Article published Jul 13, 2026. Prices below use latest available snapshots.
Question: SanDisk fell 12.2% on a day when the news was about SK Hynix's HBM4 schedule. SanDisk makes NAND. It has less to do with HBM than anything else in the complex. Why did it fall hardest?
Answer: because it is the purest exposure to the one variable that actually moved — the price of memory. The market did not reprice SK Hynix's earnings today. It repriced the memory pricing cycle, and it marked every name in proportion to how much of its revenue floats on spot.
Written intraday and now confirmed at the close. The table below is the settled 2026-07-13 close.
The close confirmed it
The gradient held to the bell, and the tell held with it:
| close | 1-day | vs the intraday read | |
|---|---|---|---|
| SNDK | 1673.97 | −12.6% | closed at the lows — no bounce |
| SKHY | 152.35 | −9.3% | kept falling into the close |
| EWY | 168.02 | −8.4% | held |
| STX / WDC | 860.66 / 555.55 | −5.5% / −4.6% | recovered off the lows |
| MU | 937.00 | −4.3% | recovered — closed at semis beta |
| SMH (semis) | 585.62 | −4.2% | — |
| SPY | 749.17 | −0.8% | — |
Micron closed at semiconductor beta. −4.3% against SMH's −4.2% — the market gave it no memory-specific discount at all, on the single worst day the memory complex has had. SanDisk closed near the lows at three times that loss.
That is the ASP thesis stated as cleanly as a tape can state it. The one name with contractual price floors on ~40% of revenue was priced like a chip company; the one name that is pure spot NAND was priced like the commodity it sells. The protection is not a talking point — it is worth roughly 8 points on a bad day.
It also answers the question we left open this morning — do MU and SNDK follow SK Hynix down, or diverge? MU diverged. SNDK did not.
The gradient is the evidence
| 1-day | RSI | Exposure to spot memory prices | |
|---|---|---|---|
| SNDK | −12.2% | 38 | Pure NAND. Almost entirely spot. |
| SIMO | −8.7% | 41 | NAND controllers — levered to NAND volumes |
| RMBS | −8.3% | 29 | Memory IP / royalties |
| EWY | −8.3% | 27 | Korea — SK Hynix is the index |
| WDC | −7.1% | 33 | HDD — benefits from memory-tightness spillover |
| STX | −7.0% | 33 | HDD — same |
| SKHY | −6.6% | — | The name itself (cushioned by ADR arbitrage) |
| MU | −5.3% | 32 | DRAM/HBM — ~40% of revenue under price floors |
| SOXX / SMH | −5.0% / −4.2% | 35 / 36 | Semis broadly |
| NVDA | −3.5% | 45 | Compute, not memory |
| QQQ | −1.8% | 41 | — |
| SPY | −0.7% | 54 | — |
This is not sympathy. Sympathy is flat — it knocks a sector down roughly evenly. This is a clean, monotonic gradient running from the broad market (−0.7%) through semis (−4 to −5%) into memory (−5 to −12%), and inside memory it sorts almost perfectly by spot-price exposure.
The tell is that Micron fell least of the memory names. Micron is the one name in the complex that has contractually removed a slice of its revenue from the spot market — its take-or-pay agreements carry price floors on roughly 40% of company revenue through 2030. The market discounted it least. The name with no such protection and nothing but NAND — SanDisk — it discounted most, by a factor of more than two.
A market pricing sympathy does not sort names by contract structure. A market pricing ASP risk does exactly that.
We also checked SanDisk for a company-specific catalyst. There is none: no earnings, no guidance, no downgrade. The move is not about SanDisk.
What the market was actually reacting to
Two things landed together, and both point at price rather than demand.
1. The downgrade names ASP explicitly. The July 13 Korea Investment Securities note cut SK Hynix 2026–27 operating profit 9–11% citing weaker HBM sales mix, HBM4 ramp delays, and — the load-bearing phrase — slower ASP growth. ASP is not an HBM variable. It is the whole-complex variable. That is why HDD names with no HBM exposure fell 7%.
2. The $26.5B raise is a supply signal, and the market read it as one. SK Hynix's CEO said, on the day of the listing, that the shortage would persist beyond 2030 and that 2027 would be the worst supply year in the industry's history. In the same week, the company raised a record $26.5 billion — and told regulators, in the prospectus, that the money is going into new memory fabs.
The most authoritative voice on the durability of the shortage has just raised the largest sum ever raised on a US exchange by a foreign company in order to end it. That is not hypocrisy; it is the mechanism. Scarcity raises the price, the price funds the capacity, and the capacity ends the scarcity. It is how every previous memory shortage has ended.
