory supply expands faster than demand, the same investment cycle could eventually become the source of the industry’s next pricing downturn.