Two green DRAM memory modules with gold contact pins, close up
Investing

The memory unwind

Memory Stocks Are Crashing — and the Stock That Caused It Is Up 466% (2026)

RaymondRates.my5 min read

📊 Stocks in this story · tap for live chart & discussion

Memory was the best trade of 2026. This week it became a bear market. SanDisk fell 14.1% on 28 July, Micron 10.3%, and SK hynix now trades below the IPO price it set 17 days ago. The trigger was CXMT's Shanghai listing — which is up 466% and is the only name in this story that hasn't fallen.

−14.1%
SanDisk, 28 July
second straight heavy session
−10.3%
Micron, 28 July
whole complex sold together
+466%
CXMT, the cause
China's most valuable listed company

What actually happened

  • Two sessions, not one. On 27 July — CXMT's debut day — SanDisk fell 12%, Micron 5% and SK hynix 8%. The next day it got worse, not better: SanDisk closed −14.1%, Micron −10.3%, with SanDisk down more than 18% at its intraday low.
  • It's a sector move, not a company story. NAND and DRAM names fell together — Micron, SanDisk, Western Digital, SK hynix, Samsung. No earnings miss, no guidance cut. That's a repricing of the whole industry's future margins.
  • Korea took the worst of it. The KOSPI fell about 11% in a single session and roughly 29% over the month — memory is that large a share of Korea's market.
  • Everything is now in a bear market. Micron, Samsung, SK hynix and the Roundhill Memory ETF are all more than 20% below their recent closing highs.
THE CAUSECXMTShanghai debut, 27 Jul+466%THE DAMAGEown scale, 0 to −15%SanDisk28 Jul close14.1%KOSPIone session11%Micron28 Jul close10.3%SK hynix27 Jul close7.5%The two panels use different scales on purpose — +466% and −14% cannot share an axis honestly.
One chart, the whole story: the listing that triggered the selloff is the only thing that went up.

Why CXMT broke the trade

  • It funded the competitor. CXMT raised roughly US$8.6–9.8 billion — mainland China's largest-ever semiconductor offering — earmarked for mass-producing memory wafers. That is new supply, financed.
  • The 466% pop was the signal, not the raise. The market read the debut as proof that China can fund a full domestic memory build-out. Cheap future supply is exactly what compresses the fat 2026 margins the incumbents were valued on.
  • Supply was already loosening. HBM capacity has been expanded aggressively and NAND/DRAM output keeps rising. The shortage that drove this year's pricing power was easing before CXMT listed — the IPO just put a date on it.
  • Washington noticed. The surge reportedly triggered a Capitol Hill probe, and SanDisk is caught up in a China NAND inquiry — regulatory risk stacked on top of pricing risk.
The same event was the biggest IPO win of 2026 and the reason the incumbents repriced. One market's windfall was the other market's margin forecast.
The uncomfortable symmetry

"China makes it cheaper" — actually, it doesn't

The instinctive explanation is that CXMT undercuts on cost. The numbers say the opposite, and the real mechanism is more interesting.

  • CXMT's chips cost more to make, not less. Its cost per bit on DDR5 is reportedly more than 30% higher than the three leading suppliers. It runs at roughly a 16nm node with a large die and yields that aren't fully proven.
  • Price in a commodity is set at the margin. Memory is a commodity — the clearing price is set by whoever is willing to sell last, not by the best producer. A state-backed manufacturer pursuing self-sufficiency doesn't need to earn a return on capital, so it can sell below economic cost and still be doing its job.
  • So the threat is volume, not cost. CXMT is projecting roughly 350,000 twelve-inch wafer starts a month by end-2026 — approaching Micron's estimated 375,000–385,000. That is the number that matters: a Micron-scale supplier who doesn't need Micron's margins.
  • It also displaces imports. Even output that never leaves China frees up supply everywhere else, because it replaces memory China would otherwise have bought from the incumbents.
The part the selloff may be over-reading. CXMT has no HBM — its IPO prospectus contains no HBM project at all, and Samsung, SK hynix and Micron hold over 99%of global HBM supply. HBM3 at competitive yield is a 2028-or-later prospect for CXMT. HBM is precisely where 2026's fat AI margins come from. So the market has repriced the incumbents for a commodity DRAM threat while the profit engine they actually depend on is, for now, untouched. That gap is the strongest argument that this week is an overshoot.
And note who fell hardest. SanDisk makes NAND flash. CXMT makes DRAM. They are different products — yet SanDisk fell the most. That tells you the market sold "memory" as a single bloc rather than pricing each business, helped along by SanDisk having run up the furthest and by a separate China NAND inquiry. Bloc selling is usually where mispricing hides.

