SanDisk's Small Print Points to the Bet Behind a 3,000% Run

The quietest filing in tech this year reads like homework: "SanDisk Enters Amendment No. 1." It is a small addendum to a deal from 2024, when SanDisk — still a unit of Western Digital — sold 80% of its Shanghai packaging plant to a Chinese chip-packaging firm called JCET for about $624 million. Amendment No. 1 only retimes a few deferred-payment details. Yet that obscure paperwork is a window onto the strategy behind one of the most explosive stock runs in the market: from roughly $38 a share at its February 2025 spin-off to about $1,600 today, a market value near $239 billion.

To see why, you have to notice SanDisk did two opposite things with its money at two opposite moments in the memory cycle.

First, the exit. The JCET sale closed in September 2024, when the NAND flash memory business was in the worst downcycle in a decade. Gross margins had collapsed to the single digits and the company was burning through cash. In that environment, offloading a commodity packaging-and-test plant for $624 million and keeping only 20% made sense as survival: SanDisk chose to own the profitable core — the design of NAND chips and the wafer manufacturing it shares with Japanese partner Kioxia — and hand the grunt work no memory buyer pays a premium for to someone else. It is a "capital-light" bet that a memory company's value lives in the technology, not in every factory step. Amendment No. 1, signed in September 2025 and tucked into its regulatory filings around January, merely fiddled with the timing of that plant's remaining payments and a roughly $42 million working-capital figure tied to shifting controller suppliers. In dollar terms, it is now immaterial — the point is what it represents.

Then, the reversal. Just as SanDisk had sold a factory to stay light, the memory market flipped from glut to scarcity, and the AI build-out turned NAND from a commodity into a bottleneck. The logic is simple to state: training a model is about compute, but running it — inference — is about moving and caching data, so storage became the "working layer" of AI rather than the shelf it once was. The earnings landed accordingly. In the quarter reported this August, SanDisk booked roughly $9 billion of revenue and about $39 a share of earnings, up from $5.95 billion the quarter before, with gross margins above 70%. Revenue is up more than 170% year over year, and operating margin is around 60%.

That is where the second capital decision begins. In January, SanDisk and Kioxia extended their flash joint venture through 2034, with SanDisk committing to pay Kioxia $1.165 billion across 2026 to 2029 for manufacturing services. Then in late August the two announced over $31 billion of planned investment in their Japanese fabs through 2032 — contingent on government support. Meanwhile SanDisk has signed long-term customer contracts that analysts describe as roughly $42 billion of minimum revenue, with over $11 billion of financial protection, already covering more than a third of next year's output.

Here is the tension every memory investor has to sit with, and it is the heart of the whole debate. A surge in capacity commitments is a dual signal: it reads as demand strength, but it is also rising leverage and delivery risk at the same time. Two things genuinely separate this cycle from the booms that wrecked memory companies before. The demand is partly contracted — those multi-year agreements with minimum volumes and penalty protection are demand that is promised, not merely hoped for. And supply cannot respond overnight: new NAND capacity takes years of fab construction, so scarcity can persist while demand stays strong. That is the bull case for treating SanDisk differently from the last cycle.

The bear case is the memory playbook, and it has never stopped applying. This is the most cyclical industry in all of technology, and every past boom has ended the same way — companies tripping over newly minted capacity precisely as demand cools, prices collapsing, and the multi-year boom turning into a multi-year hangover. SanDisk's current gross margins, in the 70s, are historically unheard-of for NAND; nothing suggests they are a permanent altitude. And the $31 billion plan is conditioned on government subsidy — a reminder that the whole industry is now leaning on state support to build.

That is why the valuation is more deceptive than it looks. SanDisk trades near 21 times trailing earnings — cheap, at a glance, for a company growing this fast. But that multiple is measured against peak-of-the-boom earnings. The classic memory trap is exactly this: a low P/E on earnings that are at the top of a cycle, which then normalize, making the "cheap" stock expensive in hindsight. The entire question reduces to whether AI inference demand plus the new contracted model have permanently smoothed out the NAND cycle — or whether SanDisk is simply a good company at a cyclical peak, worth more on paper than on normalized earnings.

SanDisk itself has pointed to annual net income of $30 billion to $40 billion down the road. Do the math at a conservative ten times, and that implies a $300 billion to $400 billion market cap — meaning the market is already paying up for much of the bull case before it is delivered. What would change my read is evidence, not price targets: whether those long-term contracts renew at pricing high enough to keep margins elevated, whether competitors add meaningful new NAND capacity, and whether near-80% gross margins hold once today's scarcity fades.

The small print earned its keep here. SanDisk's decision in 2024 was to get lean and survive the downcycle; its decision in 2026 is to spend billions betting the AI upcycle runs for years. One of those was a defensive move a dying market demanded. The other is a boom-time commitment made at the top of the best stretch in the company's history. The thesis that made the stock a 3,000% winner is now fully digested by the market — the question for anyone looking at $1,600 is whether the next few years' return curve still justifies it, which is a different and harder question than the run-up ever was.