Memory Stocks Just Had Their Worst Week Since April 2025 — Seven Forces Behind the Selloff
The memory trade finally blinked. Micron and SanDisk each fell roughly 10.6% on Wednesday, July 1, with Western Digital and Seagate dropping 6.3% and 5.2%. Thursday brought a second leg down: SanDisk lost another 11%, Seagate 7%, Micron 4%. The Roundhill Memory ETF (DRAM) — the cleanest sector proxy, launched only in April — shed nearly 11% Wednesday and another 5% Thursday. The Philadelphia Semiconductor Index posted a 7.9% weekly decline, its worst since April 2025.
For a group that has been the defining trade of 2026 — Micron up roughly 300% year to date, SanDisk up over 750%, both Micron and SK Hynix crossing the trillion-dollar market cap threshold in May — the question is whether this is routine profit-taking after a parabolic run, or the first crack in the memory supercycle thesis. The honest answer is that seven distinct forces hit at once, and they don’t all point the same direction.
The Setup: A Trade That Left No Room for Error
Context matters. Micron closed June 30 up 303% year to date. SanDisk entered the week up more than 750% for the year and over 4,000% over twelve months. These are not normal equity returns; they are the market pricing in a structural shortage with multi-year duration. Micron’s fiscal Q3, reported June 24, showed why: $41.5 billion in revenue, up 346% year over year, with non-GAAP EPS of $25.11 crushing the $20.28 consensus, and Q4 guidance calling for $50 billion in revenue and $31 in EPS at the midpoint. Gross margins around 85% — for a commodity memory maker — are the kind of numbers that only exist when supply discipline meets a demand shock.
The problem with pricing in perfection is that any input to the model becomes a catalyst. This week, several inputs moved simultaneously.
Force One: Mechanical Selling at the Half-Year Turn
The most boring explanation is also the most likely primary driver: profit-taking and institutional rebalancing at the start of the second half. Funds sitting on 300–800% gains in concentrated positions had every incentive to trim on July 1, the first trading day of H2. There was no single fundamental trigger on Wednesday — the selling started before the negative commentary circulated. When a basket is this crowded and this extended, the exit door is narrow. The DRAM ETF’s concentration — Samsung, SK Hynix, and Micron together make up roughly three-quarters of the fund — means any de-risking moves the entire complex in lockstep.
Force Two: The Morningstar Valuation Warning
Thursday’s leg down had a clearer catalyst. Morningstar’s director of research told Bloomberg TV that a large slice of AI names could give back 20–30% before becoming buyable again, and specifically flagged the biggest-gaining memory names as most exposed to a valuation reset. Her argument centers on capacity: announced supply additions from Samsung and SK Hynix should soften memory pricing as supply catches up, while AI capital expenditure is expected to peak in 2026 and taper afterward. That combination attacks exactly the pricing power that produced this cycle’s extraordinary margin expansion.
Force Three: The Supply Question Gets Real
The Morningstar call landed because it echoed a genuine development: new manufacturing capacity announced in South Korea last week revived the oldest fear in memory investing — that the industry’s legendary cyclicality will reassert itself, swinging from tight supply to oversupply as it has in every prior cycle. The bulls’ counterargument is that this cycle is different in structure: Micron has locked in 16 long-term contracts with minimum pricing provisions expected to cover roughly 40% of revenue, and SanDisk signed three contracts in its fiscal Q3 worth $42 billion in total contractual revenue. UBS reads Micron’s contract strategy as implying sustainable gross margins of 70–75% — below the current ~85% but far above the prior cyclical peak of 62% in 2018. Contracts dampen the cycle; they don’t repeal it.
Force Four: The SRAM Wildcard
The most interesting — and most underpriced — development of the week came from the inference side. OpenAI’s deal with Cerebras, signed in January, is reportedly delivering sharply reduced inference costs. Cerebras’ wafer-scale chips run on on-chip SRAM rather than the HBM DRAM that SK Hynix, Samsung, and Micron sell, and its CEO has been explicit about the read-through: HBM is the binding constraint in the market, it’s expensive, and Cerebras simply doesn’t use it. Nvidia’s acquisition of Groq — another SRAM-centric inference architecture — points the same direction.
The bear read is that if hyperscalers shift meaningful inference capacity to architectures that bypass HBM entirely, the HBM-content-per-token assumption underlying memory valuations could prove too high. The more nuanced read is that SRAM-centric systems have hard physical limits: even a full Cerebras wafer holds only 44 GB of SRAM, clustering enough wafers to hold a trillion-parameter model can cost $100 million versus a few million for an Nvidia rack, and exploding KV caches from long-context agentic workloads still demand oceans of HBM, DRAM, and SSD behind the fast tier. The likelier outcome is a deeper, more layered memory hierarchy — SRAM plus HBM plus DRAM plus flash — which arguably expands the memory TAM rather than cannibalizing it. But markets don’t do nuance during a de-risking week; they sell first.
Force Five: The Buyers Are Bleeding
A separate margin story is developing downstream. Surging memory prices — the very thing driving Micron’s 85% gross margins — are squeezing the companies that have to buy the chips. Apple’s name keeps surfacing in this conversation, and reports of Rivian struggling to source memory chips and Apple raising device prices suggest the pain is spreading. If major buyers respond by redesigning products, delaying refreshes, or renegotiating hard, the demand side of the supercycle equation softens. Memory makers’ margin expansion and their customers’ margin compression are the same phenomenon viewed from opposite ends; it is not stable indefinitely.
Force Six: The DRAM Cartel Lawsuit
Adding legal risk to the mix, Micron, Samsung, and SK Hynix were named in a US class-action lawsuit alleging deliberate restriction of conventional DRAM production to inflate prices. Memory investors with long memories will recall the industry has been here before — DRAM price-fixing litigation is practically a genre. The near-term earnings impact is negligible, but the suit hands regulators and plaintiffs a narrative at precisely the moment the industry’s “production discipline” is generating record margins. The key risk isn’t the damages; it’s any settlement or finding that forces production behavior changes and breaks the pricing structure.
Force Seven: SK Hynix Comes to Nasdaq
Finally, a flow story: SK Hynix is preparing a Nasdaq listing with a $29 billion stock sale. That does two things. It gives US investors who want memory exposure a large, liquid alternative to Micron and SanDisk — competing for the same pool of AI-memory capital — and it funds SK Hynix’s capacity expansion, feeding directly back into the supply concerns above. Some traders openly speculated whether this week’s selloff was partly positioning ahead of the offering’s pricing. SK Hynix shares also sold off in Seoul, and the weakness trickled into US names.
What Holds and What Doesn’t
The sell-side hasn’t capitulated. Bank of America raised its SanDisk target to $2,500 from $2,100 on Wednesday — into the teeth of the selloff — arguing the NAND supply/demand imbalance persists through 2027 with pricing staying firm into mid-2027. KeyBanc maintained Overweight on Micron at $1,600, citing constrained supply, production discipline, and data center demand for HBM and DDR5. Micron trades at roughly 7.5x forward earnings even after the run; SanDisk at about 32x. June contract prices for DRAM and NAND were still rising.
The frame worth holding: nothing in this week’s action refutes the demand story through 2027. What the week did was surface the three ways the trade eventually ends — supply response (Samsung/SK Hynix capacity plus the Korean expansion), demand-side substitution (SRAM architectures and buyer pushback), and the cycle’s oldest enemy, its own margins attracting capacity. All three moved from theoretical to visible this week. A stock up 300% can fall 30% without breaking a single technical pattern, and Micron is already down over 16% from its June 25 peak. Position sizing, not thesis conviction, is what gets tested at these prices — and this week tested it.