On August 27, 2026, Kioxia and SanDisk announced they expect to invest a total of roughly ¥5 trillion in Japan through 2032. Premised on government support and market conditions, the two companies will continue strengthening manufacturing equipment, related infrastructure, and technology at their Yokkaichi and Kitakami plants. The same day, they also revealed plans to build a third manufacturing building, "K3," at Kitakami, aiming for operation sometime in fiscal 2029. The new figure—equivalent to 55.6% of the roughly ¥9 trillion invested domestically over the past 25 years—marks a shift toward a long-term plan for expanded output aimed at meeting AI-driven storage demand.
However, the roughly ¥5 trillion is not the total value of equipment already on order. The companies have not disclosed year-by-year spending, the split between the two firms, the amount of government subsidy, or the resulting increase in production capacity. Even the timing of equipment installation at K3 will be decided based on market conditions. To understand what lies behind this headline figure, it helps to separate two things: the joint-venture framework the two companies have maintained for over 25 years, and the technology investments aimed at increasing the memory capacity extracted from a single wafer.
The ¥5 trillion will be deployed in stages through 2032
The roughly ¥5 trillion figure equals 55.6% of the cumulative roughly ¥9 trillion invested over the past 25 years. The intent to concentrate large-scale investment into a shorter period is clear, but the two companies have not fixed the spending all at once. The announcement states they "expect to invest," premised on government support, and explicitly notes that the pace of equipment installation will be adjusted according to market conditions.
The K3 building, announced the same day, reflects this conditional approach concretely. Kioxia will begin site preparation on the south side of the K2 building at its Kitakami plant, targeting operation sometime in fiscal 2029. Details including the construction timeline and investment in manufacturing equipment will be decided later. The land will be prepared first, while the timing of construction and equipment installation will depend on demand.
At Kitakami, production expansion is already underway without waiting for K3. The K2 building began operating in September 2025, starting with 8th-generation memory before beginning production of 10th-generation 3D flash memory in July 2026. As equipment installation and technology upgrades continue at existing buildings, K3 will expand the physical capacity beyond that. Investments with different time horizons overlap within this single funding envelope.
When comparing figures, it's also important to note differences in scope.
| Figure | Scope | What it doesn't tell us |
|---|---|---|
| ~¥5 trillion | Investment Kioxia and SanDisk expect to make in Japan through 2032 | Year-by-year breakdown, split between the two companies, subsidy amount, capacity increase |
| ~¥470 billion per year | Kioxia's capital expenditure plan for the next three years, announced in June 2026 | SanDisk's share, total through 2032 |
| Up to ~¥242.9 billion combined | Combined cap of two past government subsidies approved in 2022 and 2024 | New subsidy amount tied to the current ¥5 trillion figure |
The roughly ¥5 trillion figure and the roughly ¥470 billion annual figure don't share the same number of companies, time period, or expense categories. They cannot simply be added together to produce a total investment figure, nor can either be converted into a straightforward annual rate. This announcement establishes an upper bound on scale, but disclosure of the actual flow of funds has yet to come.
A 50.1–49.9 joint venture links investment burden to production scale
At the core of manufacturing sit three joint ventures: Flash Partners, Flash Alliance, and Flash Forward. Kioxia holds a 50.1% stake in each, with SanDisk's side holding 49.9%. These joint ventures own or lease the manufacturing equipment, while Kioxia produces wafers at its Yokkaichi and Kitakami plants. The finished wafers are then allocated to both companies, which each turn them into SSDs and embedded products to sell to customers.
This arrangement allows the two companies to share the same front-end processes and technology development while maintaining their own product strategies and sales networks. Jointly procuring expensive manufacturing equipment makes it easier to scale up production volume than either company could achieve supporting a plant alone. At a 2025 briefing, Kioxia explained that 20% of plant capacity is allocated solely to itself, with the remaining 80% split evenly between the two companies—resulting in a total capacity allocation of 60% for Kioxia and 40% for SanDisk.
The joint venture's term has also been extended in line with the large-scale investment. In January 2026, the two companies extended the Yokkaichi contract by five years, from the end of 2029 to the end of 2034, aligning it with the Kitakami contract's end date. In exchange for continued manufacturing services and supply, SanDisk will pay Kioxia $1.165 billion between 2026 and 2029. This secures a contract term covering the equipment expansion through 2032 and continued joint production afterward.
That said, joint operation also locks in mutual dependence. If judgments diverge anywhere across equipment investment, technology transitions, or demand forecasting, neither company can complete its supply plan alone. The 25-plus-year track record and the contract extending through 2034 are stabilizing factors, but since the method for sharing the roughly ¥5 trillion burden has not been disclosed, it remains impossible to judge how far investment risk will actually be shared.
