The Federal Reserve decides at 18:00 UTC today with the 10-year Treasury yield at 4.996% — its highest close since October 2023, after touching 5.00% on Monday and printing 5.008% on Tuesday morning. The 30-year is 5.36%, the 5-year 4.83%, the three-month bill 3.96%. A quarter-point hike is priced at roughly four in five and the market has stopped arguing about it.
What it is arguing about is the number after the hike, and the clearest way to read that argument is the way Ed Yardeni read it on Tuesday: the 2-year Treasury yield sits about 100 basis points above the federal funds rate, the widest gap since 2022. A 2-year that far above the policy rate is the bond market saying it expects the policy rate to rise toward it. That is a sequence, priced. Today’s quarter point is the first step of it or the whole of it, and the material that decides which is not the statement or the press conference but the dot plot at 18:00 and the 2-year’s move in the minute after.
This Pulse is about what a 5% 10-year costs. Not whether the AI build-out gets built — that argument ran all summer and the weekend’s pacing news reopened it in a different form — but what a dollar of 2027 earnings is worth when the risk-free rate is five instead of four and a third. Yardeni put a number on it. We put ours beside it, and then we run it through the two halves of the AI trade and the three levels this diary has carried since August.
Closelooknet on Google
Get our reads first in Top Stories and AI answers. One tap, no account needed here.
The arithmetic: a multiple, not an earnings story
Yardeni’s ‘Proceed with Caution’ on Tuesday cut his year-end S&P 500 target from 8,400 to 7,900 and moved 8,400 out to mid-2027. The earnings did not move: $425 a share for 2027, against $419.53 from the industry analysts, which he expects to rise to meet him by year-end. What moved was the forward multiple, from 19.8 to 18.6, ‘given the recent backup in bond yields’. His range for the year — 7,225 to 8,500, which is $425 at 17 and at 20 times — is unchanged, and 7,900 is 18.6 times, comfortably inside it. He lowered the odds of his Roaring 2020s base case from 80% to 70%, raised a bearish outcome over the next three to six months to 30%, kept 10,000 for the end of the decade and no recession through it. And he moved, in his own words, from ‘we will worry about a debt crisis when the bond market worries about one’ to ‘we are starting to worry now’, on a 10-year he sees on the verge of breaking out above 5.00%.
The shape of the cut is the point. A bull who touches the multiple and not the earnings is marking his book to the bond market, and the arithmetic is simple enough to do at the desk. With the S&P 500 near 7,600 on Tuesday’s close, 7,900 is about 4% above; 8,400 was about 10% above. The difference between those two — the six points of upside that went away — is the difference between 19.8 and 18.6 times the same earnings, and 1.2 turns of multiple is roughly what a 70-basis-point rise in the 10-year does to an equity risk premium that does not compress to offset it. Yardeni’s comfort is that the yield is still well below nominal GDP growth of 6.6% and that the surge in the 10-year TIPS yield in recent weeks reads as better growth rather than fear — the same reading this diary gave on Tuesday from breakevens that have not moved and a credit tape that has not widened. His worry is oil above $100, kept there in his account by the IRGC’s campaign against Gulf oil facilities ahead of the US midterms, because oil high for longer is what turns a hike into a cycle by keeping inflation from settling. We keep 7,900 as an outside marker beside our own lines; we do not move a line because a target moved.
The curve: a step or a sequence, decided at 18:00
Chair Warsh has said repeatedly that he does not intend to give forward guidance and would rather listen to the fixed-income market. On a day the market is guiding — a 2-year 100 basis points over the funds rate — that hands the interpretation of the hike to the Summary of Economic Projections and the dot plot, which publish with the statement at 18:00 UTC, and to the number of dissents. The reading is mechanical. If the median dot shows one further move and the committee is not badly split, the 2-year falls after the statement, the 10-year can come back under 5%, and the equity lines below hold with room. If the median shows two or more and the dissents run in both directions, the 2-year holds or rises, the 10-year goes through 5.00% on a hiking day, and every level on this page is being tested by the cost of money rather than by anything the AI trade is doing. The tell is the 2-year, not the 10-year, and it is at 18:00, not 18:30.
The other half of the curve story is not in Washington. Yardeni’s argument — and it deserves to be taken as an argument rather than as colour — is that the global sell-off in bonds is in part the Japanese carry trade unwinding: years of near-zero rates and a weak yen let hedge funds borrow in Tokyo and buy government bonds everywhere, and a Bank of Japan that has been tightening since 2024, with a hike to 1.25% expected on Friday and a yen at 155.3 that has strengthened on it, is calling that money home. Treasury Secretary Bessent’s ‘global issues’ in his testimony on Tuesday, and his ‘I am the house now’ on 8 September, read in that light as a statement that he will intervene to support the yen so Japan does not have to sell Treasuries to defend it — and, if the Bond Vigilantes run, that he will buy back bonds and issue bills to fund it. A 10-year that goes through 5% on Wednesday could be pushed further on Friday by a decision in Tokyo. The yen at 155 is the line: under 153 with Japanese yields rising is the carry trade unwinding into the Treasury market; back over 156 says the driver is elsewhere.