Why this matters more than today's drawdown
Our memory-supercycle thesis says: HBM cannibalizes DRAM capacity → all-memory tightness → pricing power. Today puts a hard question to the last two words.
Look at where Micron's record actually came from. In its fiscal Q3 (reported 2026-06-24): revenue $41.5B, +346% YoY, gross margin 84.9%, operating margin 81.2%. Extraordinary. But underneath: DRAM revenue rose 67% sequentially while DRAM bit shipments rose low-single-digits — prices rose in the low-60s. NAND revenue nearly doubled; bits rose mid-single-digits; prices rose in the mid-80s.
Micron did not sell meaningfully more memory. It sold almost exactly as much memory, for very much more money. Three years ago the same factories, the same patents and the same engineers produced a gross margin of −9.1%. Ninety-four points of margin swung on nothing but price.
That is the signature of a commodity in shortage, not of a moat. And it is why the complex sorts by ASP exposure when the ASP story is questioned.
The sharpest evidence is one nobody seems to have read closely. Micron's own prepared remarks say its largest agreements carry not only a floor but a ceiling price at the current quarter's market price, and that when the planned deals are done, agreements with fixed-or-capped prices at roughly today's levels will cover about 40% of company revenue through 2030.
Read that again. At the peak of the most extreme pricing environment in the industry's history — with a competitor's CEO publicly saying the shortage outlasts the decade — Micron has agreed to cap what it can charge its biggest customers, at today's price, for five years.
You do not sell five years of upside cheaply in a market you believe will stay this good. You sell it because you want the floor, and the floor has to be paid for. Micron bought insurance, and the premium tells you what the company privately thinks the risk is. Nvidia has not done this. TSMC has not done this. ASML has not done this. Companies with moats do not buy stability — they have it.
Verdict
The selloff is an ASP repricing. SanDisk fell hardest because it is the highest-beta expression of memory price, not because of anything about SanDisk. Neither of the two hypotheses we opened with survives intact:
- Sympathy — rejected. Sympathy does not sort a sector by contract structure.
- Liquidity drain into SKHY — mostly rejected. If capital were rotating into the new listing, SKHY would have held up while the rest fell. It didn't: SKHY (−6.6%) landed in line with the complex median (≈−7.1%). The drain story needs SKHY outperforming, and it isn't.
What replaces them is more consequential than either. The thesis is not wrong about the present — it is exposed on durability. The tightness is real; the 82% operating margin is real. The open question is whether it is a moat or a shortage, and this week produced two pieces of evidence that it is a shortage: a record capital raise pointed directly at new supply, and the industry's best-informed operator selling a five-year price ceiling to buy a floor.
This does not break the trade. It dates it. A commodity shortage is investable — but it is investable on a clock, and today the market started looking for that clock.
What would settle it
- Micron's next guide — the only thing that speaks directly to HBM4 timing, mix, and whether ASPs are decelerating or just pausing.
- Does the SNDK-vs-MU spread persist? It opened at 8 points on day one. If it holds or widens, the market is pricing an ASP roll and the contract structure is the whole ballgame. If it closes, today was a one-day shock and the sort was noise.
- Capacity announcements. SK Hynix's $26.5B is the first. Watch whether Samsung and Micron follow — the shortage ends when the industry as a whole spends, not when one player does.
- Bits vs price in the next quarter. If bit shipments finally accelerate while prices hold, the demand story is real. If price does all the work again, the clock is running.
Prices and RSI from an internal tape refresh (2026-07-13 intraday snapshot vs the 2026-07-10 close; market open at the time of writing, so these are not settled numbers). SanDisk checked for company-specific news via the desk news pass — none found. Micron fiscal Q3 figures (revenue $41.5B, gross margin 84.9%, operating margin 81.2%, DRAM/NAND bit-vs-price split), the take-or-pay floor/ceiling structure and the ~40%-of-revenue coverage from Micron's own earnings materials and prepared remarks, reported 2026-06-24. SK Hynix listing terms ($149 pricing, $26.5B raised, 7x oversubscribed, +13% debut) and the prospectus statement that proceeds fund new memory fabs, consistent with our 2026-07-06 SKHY listing investigation and its SEC F-1 sourcing. The SK Hynix CEO remarks on shortage duration and the reading of the raise as a supply signal from "Micron Stock and the $26.5 Billion Tell" (Wonderful at a Fair Price, 2026-07-12), surfaced by the curated feed sweep. The forecast cut reported by Semiconductor Insider (@SemiconductorsX) citing a Korea Investment Securities note dated 2026-07-13.