The bond market had its worst week in twenty years and equities went up anyway. The 30-year Treasury yield touched 5.501% on Thursday, its highest since June 2004. The 10-year reached 5.223%, the highest since June 2007, and closed the week around 5.17%. Both went straight through the 4.85% to 5.10% range this column called for seven days ago. The S&P 500 still closed at 7,743.41, up 1.2%. The Nasdaq gained 2.1% to 27,068.72 after a record close of 27,244.28 on Tuesday. The Dow added 0.3% to 51,828.62 and snapped a three-week losing streak.
The cause of the rout was good news. S&P Global's flash composite PMI jumped to 58.4 in September from 56.0, the strongest since July 2021, with input costs rising at the fastest pace since October 2022 and the survey implying roughly 5% annualized growth. Jobless claims fell to 197,000, one of the lowest readings since 1969. Futures now put about 72% odds on another hike on October 28. Brent swung from $97.69 on Tuesday to $108.40 on Thursday and settled just under $106. Gold fell about 2% to near $4,280.
Scorecard: September 21
Seven rows. Five right, one badly wrong, one never graded because I put it on the wrong day.
- August PCE on Friday, 3.6% headline and 3.2% core or lower. Void, and my fault. The BEA publishes August personal income and outlays on Wednesday, September 30, not Friday the 25th. A calendar error, not a forecast error. The call carries over below, on the right day.
- 10-year in a 4.85% to 5.10% range. Wrong, and not narrowly. Thursday's high was 5.223%.
- S&P 500 retakes 7,700 by Friday if Brent stays under $105. Right on the level, wrong on the reason. The index closed at 7,743 while Brent spent Thursday above $108.
- Nasdaq makes a new closing high above 26,700. Right. 27,244.28 on Tuesday.
- Micron holds $1,000 into the print. Right. An all-time intraday high, and Citi raised its target from $1,150 to $1,300.
- Bitcoin confirms the September 16 low with three straight days of ETF inflows and reclaims $80,000. Right, emphatically. Spot ETFs took in $2.4 billion, the best week since October 2025, flipping 2026 flows positive. Bitcoin traded near $84,250 on Friday.
- Brent trades below $100. Right. $97.69 on Tuesday, after Iran offered to reopen the Strait of Hormuz within a week. It did not stay there.
The rate miss is the one worth understanding. I built that ceiling on the view that the dot plot had already priced the tightening, so the hike was the event. It was not. Wednesday's PMI was. A yield set by a Fed that has published its plan behaves like a range. A yield set by growth reaccelerating does not.
A 5.2% Ten-Year Is Not the Same Trade as a 5.0% One
The September 21 argument was that a market shrugging at 5% was bullish, because the drivers Chair Warsh named, economic strength, competition for capital from hyperscaler capex, and geopolitics, are not recession signals. Last week strengthened that case. Even housing, the soft spot two weeks ago, cleared at a lower price rather than freezing: new home sales jumped 6.4% to 684,000 with the median price down 5.8% year over year.
What changed is the composition. Nearly all of the index gain came from two places. AMD rose about 10% on Monday to an intraday $616 and crossed a $1 trillion market cap for the first time, the fourth semiconductor company ever to do it, with Intel up roughly 12% the same session and the chip index up more than 3% into Thursday's state dinner, where Jensen Huang, Lisa Su, Tim Cook, and Satya Nadella sat with China's head of state. Meta rose about 14% on the week after Muse, the AI agent it launched September 8, passed 3.4 million App Store downloads.
A narrow rally on a broad rate shock only works if AI earnings are rate-insensitive, and they are not. They are growing fast enough to outrun the discount rate, which is a different thing. At a 5.5% long bond the terminal value in every AI model shrinks, so the names that keep clearing the bar are the ones with cash flow this year rather than in 2030. That filter tightens with every basis point.
Micron Is the Week
Micron reports fiscal fourth quarter results after Wednesday's close. Consensus is roughly $51 billion of revenue, up about 350% year over year, and $31.50 of earnings per share against $3.03 a year ago. The stock is up roughly 270% in 2026 and made an all-time high last week.
None of that is the trade. The trade is the November guide and the fiscal 2027 capital plan. The August 3 post argued that July's 29% drawdown on supply-wave fear was a buying opportunity because the shortage is structural, not cyclical, and Samsung and SK Hynix both say it runs to 2028. The August 17 capex framework inverts here: for a hyperscaler, capex discipline protects margins, and for Micron it protects pricing. A guide above $55 billion with slow capacity additions keeps the cycle intact. An aggressive 2027 build gets the print sold however good the quarter was.
This Week
- Wednesday: August PCE, the third estimate of second quarter GDP, and ADP. Warsh put headline near 3.6% and core near 3.2% at the September press conference. Core ran 3.3% in July. At or below 3.2% is the first real inflation relief since spring. At 3.4% or higher, October stops being a question. Government funding also runs out that day; a clean extension is the base case.
- Thursday: ISM manufacturing, Accenture, and Nike after the close. Nike's quarter ended August 31 with consensus at $0.44 on $11.35 billion of revenue, a year over year decline. Accenture reads whether enterprises are still paying for AI implementation.
- Friday: September payrolls. Consensus is about 90,000 with unemployment at 4.1%, against 162,000 in August. With claims at 197,000 the risk is to the upside, and a surprise stacked on a 58.4 PMI is what puts the 10-year through 5.30%.
- Bitcoin. Seven straight sessions of ETF inflows and 2026 flows back in the black is the setup the January call needed. The $150,000 target stands and the retirement level is still a weekly close below $55,000.
Key Levels for the Week
| August core PCE on Wednesday | 3.2% or lower year over year |
| September payrolls on Friday | Above the 90,000 consensus, unemployment at 4.1% |
| 10-year yield, the level that now matters | Weekly close below 5.30% |
| Micron fiscal Q4 on Wednesday | Revenue above $51 billion, November guide above $55 billion |
| S&P 500, support on a closing basis | 7,700 holds |
| Nasdaq | Retakes the 27,244 record close by Friday |
| Bitcoin | Holds $80,000 and the ETF inflow streak survives the week |
My Take
The week proved the thesis and broke the range at once. Growth is accelerating, nobody is being laid off, and the long end is repricing to fit an economy running near 5% rather than to a Fed that has already published its intentions. Equities can live with that as long as earnings growth outruns the discount rate, which is exactly what last week's leadership was saying: sold-out memory, and chips with cash flow now.
Playbook: stay with memory, banks, and short Treasuries, and retire 5% as the line in the sand, because 5.30% on the 10-year is the level that matters now; own Micron into Wednesday for the November guide rather than the quarter, and treat an aggressive fiscal 2027 capex plan as the reason to trim into strength; extend duration only on a core PCE print at or below 3.2%; keep Nvidia at BUY with the $268 target intact after a week that left it near $225; hold Bitcoin now that the streak has confirmed the low, and add on a hold of $80,000 rather than chasing.