What 5% does to the AI trade’s two halves, and to our levels
Monday’s sort — the sellers of compute sold, the buyers bought — was the market lowering the premium it pays for the scarce inputs to a race, on a new fact about the frontier. Tuesday marked the buyers on the discount rate: Microsoft −1.5%, Alphabet −1.4%, Amazon −1.5% at the close, while the semiconductor ETF rose 1.0% to $502.07 and the equipment names that were sold at midday — Lam, Applied Materials, KLA — recovered to flat. A 5% 10-year reads differently on the two halves and it is worth being precise. The sellers of compute — the chip ETF under $505, the memory ETF under $58, the Rubin build-out down 14% on the month — have already been repriced on the premium; a higher discount rate hits them again, but the larger part of their move is behind them, and equipment closing flat after being sold at noon is the first sign the volume question is being held open rather than answered in the negative. The buyers of compute — the four largest software and platform names and, with them, the security group — carry the longest duration in the index; they were bought on Monday on the premium argument and they are the names a 5% 10-year hurts most from here. The security group holding its bid through a rates day, CrowdStrike +2.4% and Zscaler +0.9% while the software ETF fell 1%, is the one part of the trade both facts are leaving alone, and the agentic index’s internals said the same for a second session: Foundation Models −5.7%, Security and Operations +2%.
The print record scored the same split in two names. Oracle beat on every line and was sold 8.3% over its three-session window, because its guide is a claim about contracted compute converting on schedule and the week put the schedule in question; Adobe beat on both lines and was paid 3.9%, because its guide is a claim about applications selling. Two clean beats, one sold as a seller of compute, one paid as a user of it — the first pair on the record split by AI exposure rather than by the numbers or the tape.
On the index, the levels this diary has carried since August were drawn in a market with a 10-year in the 4s: 757.83 on the S&P 500 ETF, 704 and 708.69 on the Nasdaq-100 fund, 505 on the chip ETF. The S&P 500 ETF spent Tuesday afternoon under 757.83 and closed at 758.83, a dollar above it — the market bought the close on a 5.008% intraday print, which is not what a market that has decided the cycle does. 704 held. 505 did not, but the chip ETF closed higher with the design names bought and the equipment names flat, the pattern of a repricing finding its level rather than a liquidation looking for one. What a 10-year above 5% does to these lines is what Yardeni did to his target: it lowers the multiple they were drawn at. A close under 757.83 and 704 together, on a hiking day, with the 10-year above 5.00%, is the index marking its own multiple — October’s resolution arriving two weeks early. One without the other is a rates day inside the range.
Our signal: four boards, one reading
Money Temperature fell to 44 on Tuesday’s closes from 46: the S&P 500 and Nasdaq-100 funds at 44, the 20-year-plus Treasury fund at 29, gold 33, the dollar 69 — nothing is a haven in this move except the currency, which is the shape of a duration repricing rather than a flight. The factor regime reads momentum leadership cooling but still above trend: the momentum-to-low-volatility ratio at the 88th percentile of its window and −4.5% over twenty sessions, −10.5% over sixty — defensive rotation in progress, not complete. Sovereign pressure is 0.45 from 0.56 three weeks ago, with the slope component at −2.8: the long end is doing the work and the curve is flattening into the hike, which is the bear-flattening this diary has written about since the yield march began. And the credit tape, as of Monday’s data, has not moved: high-yield spreads at 2.65%, the 4th percentile of three years, investment grade at 0.80% — the price of AI debt is up because the risk-free rate is up, not because the market doubts who can pay it back. That is the reading that separates a market repricing the cost of money from one repricing solvency, and it is the reading that would change what we do if it changed. It has not.
The indices carry the same message with a week on them. Rubin −6.8% on the week and −14.2% on the month, the longest-duration basket in our universe and the one a multiple cut lands on hardest; the Agentic Winners 40 +1.2% and the agentic ecosystem +3.6% on the week — the buyers and the guards of the frontier on the other side of the same rate. Retail sales at 12:30 UTC are the last data before the decision and the one that can move the dots if it is strong, because a resilient consumer at $100 oil is the Fed’s inflation case. The decision and the dots at 18:00, the chair at 18:30, the Bank of Japan on Friday. We do nothing before 18:30; we read the dots, then the 2-year, then the presser, then the close against 757.83, 704 and 505 with the 10-year against 5.00%. The market has been told in one week that the frontier will be paced and the cost of money is 5%, and it has answered by sorting the AI trade rather than selling it and by marking duration rather than credit. Tonight decides whether the second of those holds — and the number that decides it is not the one at 18:00. It is the one the 2-year prints at 18:01.