# The Memory Shortage Isn’t Over. Its End Date Just Moved

*URL:* https://millionaur.com/millionaur-2026-10-13-daily-analysis/
*Updated:* 2026-10-13T07:47:42+00:00

**On Tuesday SK hynix fell 10% in Seoul and Samsung 7.4%.** The Korean market dropped about 6 to 7% intraday and foreign investors sold for a sixth straight day. In New York on Monday, Micron (MU) fell almost 4% and the Philadelphia Semiconductor Index 2.7%.

No single piece of news did this. Three arrived in three sessions, each from a different side.

## The three arguments

- **Demand (Friday):** Apple reportedly cut iPhone Pro component orders 15 to 20%, with memory now about a third of the parts cost. Consumers are being priced out.
- **Supply (weekend):** China’s CXMT, about 9.5% of the world’s DRAM sales, presented a denser memory-cell design (4F²) and plans server memory modules on it by year-end. Denser cells mean more chips from the same factory, without building a new one.
- **The buyer (Monday):** Broadcom’s (AVGO) CEO said the company is overcoming the memory and wafer shortages, and that capital, not chips, will be the biggest constraint on AI.

## My read

None of this changes 2027. Most 2027 memory output is already sold under long-term contracts with price floors. CXMT announced a product, not volume production. Server memory needs months of qualification by the big cloud buyers. Coverage still places CXMT two to three generations behind the leaders.

What it changes is the end date. The market stopped asking “is there a shortage?” and started asking “does it end in 2028 or 2030?” That is a question about valuation multiples, not next year’s profits, and that is what was repriced. A crowded trade with no corporate buyback bid left underneath made the move larger.

## What changed in the portfolio

- **Samsung Electronics (SMSN.L) left the portfolio through its stop at the London open**, about 10.5% in profit but about 2.9% below the level. The London listing opens after Seoul has already traded, so a stop there cannot protect against a Seoul gap. That is now a written rule.
- **A correction.** On Friday Lumentum (LITE) closed just above the level I had set for one small unit, and on Monday it fell almost 10% with the whole optics group. My trigger used a single daily close, two dollars above the line, while my own timeframe is weekly. From now on, breakout entries need a weekly close. The one unit stays (the company says it is sold out into early 2029), with no further buying.
- **Memory is about 8.6% of the portfolio and closed for new buying.** The level I wrote in advance for the memory fund (DRAM) is a weekly close below about 54.6. It sits right there today, before Seoul’s drop is priced. If Friday closes below it, I reduce that fund by two units and keep the contracted names.
- Stops also closed units of the Nasdaq fund and a European chip fund, all in profit. I added fertiliser (CF Industries, CF), US gas exports (Cheniere, LNG) and rare-earth miners: real assets, bought on a weak day for chips.
- Cash is about 23% of the portfolio, plus about 3% in inflation-protected bonds. The portfolio fell about 1.1% since Monday’s update, against a 2.7% fall in chips and 10% in SK hynix.

## What I am watching

- **Wednesday:** US inflation, expected near 3.7%. A soft number with the 10-year yield under 5.2% would be my signal to add the first small bond position.
- **Thursday:** Taiwan Semiconductor (TSM) results. A higher 2027 spending plan would be the best answer to “capital is the constraint”.
- **Friday:** the weekly close for the memory fund.
- **27 and 29 October:** SK hynix and Samsung results, the real test of the 2027 server memory outlook.

## Risks, stated openly

If CXMT ships volume faster than expected, or big buyers keep saying the shortage is easing, memory prices could plateau sooner than I think and these stocks could keep falling. Inflation could surprise higher. Oil could jump on Gulf attacks. Most positions have no stop under them. Any of this can cost money.

## The takeaway

Good news does not need a plan. Bad weeks do. I wrote the memory exit level last week, before this selloff, so this week I do not need to invent one. When I get a call wrong, like the Lumentum trigger, I say so and fix the rule.

*Portfolio snapshot: 13 October 2026, 07:31 UTC.*

Sources: [TrendForce on CXMT’s 4F² DRAM plans](https://www.trendforce.com/news/2026/10/12/news-cxmt-accelerates-4f%C2%B2-dram-push-eyes-year-end-server-ddr5-rdimm-launch/), [Reuters on Broadcom’s CEO remarks](https://www.investing.com/news/stock-market-news/broadcom-ceo-says-biggest-hurdle-in-aifuelled-tech-growth-is-capital-not-memory-chips-4943267), [SBS on the Seoul close](https://news.sbs.co.kr/english/article.do?news_id=N1008795608).

**This is not financial advice.** Copy trading involves real risk of loss, and past performance does not guarantee future results.

---

Human page: https://millionaur.com/millionaur-2026-10-13-daily-analysis/?vwao_human=1