The 5-year Treasury yield rose to 4.859%, the 10-year closed at 5.006% — its first close above 5% of this cycle — and the 30-year fell to 5.349%. The equity tape gave back a midday relief: the S&P 500 ETF closed at $754.05, its second close under Thursday’s $757.83; the software ETF fell a third day; gold was sold; the Magnificent Seven fund closed on its 69.5 line.
Nothing in that list was about earnings. The companies that were sold on Wednesday will report in October roughly what they were going to report on Tuesday. What changed was the other half of the fraction, and this Pulse is about the fraction — because the same arithmetic that sold software and gold on Wednesday is the arithmetic that decides how the AI trade is priced for the rest of the year.
Closelooknet on Google
Get our reads first in Top Stories and AI answers. One tap, no account needed here.
The fraction: a numerator, a denominator, and a direction
Every asset that pays in the future is priced as a fraction. On top sit the cash flows — earnings, dividends, the output of the business; that is the numerator. Underneath sits the rate the market uses to bring those future dollars back to today — the risk-free yield plus a premium for the risk; that is the denominator. The discounted-cash-flow model that every analyst runs is only that fraction extended over many years, and the further out the cash flows sit, the more the denominator matters. That sensitivity is what the bond market calls duration, and it applies to stocks exactly as it applies to bonds.
The part that gets missed is that the price does not respond to the level of the denominator. It responds to a change in the direction of the path the market expects the denominator to take. A 5% 10-year that the market has expected for months is in the price; a 10-year that the market expected to fall toward 4% and that now looks like it will sit at 5% or rise is a repricing, because every model that discounted 2028 cash flows at a falling rate has to be re-run at a flat or rising one. The change is in the anticipated forward path, and the forward path is what moved on Wednesday: from a hike that was priced as one step to a hike that the wires now count as the first of a sequence.
We have a house term for the other half of this, the numerator regime: a market phase in which returns come from the top of the fraction, earnings and physical capacity growing, rather than from the bottom, multiples expanding as rates fall. The AI build-out has been a numerator regime since 2023 — Rubin, our build-out index, is +83% this year on earnings and capacity, not on a lower discount rate. The point of Wednesday is that a numerator regime does not exempt anyone from the denominator. It only decides who has enough numerator to absorb it.
2022, the year the denominator did all the work
The cleanest example in recent memory is the first half of 2022. The 10-year went from roughly 1.5% at the start of the year to roughly 3.5% by mid-June as the Federal Reserve turned from zero to the fastest hiking cycle in forty years. The software companies that were sold hardest did not miss. Quarter after quarter, most of them reported revenue and earnings above what the Street expected; the software ETF still fell by about a third in six months, and the index of cloud companies that Bessemer maintains more than halved from its November 2021 peak. That was the first software drawdown of this cycle, and it was a pure denominator event.
The arithmetic explains why the beats did not help. A subscription-software company in 2021 was valued on cash flows that sat five to ten years out, because that was when the growth was supposed to convert into margin; at a 1.5% risk-free rate those distant dollars were worth nearly what they would be worth today. At 3.5%, and with a path that pointed higher, the same dollars were worth a fraction of that — and a beat of 3% on this quarter’s revenue does not move a model whose value sits in 2029. The numerator improved slightly; the denominator changed direction; the denominator won by a wide margin.
Wednesday was the small version of the same event. The Agentic Winners 40, our index of the companies selling agentic software, fell 1.3% with Adobe −2.8%, Intuit −3.4%, Duolingo −3.7%, UiPath −4.2%; the software ETF −0.6% to $104.96 on its third day of selling. Adobe beat last week and was paid for it over its three-day window; on Wednesday it was marked on the 5-year yield like everything else in its index, and it closed under the line it was paid above. That is not the market changing its mind about Adobe’s quarter. It is the market re-running Adobe’s model at a different path.
Gold and bitcoin: all denominator, no numerator
The assets that pay nothing are the purest case, because they have no numerator at all. Gold and bitcoin produce no cash flow, so their price cannot be a fraction of earnings over a rate; it is the inverse of the opportunity cost of holding them. That opportunity cost is the real yield — what a Treasury pays after inflation — and the direction of its expected path. When the market expects real yields to fall, holding an asset that pays nothing costs less every year and its price rises; when the path turns up, the cost of holding it rises and the price falls, whatever the level.
