Rate-Hike Bets Collapsed and Long Yields Rose Anyway. Who Is Setting the Long End?
The indices: a weak jobs report, and all three still closed higher
All three indices closed higher on Friday: the $NASDAQ(.IXIC)$ Composite rose 1.19 per cent to 27,190.86, having reached 27,353.68 at its intraday peak; the $S&P 500(.SPX)$ rose 0.73 per cent to 7,722.72; and the $Dow Jones(.DJI)$ rose 0.49 per cent to 51,176.96.
The September employment report came out that morning: non-farm payrolls added 29,000, where the market had looked for somewhere between 85,000 and 90,000, and the unemployment rate went from 4.1 per cent in August to 4.2 per cent, where a flat reading had been expected. The previous two months were revised down as well: July and August lost a combined 60,000, with July turned from a gain of 21,000 into a net loss of 10,000. The report itself was weak and equities rose anyway, because of what it did to rates: the case for another rate rise has been pushed down. Two paths follow. Should October's data stay soft, the market moves from "no rise" to counting cuts, and Friday was only the opening move. Should October bounce back to seventy or eighty thousand, September gets treated as statistical noise and Friday's reasoning expires on the spot. The thing to watch is whether the next payrolls report puts these revisions back.
Rates: the bets on a rise collapsed and long yields went up regardless
By Friday's close, futures markets put the odds of an October rate rise at around one in five, against a range of three in ten to seven in ten a week earlier depending on whose reading you take. That combination ought to have let both equities and bonds breathe out, and the bond market did nothing of the sort: the 10-year Treasury yield rose another 4 basis points to close at 5.28 per cent, having traded near 5.34 per cent on Thursday, the highest since the spring of 2002; the 30-year touched 5.69 per cent on Thursday, also a first since 2002, and finished Friday around 5.63 per cent.
The UK's 30-year gilt yield broke above 6 per cent on Thursday for the first time since 1998, easing back to about 5.9 per cent on Friday, so this end of the market is not only an American story. Put the two together and the real information in the day is this: whatever is pushing the long end up does not sit with the Federal Reserve — the odds of a rise fell to one in five and long yields went up anyway. Two paths again. Should the long end come back down alongside rate expectations, this run was about rate-rise worries and following the Fed's pace is enough. Should rate expectations keep falling while the long end holds, what is weighing on valuations sits with public finances and with supply, and nothing the Fed does will fix it. The thing to watch is whether the 10-year can drop back below 5 per cent.
Hard drives: Toshiba plans to double capacity, Seagate and Western Digital each fell a tenth
$Seagate Technology PLC(STX)$ closed 10.21 per cent lower at US$848.99 and $Western Digital(WDC)$ 10.22 per cent lower at US$415.29, each losing a tenth of its value in a day. The trigger was a report in Nikkei Asia: Toshiba plans to spend about ¥60 billion, roughly US$380 million, expanding its plants in the Philippines so as to double hard-drive capacity for AI data centres by its 2027 financial year, aiming for a 30 per cent share of the market.
What let this line run over the past year was the premise that capacity could not be expanded and the shortage had several more years to go. Public reporting has Seagate's and Western Digital's 2026 output essentially sold out, with most of Seagate's nearline capacity committed through 2028. Those long-term agreements lock in volume and duration, not the premise about how many years the shortage lasts, and that premise is exactly what the Toshiba report moved — the market was not recalculating this year's orders but how much pricing power is left after 2028. Two paths follow. Should no second supplier follow with its own expansion, this fall was an overreaction to a plan, with two years of construction and ramp-up still in between. Should Samsung, Western Digital or others with capacity announce expansions too, a year comes off the shortage across the board and the move up has to be worked out again. The thing to watch is whether a second expansion plan appears over the next month.
Memory and compute: memory followed the drives down, the compute names closed higher
Memory followed the drives down: $SanDisk Corp.(SNDK)$ closed 3.79 per cent lower at US$1,719.99 and $Micron Technology(MU)$ 2.05 per cent lower at US$1,074.89, while $SK hynix(SKHY)$ did not follow and closed 0.84 per cent higher at US$195.13. The names doing compute ran the other way: $NVIDIA(NVDA)$ closed 1.34 per cent higher at US$233.95, $Advanced Micro Devices(AMD)$ 2.95 per cent higher at US$633.91, a record close for it, having first finished above a US$1 trillion market value back on 21 September; $Broadcom(AVGO)$ closed 3.35 per cent higher at US$355.14. The one left behind in that group was Intel, down 0.56 per cent at US$119.33.
