To The Moon

CXMT Targets Server Memory: Is the Memory Repricing Done? Here's What the Numbers Say, and Why I'm Not Selling Puts Yet

Written Friday 16 Oct, SGT, after Thursday's US session.

Last weekend in Shanghai, CXMT president Cao Kanyu said China's largest DRAM maker plans to launch a DDR5 RDIMM server module built on its new 4F² cell design by the end of 2026. A 4F² cell takes up roughly a third less area than the 6F² design the industry still uses, so you get more memory per chip. CXMT says it holds more than 300 patent families on the architecture.

That's the headline. Here's what wasn't said: no speed, no capacity, no yield and no volume. It's a product launch target, not a commitment to mass production or to qualification at the big server platforms.

The market didn't wait for the details. From the Oct 2 close, six of the largest memory and storage makers lost about $1.09 trillion in combined value. SK Hynix dropped 31% in seven sessions, its worst seven-day run since 2008, and its Seoul shares now sit nearly 58% below their record. Shortly after Thursday's US open, Micron was 30.3% off its record, SanDisk 44.3% and Western Digital 57.2%. One 2x SK Hynix product lost 22.69% in a single session.

CXMT wasn't the only reason. Two hot inflation prints landed back to back. September CPI came in at 3.7% against 3.6% expected, and PPI jumped 1.3% against a 0.5% forecast. That revived fears of an October rate hike, and higher rates make the data centers that buy these chips more expensive to build.

Thursday also showed how stretched positioning was. SanDisk was down 6.6% near the open and closed down only 0.61%. Micron reversed hard off its lows too. When a sector can swing that much inside one session, the selling is about positioning as much as fundamentals.

The bear case: why this isn't just noise

CXMT is already big. TrendForce puts its DRAM revenue share at 9.5% in Q2, up from 7.6% a quarter earlier, behind only Samsung, SK Hynix and Micron. It raised $8.6 billion in its Shanghai IPO in July to fund more capacity.

Server memory is where the money is. Micron's gross margin reached 86.76% in the quarter ended September. That margin exists because server memory is short. A credible new supplier is a direct threat to pricing.

SK Hynix's HBM lead is already shrinking. Counterpoint data shows its HBM share fell from 58% to 50% in Q2, while Samsung jumped from 21% to 33%.

The oversupply story has a date now. A Maeil Business column warned that Korean memory could face oversupply from 2029 as new fabs come online. Elon Musk's reply that China will make "a LOT of RAM in 2028" didn't help.

This is not HBM. CXMT's planned module is conventional DDR5. The highest-value part of the market, high-bandwidth memory for AI accelerators, isn't what was announced.

Launch is not volume. Server memory needs continuous-load stability, error control and long validation cycles with each platform. Even a Korean analyst quoted in the press questioned whether CXMT can produce enough to matter, even if the product shows up on time.

Demand hasn't cracked. Micron says its HBM capacity is sold out through 2027 and expects HBM demand to outpace the broader DRAM market through 2028. TSMC reported Q3 revenue of $46.72 billion on Thursday and raised its 2026 growth outlook to about 44%.

Analysts split on who gets hurt. D.A. Davidson's Gil Luria told CNBC any impact lands on SK Hynix and Samsung, not Micron. Mizuho's Jordan Klein said it changes little in the near term. Rosenblatt called the selloff overblown.

We have seen this movie twice already. Micron fell 8.02% on July 15 after CXMT disclosed its IPO terms and 3.5% on August 3 on another China supply report. Both times the stock recovered.

So, is the repricing done?

The CXMT news itself is now priced. The bigger question underneath it, how long this memory upcycle lasts, is still open, and it won't be settled by price action alone. These are the four things I'm watching:

SK Hynix Q3 results, Oct 28 Seoul time. The first big read on HBM pricing and long-term contract coverage after the selloff.

SanDisk results on Oct 29. The NAND side of the story, and a test of whether the flash names were sold on CXMT headlines that are really about DRAM.

CXMT's year-end deadline. If a qualified server module with real specs shows up by Dec 31, the threat gets a timeline. If it slips, the market will remember.

Rates. Another hot inflation print matters more for these valuations right now than anything CXMT says.

To answer the topic directly: SK Hynix carries the most direct risk because of its HBM concentration and shrinking share. Micron has the cleanest balance of HBM exposure, margin and capital return, including a $35.16 billion buyback authorization from Dec 9. SanDisk and WDC are mostly collateral damage from a DRAM headline, but they still need to prove it on earnings.

How I run my sell-put strategy on this

Memory names aren't on my core watchlist, but this is exactly the type of setup where my rules earn their keep. Here is how I'd work through it.

1. The earnings calendar comes first, before the chart or the story.

SK Hynix reports on Oct 28 Seoul time. SanDisk reports Oct 29. My rule is no earnings inside the expiry and no new position within 7 calendar days of a report. Any 30 to 45 day put I open today expires in mid-to-late November, which straddles both reports. So both are blocked. That's not a view on the companies. It's the rule.

2. Micron is the only name the calendar lets through.

Micron's next report is estimated for mid-December but hasn't been confirmed by the company. A November expiry would clear both the report and the blackout. I'll only treat it as eligible once the date is confirmed.

3. Check the size before anything else.

Micron trades around $900. A cash-secured put at a hypothetical $800 strike ties up $80,000 per contract. Non-core names are capped at 2% of my account. For most people that one contract is too big, and that alone ends the conversation.

4. Fear raises the premium. It doesn't make the trade.

A selloff like this pushes implied volatility up, which is good for put sellers. But a high VIX or a scary headline isn't a signal by itself. The specific chain still has to pass my volatility gates:

IV Rank of at least 40%

IV Percentile of at least 45%

Implied volatility at least 1.2 times historical volatility

5. Structure decides the strike, not valuation.

The stock has to be above its 200-day moving average. For the strike, I take the lower of two anchors: the one standard deviation floor or the key moving average × 0.98, and the delta has to land between 0.15 and 0.25. A low forward P/E on a memory stock isn't a margin of safety. Cyclical stocks look cheapest when earnings peak.

6. Leveraged single-stock products aren't underlyings.

That 22.69% one-day loss is what daily-reset leverage does in a volatile sector. Those products decay in choppy markets, and I don't write puts on them.

7. If I do open one, the exits are set on day one.

A 50% take-profit GTC order on fill, a stop at 2.5 times the premium for a non-core name, and one roll at most.

My plan

Let SK Hynix and SanDisk report, watch CXMT's December deadline, and keep an eye on rates. If memory sells off again into earnings and Micron's chain passes every gate with a confirmed date, that's the fear I want to sell into. Until then, the best trade here is patience.

Is the memory repricing done for you, or are you waiting for the earnings week to decide? Which name would you hold through it: Micron, SK Hynix or SanDisk?

This is my personal trading framework, shared for education. It is not investment advice. Options carry significant risk, including assignment and losses larger than the premium collected. Always verify live chain data and confirmed earnings dates before trading.

$MU $SKHY $SNDK $WDC