2022 showed this too. Gold traded above $2,000 in March 2022 and near $1,650 by the autumn while real yields went from about −1% to about +1.5%; bitcoin went from roughly $47,000 at the start of the year to under $20,000 by June. Neither asset’s ‘fundamentals’ changed — there were none to change. The path of the denominator changed, and the assets that are all denominator moved most.
Wednesday repeated the gold half and skipped the bitcoin half. Gold was +1.4% at midday when the tape expected relief, closed −0.6%, and is −1.2% overnight at $4,335: the hedge against a central bank behind the curve was sold when the central bank showed it was not. Bitcoin held, +0.5% at $76,318, trading with the equity futures rather than with gold — which our cointegration board has been flagging for weeks as the crypto-tech pair breaking. We note the divergence and do not explain it; an asset that is all denominator and did not fall on a hike is either being held for a different reason than gold or is late, and the next two sessions will say which.
Who absorbs a hike: the megacaps, chips, software, and the sectors whose numerator rises with rates
Our rule of thumb from the rate-up phases of the last decade is a ranking by how much numerator an asset has close to today. The Magnificent Seven tend to perform about as well as the market in these phases, because their cash flows are now, not in 2029, and their buybacks shorten their duration; the fund that holds them closed at $69.52 on Wednesday, on its 69.5 line, flat on a day software fell. Semiconductors do better than software, because their numerator is a volume — units shipped this quarter and next — that rises with a strong economy; on Wednesday the chip index split, the design names and the optical interconnect names bought while the equipment names were sold a third time, which is the market sorting chips by how near their volume is. Software does worst, because its cash flows sit furthest out and its beats do not move the denominator; that is 2022’s lesson and it printed again on Wednesday.
Then there are the sectors whose numerator rises with the reason rates are rising. When rates go up because the economy is stronger than perceived and inflation is stickier, the businesses that sell the inputs to that economy earn more: agribusiness, materials, energy, the banks whose lending margins widen with the curve. In 2022 the energy sector rose by more than half while the index fell 18%; the agribusiness fund finished the year roughly where it started; materials fell, but by less than the index. Those are the numerator-positive sectors of a denominator-negative year, and they are where a diary that carries the AI build-out looks for the other side of its book when the path turns.
Inside our own indices the same ranking is visible. Rubin, whose constituents sell the physical inputs to the build-out — the wiring, the power, the materials — rose 1.9% on Wednesday with its Interconnect layer +4.6% and Advanced Materials +4.2%, on a day the software index fell; HALO, our growth index outside the megacaps, was −0.3% with its energy-transition sleeve −2.6%, the longest-duration part of a growth basket taking the hit. The build-out has a near numerator. The subscription has a far one.
The level a market can absorb, and where Warsh drew it
None of this means rising rates are simply bad for equities. There is a level up to which a market can absorb a stronger-than-perceived economy coupled with moderately rising rates, and can absorb it well — because a stronger economy raises the numerator for most companies at the same time as the rate raises the denominator, and as long as the earnings revisions come faster than the path steepens, the fraction goes up. That was the chair’s message on Wednesday as the wires read it: the economy is stronger than perceived, the committee can raise moderately, and the market can take it. The years from 2023 to now, with the 10-year rising from the mid-3s toward 5% while the AI earnings cycle ran, are the proof that this works when the numerator is strong enough.
The level is crossed when the path steepens faster than the earnings revise. In 2022 that took weeks, because the path went from zero to three-and-a-half in a straight line and the numerator was not growing fast enough to matter. In 2023 and 2024 it was never crossed, because the AI numerator grew faster than any denominator the Fed put under it. The question the tape is asking now is which of those two years this one resembles, and the honest answer is that it depends on the slope of the path — ‘one more this year’, as CNBC counted the projections, is a moderate slope that a strong numerator absorbs; a 2-year yield about 100 basis points above the funds rate, as Ed Yardeni pointed out on Tuesday, is a steeper one, and the belly of the curve selling on Wednesday moved the market toward the steeper reading.
So we carry the lines and the fraction together. On the denominator: 5.00% on the 10-year, which closed above it, and the 5-year as the meter of how much sequence is priced. On the numerator: Micron’s report next week as the first order-book fact of the quarter for the volume story, and the equipment names against the design names at each close. On the assets with no numerator: gold, which is telling us the market believes the central bank, and bitcoin, which is not yet telling us anything. And on the line Warsh drew: the equal-weight S&P, which fell 0.8% on Wednesday and has fallen on both post-decision days — because if the economy is strong enough to absorb the hike, the average stock is where that should show first, and it has not shown yet. This is our diary of what we hold and why; it is not a recommendation to anyone else.