Can SanDisk Avoid the Memory Trap That’s Burned Investors Before?
SanDisk just committed to a $31 billion NAND expansion at the exact moment investors are asking whether the AI memory boom is just another cycle waiting to collapse. The answer hinges on a contract structure unlike anything the industry has…
SanDisk (NASDAQ:SNDK | SNDK Price Prediction) and Japanese partner Kioxia said this week they will invest more than $31 billion in Japan over six years to expand NAND flash capacity for the AI era, including roughly $11.3 billion for a new facility at Kioxia’s Kitakami plant to produce jointly developed tenth-generation BiCS Flash.
This is a joint figure covering both partners, and the company’s share has not been disclosed. The plan leans on Japanese government support, so treat it as a proposal at this stage. The reason it matters is that this is a NAND bet, while most AI headlines focus on high-bandwidth memory. NAND is used in enterprise SSDs that store training data, checkpoints, and inference results. Shares closed at $1,484.98 on Friday, up 525.57% year to date, so the market has already priced in significant upside.
Why This Bet Looks Different From Past NAND Cycles
Memory has burned investors before because capacity gets built into peak pricing and arrives during a glut. That history is the single most important context for this deal.
What differs now is that SanDisk is locking down demand before pouring concrete. Management said it has signed new business model agreements with eight data center and edge customers, with a weighted average duration of over four years.
CEO David Goeckeler put the visibility bluntly: “A year ago, we were talking about visibility in this business of three months,” and now the company has “over four years of visibility.”
The backlog is real. Remaining performance obligation stood at $59.8 billion and, including two agreements signed after quarter-end, reached $91.1 billion, backed by $16.5 billion in customer financial guarantees. Total minimum expected NBM revenue at floor pricing is $93.9 billion. Those are contracted floor economics tied to fixed volumes.
Numbers Behind the Rally
Fiscal fourth-quarter revenue reached $8.96 billion, up 371.59% year over year, with non-GAAP EPS of $39.25 against a $33.28 estimate. Datacenter revenue rose 437% for the full year, and non-GAAP gross margin expanded to 84.6% from 26.4% a year earlier.
Roughly two-thirds of sequential revenue growth came from pricing, which reverses fastest when supply catches demand. Free cash flow of $11.494 billion and a zero-debt balance sheet give SanDisk room to fund its share of Kitakami without breaking the model. The board authorized an additional $14 billion in buybacks.
At a forward P/E near 23x, the stock is priced for NBM economics to hold, not for a return to spot NAND volatility.
Verdict on the $31 Billion Question
Can AI demand absorb this capacity without recreating the oversupply that has repeatedly hurt memory pricing? Likely yes, provided the NBM structure holds, and hyperscaler forecasts do not compress. Management expects bits to remain in allocation beyond calendar year 2027, and analysts note a structural memory shortage that is unlikely to ease before 2028. New Kitakami capacity arrives inside that window.
The risks are the ones the Palo Alto Networks CEO flagged this week: structural memory shortage unlikely to ease before 2028. Reddit sentiment has already turned on concerns that the memory supercycle will eventually become cyclical. The counter is that bearish will sit inside contracts with floors, insulating the mix even if spot NAND rolls over.
The bet is defensible because SanDisk is expanding into demand it has already sold. That is the argument for owning the stock through the next headline about cycle risk.
Contact [email protected] for any questions or corrections.







