# Memory’s Second Derivative: Why Chip Stocks Fall While Profits Rise *URL:* https://millionaur.com/millionaur-2026-10-20-daily-analysis/ *Updated:* 2026-10-20T08:35:42+00:00 On Monday the US chip index rose 1.6% and ended its eight-day losing streak. Twelve hours later Korea’s KOSPI fell 3.7%, and SK hynix headed for its lowest close since April. Two markets read the same week in opposite directions. When that happens, the useful question is not which one is right. It is what each one is pricing. ## The edge cracked in a new place Last week the stress in the AI build sat in the bond market: CoreWeave, which borrows to rent out AI chips, saw some of its bonds yield 15 to 16%. This week it showed up in the share market. Elice, a Korean company that rents out GPUs from its own data centres, listed on Tuesday. It opened 16% below its IPO price and fell as much as 33%. That is still the edge of the AI build: companies that borrow or raise money to rent out chips. The core is the builders paid in cash, and nothing there changed overnight. The alarm I watch for is the stress moving into the bonds of the large cloud companies, or into the long-term memory contracts. Neither has happened. ## The memory number that matters The research firm IDC says smartphone shipments fell 22% in the third quarter, the worst quarter it has ever recorded. Memory costs are up about 350% in a year, and phone makers are building fewer phones. The detail most people skipped: the quarter-on-quarter increase in memory costs slowed to low double digits, from roughly 50 to 100% in earlier quarters. This splits the memory debate in two: - Memory makers’ revenue depends on the price **level**, and most of 2027 server and AI memory is already under contract. - Memory shares trade on the **rate of change**. In past cycles the stocks peaked when price increases slowed, even while profits were still rising. So the selloff may be the market pricing the slowdown in price increases, not a collapse in demand. The test has dates: SK hynix reports on 27 October and Samsung on 29 October. If they still see solid quarterly contract increases and a sold-out 2027, the level story holds. If quarterly prices go flat, the shares were early, not wrong. One more wrinkle: a union at Micron Technology (MU) in Taoyuan, Taiwan, says it will strike on Thursday. That site makes DRAM and high-bandwidth memory. A short strike costs Micron output and helps Samsung and SK hynix. ## What changed in the portfolio - Added the first US Treasury position (7 to 10 year bonds, about 0.5% of the portfolio) at Monday’s open. The pre-written condition was met on Friday. Yields rose afterwards, so it starts slightly down. That is fine: it is a hedge for the next credit scare, not a trade. - Reset Constellation Energy (CEG) and a solar ETF to standard position size. - Raised the profit floors on two energy ETFs. - No new AI purchase yet. The one AI addition I allow per week while the 10-year yield is above 5% is being kept for after Wednesday’s results. ## What I watch this week - Wednesday: Meta Platforms (META) on capital spending and free cash flow, and Lam Research (LRCX) on chip-equipment demand for 2027. - Thursday: the Micron strike and Western Digital (WDC) on storage pricing. - Friday: Oracle (ORCL) exits on a weekly close below about $130. It trades near $128 now. - Next AI addition: Taiwan Semiconductor (TSM) on a weekly close above about $430 or near $400, or a second ASML Holding (ASML) position after Lam confirms demand. - No memory additions before 27 and 29 October. ## Risks, stated openly - If quarterly memory prices go flat, memory shares can keep falling even with record profits. Memory and Korea are about 10% of the portfolio. - If the 10-year yield closes a week above 5.20%, the bond position comes out. - Oil is near $110 with Hormuz still restricted. Energy is a hedge and will fall fast on a deal. - Monday’s US bounce looked like short covering by funds that had sold tech hard. It can fade. ## The lesson Prices are a story about direction as much as level. A business can earn more every quarter while its shares fall, because the market is already pricing the next derivative. The answer is not to guess. It is to write the test down, date it, and keep enough cash and hedges to act on the result. *Portfolio snapshot: 20 October 2026, 08:20 UTC. Market data as of the 19 October US close and Tuesday’s Asian session.* Sources: [IDC Q3 2026 smartphone shipments (Fonearena)](https://www.fonearena.com/blog/495163/global-smartphone-shipments-q3-2026-idc.html/amp) · [Asia stocks slump as AI jitters hammer Korean chipmakers (Investing.com)](https://www.investing.com/news/stock-market-news/asia-stocks-slump-as-ai-jitters-hammer-korean-chipmakers-ahead-of-tech-earnings-4958731) · [Micron union in Taiwan to launch strike (AsiaOne)](https://www.asiaone.com/money/micron-union-taiwan-launch-strike-bonus-dispute-escalates) **This is not financial advice.** Past performance does not guarantee future results, and copy trading carries a real risk of losing money. Always do your own research. --- Human page: https://millionaur.com/millionaur-2026-10-20-daily-analysis/?vwao_human=1