SK hynix: 17 days from record to underwater

The cleanest measure of how fast sentiment turned is the newest listing in the group.

$140$150$160$170IPO price $149$168.01$143.02 · below IPO10 JulIPO price10 JulDay-1 close17 JulBreaks IPO27 JulNow17 days from record listing to underwater
SK hynix's US ADR: priced at $149 on 10 July in the largest foreign listing in US history, closed day one at $168.01, and was back under the IPO price within a fortnight.
A hand holding a phone showing a falling price chart in a dark room
No earnings miss, no guidance cut — the whole complex simply repriced at once. · Photo: Unsplash

Refreshing our call — including the one we got wrong

  • 26 June — we called the top. Our first top call flagged the AI trade cracking while the mega-caps still masked the damage. That has held up: a month later the memory complex is in a bear market.
  • 17 July — we called the shakeout and rebound. The shakeout piece expected a washout into end-July then a bounce. The bounce did arrive on 20 July — SanDisk +11.6%, Micron +6%.
  • 26 July — this is the one we got wrong. In the dead-market piecewe wrote that the AI lows were "likely already in." Two days later memory broke to new lows. We were writing about CXMT that same week and did not connect it to the memory complex. That was the miss.
  • What changed the picture. A rebound needs the thing that caused the fall to stop getting worse. Liquidity draining into an IPO is temporary. A funded competitor changing the industry's supply curve is not.
Our read now: lower before better. We think the more likely path from here is another 10–20% down in the memory names before a durable rebound, rather than a V off this week's low. The reasoning: the selling is repricing long-run margins rather than reacting to one headline, positioning was crowded after a year of gains, and the catalyst is structural. This is our scenario, not a forecast or a recommendation — see the levels below for where that lands, and the "what would break this" note after it.

What to watch

  • 30 JulMemory earnings land. Guidance on 2027 pricing matters far more than the quarter itself — that is the number that either confirms or kills the margin fear.
  • DaysWhether CXMT holds its debut gain. If the cause cracks, the incumbents get relief; if it holds, the supply story stays credible.
  • WeeksContract DRAM and NAND pricing. Spot noise is not the signal — contract prices rolling over would confirm the cycle has turned.
  • OngoingThe Capitol Hill probe and the China NAND inquiry. Policy can cut either way here: restrictions on CXMT would be a reprieve for incumbents.
What would break this view. A strong 2027 pricing outlook at the 30 July earnings; CXMT giving back its debut gain; US restrictions that slow Chinese memory expansion; or a fast capex response from the incumbents. Any of those would make this week a bottom rather than a waypoint — and we would say so.
Not financial advice. This is general market commentary using publicly reported figures as at 28 July 2026. Rates.my is not a licensed investment adviser, does not issue price targets, and none of the above is a recommendation to buy or sell any security. The 10–20% figure is our scenario for how a repricing typically plays out, not a prediction — markets routinely do neither what we nor anyone else expects. Verify live prices and do your own research.

Sources & further reading

Keep reading

💬 Market Chat

Join the discussion — share your take on this with other Rates.my readers.

💬 Discussion (1)

Live · last activity 7h ago

Your rating:
(optional)
F
Fang Yuan·7h ago
Time to buy the dip!

This article is general information, not personalised financial advice. Rates.my is not a licensed financial adviser — always verify rates with the institution and consider your own circumstances.

← All insights