No single piece of public news from Friday accounts for the size of the move in chips. Two things can be put on the table: the morning's employment report, and the buyback Nvidia announced the previous Monday — the board approved another US$150 billion, taking the remaining authorisation to US$235 billion, to be used within the financial year ending in January 2028. That second one is Monday's news rather than Friday's; it sits under the share price without explaining why this particular day. So what can be established is that the two ends ran opposite ways, not that either end had fresh news. Two paths follow. Should drives and memory keep falling while compute keeps rising, the market has split the AI chain into two separate books, with upstream supply worked out apart from downstream demand. Should compute turn down in a few days as well, Friday was simply the drive makers moving first. The thing to watch is whether SanDisk and Micron track the drive makers this week.
Side by side: the Fed no longer sets the rate alone, and this year's orders no longer set the shortage
Two things happened on Friday, and they pointed opposite ways. On one side the odds of a rate rise fell to one in five and long yields kept climbing, which says the rate that prices every asset is no longer set by the Federal Reserve alone. On the other side one company announcing an expansion was enough to take a tenth off two drive makers, which says what prices this chain has never been this year's orders but the assumption about how severe the shortage stays over the next several years. The first changes how steeply future money is discounted back to today; the second changes how many years that money keeps arriving. Friday put both on the table at once.
The above is personal analysis, not investment advice.
💬 【Talking Point】
Memory fell and compute rose on the same day. Which of the two do you think is resting on the more fragile assumption?
💰 【Bounty】
Drop your view in the comments and there are Tiger Coins in it for you! 🎁
🔔 Better shared than saved — tag a friend and split the coins!
Disclaimer: Investing carries risk. This is not financial advice. The above content should not be regarded as an offer, recommendation, or solicitation on acquiring or disposing of any financial products, any associated discussions, comments, or posts by author or other users should not be considered as such either. It is solely for general information purpose only, which does not consider your own investment objectives, financial situations or needs. TTM assumes no responsibility or warranty for the accuracy and completeness of the information, investors should do their own research and may seek professional advice before investing.
What say you!
因为算力这边现在交易的核心还是 AI CapEx、GPU/ASIC需求和实际计算负载增长，只要超大规模云厂商继续扩张、AI Agent 推理需求继续增加，算力需求就有基本面支撑。
但存储这边除了需求，还多了一层非常关键的前提：
需求增长 + 供给纪律 = 高价格和高利润率。
一旦东芝扩产只是开始，后面再有更多厂商跟进，那么即使 AI 存储需求继续增长，也可能出现 供给增速跑赢需求增速，最后先压 ASP，再压毛利率。
所以我会把两边拆开看：
算力最大的风险是需求不及预期；
存储最大的风险是需求没问题，但供给突然多了。
而周期行业里，第二种情况往往更危险，因为利润率拐点可能出现在需求仍然很强的时候。
我接下来最关注的不是某一天 SNDK、MU 跌不跌，而是 长协价格、库存、产能计划和毛利率是否开始同步松动。
一句话：
算力靠需求证明自己，存储除了需求，还必须继续证明供给纪律没有崩。
Memory looks more fragile.
Memory/storage prices depend heavily on supply shortages. If Toshiba or other companies increase production, prices and profits could fall quickly.
Compute like NVIDIA, AMD and Broadcom has a stronger long-term driver: AI demand. Even if hardware supply improves, AI companies still need more computing power for training and inference.
However, compute stocks are not risk-free. Their valuations are already high, so slower AI spending could cause a sharp correction.
The weak jobs report is also important. If the economy weakens and the Fed cuts rates, that could support high-growth tech stocks—but falling yields caused by a recession would be a different story.
Bottom line: I would be more cautious about memory/storage because the shortage can disappear when supply increases. For long-term investors, I prefer profitable AI compute leaders over companies whose profits mainly depend on a temporary supply shortage.