Kioxia and Sandisk's $31 Billion NAND Plan Produces No New Bits Until Fiscal 2029
Kioxia and Sandisk said Thursday they intend to spend roughly ¥5 trillion, more than $31 billion, in Japan through 2032 to expand flash memory production at the Yokkaichi and Kitakami plants. Kioxia said separately that it has begun site preparation for a third fabrication building at Kitakami. Hiroo Ota carried the plan to Prime Minister Sanae Takaichi’s office in person. The wires ran it as a capacity announcement. The two qualifiers attached to it are more informative than the headline number: the spending is contingent on support from the Japanese government, and Fab3 is not expected to begin operations until fiscal 2029.
The disclosure trail is stranger still. On the same day, Kioxia Holdings published a notice regarding certain media reports stating that reports of a new manufacturing facility at Kitakami were not announcements made by the company or its subsidiary, that it is considering various ways to increase corporate value including new facility construction, and that it will announce promptly if any matter requiring notification arises. That is the listed entity declining to treat the number as material disclosure. Whatever ¥5 trillion is, it has not yet passed through the machinery that would make it a commitment.
Measured against what Kioxia has actually guided, the figure is a step change rather than a continuation. At the June Investor Day the company set capital spending at roughly ¥470 billion a year across fiscal 2026 through 2028, itself up about 66% from fiscal 2025. Spread evenly from fiscal 2026 to 2032, ¥5 trillion implies something closer to ¥710 billion a year, better than half again the guided run rate. The longer comparison is starker. The two companies put more than $50 billion into Japan across the previous 25 years, about $2 billion a year. The new plan runs at roughly $4.4 billion a year over seven. The rate of investment roughly doubles.
Memory bears will file this as the first brick in the wall of supply. The timing argues otherwise. Sandisk’s August 13 Investor Day disclosed long-term pricing agreements with eight customers covering about half of fiscal 2027 bit shipments and roughly two-thirds of fiscal 2028, at floors near $0.29 per gigabyte. Fab3 output arrives after that window closes. It does nothing to the pricing that determines the next eight quarters of earnings, and everything to the terminal multiple the market is willing to attach to them.
The more useful detail sits in the existing buildings. Reporting in June put utilization at Yokkaichi Fab 7 and Kitakami Fab 2 at around 50%, with the company prioritizing equipment installation inside available cleanroom space before adding shells. Buildings do not produce bits. Tools do. Announcing a third shell for 2029 while two existing shells are half equipped tells you the binding constraint has been capital pacing and tool procurement, not physical footprint, and that the pacing is a choice the operator retains. A shell is an option. It can be equipped fast, slowly, or not at all, and the decision gets made against contract prices three years from now rather than today’s.
The competitive structure is what makes the announcement asymmetric between the two partners. Flash Ventures is 49.9% Sandisk and 50.1% Kioxia, and it buys wafers from Kioxia at cost and resells them to both parents at cost plus a markup. Kioxia owns the fabs. Sandisk gets leading-edge NAND, and now the CBA architecture that lets the CMOS and array roadmaps advance independently, while carrying $674 million of net property, plant and equipment against $20.2 billion of annual revenue. That asset-light position is the moat, and it is durable in a way layer counts are not, because a competitor cannot replicate 25 years of shared process development by writing a check. It is also not free. Sandisk’s economic share of ¥5 trillion is roughly ¥2.5 trillion, and under the January agreement that moved the relationship from an at-cost model to compensation for manufacturing services, Sandisk already owes $1.165 billion in installments through 2029 on top of building depreciation prepayments that stood at $881 million outstanding in January. Kioxia collects manufacturing service revenue on a larger footprint. Sandisk funds it. The same press release is an income statement event for one partner and a funding obligation for the other, which is roughly how the two tapes traded.
The government contingency deserves more weight than it is getting. The precedents are up to ¥92.9 billion for Yokkaichi Fab 7 in 2022 and up to ¥150 billion across Yokkaichi and Kitakami in 2024. Asking Tokyo to underwrite a proportional share of ¥5 trillion is a request an order of magnitude larger than anything the program has previously granted a NAND producer, which is why the plan was delivered to the prime minister’s office rather than a ministry. If the subsidy comes in below the implied ask, the number does not get revised down in public. The schedule stretches instead.
Sandisk closed Wednesday at $1,474.99, about 37% below the June 22 record of $2,354.39, on a 52-week range running from $47.40 to that high and a market capitalization near $217 billion. Consensus across 24 analysts sits around $2,125, with a high of $3,600 and a low of $1,000, and Mizuho trimmed to $1,875 from $1,900 on Monday. Kioxia rose better than 5% on the Tokyo session. Base case is $1,400 to $1,800, where the announcement is correctly priced as a 2029 event and the stock trades on contract NAND pricing and the November print. Bull case retests the record on evidence that the subsidy is committed at scale and that Kioxia guides bit growth below what the contracted book implies, which would mean the capital is buying share rather than surplus. Bear case is cohort derating to $1,000 to $1,150 alongside Micron and SK Hynix if the market reads ¥5 trillion as the top three and CXMT all committing to 2029 and 2030 capacity simultaneously, in which case the terminal multiple compresses no matter what the next four quarters print.
The number to watch is not the ¥5 trillion. It is the Flash Ventures building depreciation prepayment commitment schedule in Sandisk’s next 10-Q. That is where a real capacity decision surfaces as a dated, dollar-denominated obligation with a fiscal year attached to each tranche. Until that table grows, the $31 billion is a subsidy application with a press release around it.