332-layer 10th-generation memory boosts capacity per wafer
The 10th-generation BiCS FLASH now in production at the Kitakami K2 building is a 1Tb TLC product built by stacking 332 memory cell layers. According to Kioxia's published figures, the NAND interface speed reaches 4.8Gb/s. Combined with increased planar density, bit density has been raised by 59%. The company says power efficiency improved by 18% during writing and 30% during reading.
Manufacturing capacity cannot be measured by wafer input alone. Higher bit density allows more memory capacity to be extracted from the same wafer area. However, chip yield, product mix, and equipment utilization rates all affect actual supply volume, so a 59% increase in density does not directly translate into a 59% increase in shipment volume.
CBA technology, used since the 8th generation, forms the memory cell array and the control CMOS circuitry on separate wafers before bonding them together afterward. Because the memory side and the circuit side can each be manufactured using processes suited to them, it becomes easier to combine vertical improvements—increasing the number of stacked layers—with horizontal improvements that pack circuits and cells more densely. Plant investment goes not only toward constructing new buildings but also toward the equipment, cleanrooms, and infrastructure required for each generational transition.
As of July 3, 2026, the 10th-generation product was still at the sample-shipment stage, meaning specifications could change once mass production begins. Even so, production of this generation has already begun at the K2 building. The results of the roughly ¥5 trillion investment should be measured not by the number of buildings, but by the speed of transitioning to new generations and the time it takes to reach stable yields of good chips.
AI inference is pushing NAND demand, but market volatility hasn't disappeared
One tailwind for this investment is the growing volume of data handled by AI inference. Even when GPUs handle the computation, storage devices are needed to hold training data, models, and intermediate computation results. As inference services scale to serve large numbers of users, demand grows for fast, high-capacity enterprise SSDs deployed across data centers. Kioxia has set a mid- to long-term target of raising its data center and enterprise sales ratio to over 60%.
Current supply-demand conditions are strong. According to TrendForce, the enterprise SSD market in the first quarter of 2026 saw revenue rise 86.1% quarter-on-quarter to over $18.46 billion, with contract prices up roughly 80%. Kioxia reportedly saw certification and mass production progress with North American customers for existing-generation products, while SanDisk also increased shipments of high-capacity products. This tight supply-demand balance is the reason both companies are moving to secure supply capacity through 2032 in advance.
However, rising prices reflect both strong demand and supply shortages at the same time. NAND is a cyclical industry where oversupply drives down prices and makes investment recovery difficult. Even as AI-related demand grows, fluctuations in PC, smartphone, and consumer storage demand remain. This is why both companies say they will decide the construction timeline for K3 and the details of equipment investment based on market conditions.
There is no guarantee that AI demand will continue at the same pace through 2032. If the current surge in enterprise SSD prices were treated as a permanent growth rate, expanded production could lead to oversupply. What the two companies have announced is an investment envelope that can flex with demand—not a plan to spend the same amount every year.
Government support drives investment, but the new subsidy amount remains unspecified
The Japanese government has already supported joint production between the two companies. In 2022, it approved a subsidy of up to roughly ¥92.9 billion for Yokkaichi; in 2024, it approved up to ¥150 billion for 8th- and 9th-generation production at both Yokkaichi and Kitakami. The latter also served the policy goal of spreading production across two regions to reduce supply risk in the event of a disaster.
The scale of current policy is even larger. The government's "AI and Semiconductor Industry Foundation Strengthening Framework" calls for over ¥10 trillion in public support over the seven years through fiscal 2030, aiming to spur more than ¥50 trillion in combined public and private investment over ten years. Kioxia and SanDisk's plans align with this direction of expanding Japan's advanced semiconductor capacity.
Still, the combined roughly ¥242.9 billion cap from the two past subsidies cannot simply be added as the support amount for this new plan. The August 27 announcement did not name any new subsidy program or amount. While K3 is also premised on government support, its construction cost and equipment installation timeline remain undisclosed. How far public funds will cover buildings, equipment, and R&D—and how much sustained production the government will require from the companies in return—will determine the conditions under which the ¥5 trillion figure becomes actual spending.
The credibility of this investment plan should be verified, in order, through subsidy decisions, annual spending disclosures, equipment shipments to K3, and increases in bit supply—well before looking at the total figure. If these are disclosed and new-generation products can be brought online in step with demand, Japan's two production sites could become a long-term manufacturing network supporting storage supply in the AI era.