Study · Semiconductors & Memory · August 2026
The NAND Supercycle
Sandisk just reported fiscal Q4 revenue of $8.97 billion, up 371.6% year over year, and beat EPS estimates by nearly 15%. That is not a story about Sandisk executing better. It's a story about a commodity — NAND flash memory — going through the steepest pricing cycle in its history, at the same moment AI datacenter buildout is colliding with years of underinvestment in memory fab capacity. SNDK (9.48% of my book, my largest position) and MU (5.61%) are both, structurally, the same trade: long a memory cycle I don't control and can't time precisely, on the belief that AI-driven enterprise storage demand is a structural shift, not just a spike.
1. How steep, exactly
Every memory upcycle sounds dramatic in the moment. What separates this one is the size of the quarter-over-quarter moves, revised upward multiple times this year as TrendForce kept underestimating how tight supply actually was. NAND contract prices rose roughly 55–60% quarter-over-quarter in Q1 2026 — itself a revision up from an initial 33–38% estimate — then accelerated again to a forecast 70–75% in Q2. DRAM moved even faster: initial Q1 estimates of 55–60% QoQ were revised to 90–95%.
| Quarter | NAND Flash QoQ | Conventional DRAM QoQ |
|---|---|---|
| Q1 2026 | +55–60% | +90–95% |
| Q2 2026 | +70–75% | +58–63% |
| Later 2026 (TrendForce projection) | +8–13% | +8–13% |
2. Why now: two demand curves crossed
The driver isn't a single event, it's a structural collision. Enterprise and datacenter SSD demand — the storage layer of the same AI buildout covered in Data Center Alley — has become the primary growth engine for the entire NAND industry, pulling wafer capacity away from consumer applications. Micron has disclosed it can fulfill only 55–60% of core customer demand at current output. Meaningful new fab capacity isn't expected until late 2027 or 2028 at the earliest, because memory fabs take years to plan and build — supply cannot respond to a demand shock the way a software company can just hire more engineers.
The result is a genuinely bifurcated market. Enterprise MLC/SLC SSD contract prices have surged on AI-inference demand, while consumer TLC SSD spot prices actually fell 30–40% in early 2026 on weak DIY-PC and smartphone demand — and even that consumer segment has since reversed, with 1TB consumer SSDs roughly doubling from about $45 to nearly $90 as manufacturers reallocate capacity toward the more profitable enterprise tier. Both SNDK and MU sell into both tiers, but the enterprise/datacenter mix is where the margin expansion is actually coming from.
3. Two holdings, same trade, different exposure
SNDK and MU are the two ways I have this exposure, and they aren't identical bets even though they're both "long NAND."
Sandisk — 9.48% of book
A NAND-flash pure play since its 2025 spinoff from Western Digital — no DRAM business to diversify the cycle. Fiscal Q4 revenue of $8.97B (+371.6% YoY, +51% sequentially) and non-GAAP EPS of $39.25 beat consensus by nearly 15%. Because it's pure-play, SNDK's results are the cleanest single read on NAND pricing of anything I hold — and the most exposed if pricing reverses. See the full FQ4 2026 report.
Micron — 5.61% of book
A diversified memory maker across both DRAM and NAND, including HBM (high-bandwidth memory) for AI accelerators — a product line SNDK doesn't have. That diversification cuts both ways: less pure-play NAND torque than SNDK, but a second, distinct AI-driven demand curve (HBM for GPUs, not just SSD capacity) that SNDK doesn't participate in at all.
4. The bull case and the bear case, stated honestly
This time the demand shift is structural, not cyclical
Past memory upcycles were driven by PC and smartphone unit demand — genuinely cyclical, prone to inventory whiplash. This one is driven by enterprise AI datacenter buildout, which multiple hyperscalers have committed hundreds of billions of dollars to over multi-year horizons (see The AI Capex Reality Check). If that capex holds, enterprise SSD and HBM demand doesn't collapse the way consumer demand cycles typically do.
NAND has never once escaped its own cycle
Every NAND upcycle in the industry's history has been followed by a downcycle, usually because the same pricing signal that's rewarding SNDK and MU today is also the signal that tells every memory maker on earth to build more capacity. Capacity additions arrive with a multi-year lag and then land all at once — the classic commodity-cycle setup for a painful oversupply correction two to three years out.
The two readings describe different time horizons, not a contradiction
The structural AI-demand story can be real over a 5–10 year horizon while the pricing cycle still corrects sharply over the next 18–36 months, once currently-planned 2027–2028 fab capacity comes online. A structural bull case does not exempt a commodity producer from its own commodity cycle — it just means the trough of the next cycle is likely to sit at a structurally higher demand floor than the last one.
5. What would actually change my mind
I'm not going to pretend I can time a commodity cycle better than the industry itself, which has a poor track record of doing exactly that. What I am tracking, as falsifiable markers rather than a vibe: TrendForce's own projected deceleration to single-digit QoQ price growth in the second half of 2026 actually materializing (it would confirm supply is catching up, right on schedule); any signal of new fab capacity commitments being pulled forward ahead of the currently-expected late-2027/2028 timeline (bullish for a longer cycle); and, most importantly, whether SNDK and MU's own forward guidance starts citing softening order books rather than continued shortage — that's usually the first real tell a memory cycle is turning, well before the headline pricing data confirms it.
Related research
This is the storage layer of the same AI-infrastructure buildout examined from the power/cooling side in Data Center Alley and the hyperscaler-capex side in The AI Capex Reality Check. SNDK's and MU's own concentration inside AEA's broader semiconductor cluster is discussed in Return Concentration.
Methodology & sources
Cumulative NAND price increase (+246% since start of 2025). Cited by Kingston's datacenter SSD business unit manager, as reported in Utmel, "The 2026 Memory Super-Cycle."
Quarterly contract-price increases. TrendForce press releases: Jan 5, 2026, Feb 2, 2026 (Q1 revision), and Mar 31, 2026 (Q2 forecast), cross-checked against Tom's Hardware's coverage. Later-2026 deceleration projection per the same TrendForce coverage, aggregated in Sourceability's quarterly tracker.
Micron fulfillment rate (55–60% of core demand). NAND Research, "Memory & Flash Crisis: March 2026 Update."
Consumer SSD price reversal. Consumer TLC spot-price decline and subsequent 1TB SSD price roughly doubling ($45→$90): Oreton Storage, "Why SSD and DDR5 Prices Are Still Rising in 2026."
Fab capacity timeline (late 2027/2028). Avnet, "Riding the AI Supercycle: Navigating the 2026 Memory & Storage Market."
SNDK fiscal Q4 2026 figures. AEA's own SNDK FQ4 2026 earnings report, sourced from Sandisk's own investor-relations release.
Chart. Hand-built SVG using the midpoint of each quarter's cited TrendForce range; 2H26 bars are a projection, not a confirmed outcome, and are shown at reduced opacity to reflect that.
Limitations. Contract and spot pricing figures vary by source, product tier, and methodology; this piece cites ranges rather than single precise figures wherever the underlying sources disagreed. Memory pricing is a genuinely volatile, fast-moving data series — figures here may already be stale by the time you're reading this.
Not investment advice. Nothing here is a recommendation to buy, sell, or avoid any security.