The AI trade has spent the year arguing about demand. This week the argument is about the discount rate, and the cleanest way to see it is not in the Fed funds futures but in TIPS — Treasury Inflation-Protected Securities, the bonds whose yield is the real rate the market charges after inflation. On Friday the 10-year real yield closed at 2.60%, according to the Federal Reserve’s daily series, up from 2.32% on 25 August and from 2.43% a week earlier. The 10-year breakeven — the inflation the market expects over ten years, read as the gap between nominal and real yields — was 2.37% on Monday, within five basis points of where it was in late August.
Put those two together and the move in Treasuries has a clear shape. The 10-year nominal yield touched 5.00% on Monday and closed 4.975%; a month ago it was 4.70%. Of the roughly 30 basis points added, five are inflation expectations and the rest is real. The five-year says it louder: the real five-year yield rose from 2.04% to 2.38% between 25 August and Friday, the five-year nominal from 4.41% to 4.79%, and the five-year breakeven from 2.31% to 2.40%. The market is not pricing an inflation problem into the Fed’s hike; it is pricing a higher real cost of money into everything with a long cash flow — and nothing in the AI build-out has a short one.
Closelooknet on Google
Get our reads first in Top Stories and AI answers. One tap, no account needed here.
This edition connects the rates discussion of Fed week to the build-out through three of our own boards: the sovereign-pressure board for the curve, the AI credit stress tape for what the issuers actually pay, and the Money Temperature and factor gauges for how equities are absorbing it. The reading is the same on all three, and it is more specific than “rates up, tech down”.
The rise is real: breakevens flat, TIPS down, the curve flattening
The TIPS funds show the move in prices. The broad TIPS ETF (TIP) closed $105.82 on Monday, down 1.7% from its 25 August high of $107.64; the long-dated TIPS fund (LTPZ) $47.28, down 2.6% from $48.52; the short-dated fund (STIP) $100.29, down 0.7%. A TIPS fund falls when real yields rise — it is the one part of the bond market that cannot blame inflation for its losses — and the long-dated fund falling most is the signature of a real-rate move that reaches the far end of the curve. The nominal funds fell with them: the 20-year-plus Treasury fund (TLT) $80.93, down 3.0% from its late-August high, the 7-to-10-year fund $90.93.
The curve is flattening from the front. The 13-week bill yield went from 3.70% to 3.91% over the month, 3.94% this morning — that is the hike being priced, a quarter point at roughly four in five for tomorrow — the five-year rose 43 basis points, the ten-year 27, the thirty-year 8, to 5.35%. Our sovereign-pressure board carries the two-year at 4.68%, the ten at 4.99% and the thirty at 5.35% on Monday’s closes, a two-to-ten spread of about 31 basis points; its Sovereign Pressure Index reads 0.45, down from 0.56 three weeks ago, because the slope component is deeply negative — a bear flattening scores as lower pressure on the long end even while the level of yields rises. And it is not an American story: the G7 equal-weight ten-year is 4.24%, up 30 basis points in 21 days, with Germany at 3.53% (+31), France 4.50% (+44), Italy 4.40% (+40), the UK 5.32% (+26), Canada 3.96% (+34), Japan 2.99% (+11).
The Money Temperature board adds the piece that matters for positioning: Treasuries are not a haven in this move. The long-bond tracker’s temperature is 31, gold’s is 32, equities 52 and 49; the regime board’s Duration Rotation spread is −21, which in its language means no flight to quality is under way, and its Risk-versus-Haven spread is +20. The composite Money Temperature fell to 46 on Monday from 50 on Friday. A market where equities lose their line and bonds do not catch a bid is a market repricing the discount rate, not fleeing risk — which is exactly what the real-yield series says.
The credit tape: the price of AI debt is up, the doubt is not
Now the issuers. Our AI credit stress tape decomposes what an AI borrower pays into the Treasury rate, the market’s credit spread, its rating cohort, and a residual that only the issuer explains — the residual is the product, because spread levels alone print “no stress” forever when they are as tight as they are. The environment on Friday: investment-grade option-adjusted spread 0.80%, unchanged over twenty days and at the 30th percentile of the last three years; BBB 0.97%; single-B 2.73%, the 7th percentile; high-yield 2.65%, the 4th percentile — about as tight as the last three years have seen. European high yield 2.68%, up 14 basis points in twenty days, is the only spread on the board that widened. Leveraged loans, which float with the front end, are flat on the month (BKLN $20.61, +0.1%) while the fixed-rate credit funds fell with duration: investment-grade (LQD) −1.7%, high-yield (HYG) −1.5%. The price of credit fell because Treasuries fell; the spread over Treasuries did not move.
The bellwether bonds say the same thing name by name. Oracle’s 4.375% of May 2055 yielded 7.62% on Monday at a price of $62.37 — up 63 basis points since June, 14 of them in the last twenty days. Nvidia’s 3.5% of April 2050 yielded 6.26%, up 74 basis points since June, 9 in the last twenty days. Alphabet’s 2.25% of August 2060 yielded 6.03%, up 52 since June. The quality names’ long bonds rose more than the stressed name’s over the summer, which is what a Treasury-driven move looks like; Oracle’s spread over Nvidia is 1.36 percentage points, wider by only 5 basis points in twenty days, its spread over Alphabet 1.59 points, wider by 9. The AI residual is small. The market charges Oracle for being Oracle — a company whose capex is not covered by operating cash flow, with interest covered 4.5 times and net debt at 3.2 times EBITDA — and it has charged that for months. What changed since June is the risk-free rate under all three.
The funding ladder puts the stress where the cycle normally puts it. Our six rungs run from the balance-sheet funders at the top (Microsoft, Alphabet, Meta, Amazon — stress score 7, free cash flow covering capex 1.34 times) through Oracle alone on the corporate-debt rung (score 41, cover 0.62) to the project-funded builders (CoreWeave, Nebius, TeraWulf, Applied Digital — score 86, cover 0.18, interest not serviced from the business, external funding at 1.10 times capex). The ladder reading is T3: stress at the project-funded rung, which the board labels normal cycle behaviour and asks you to watch for climbing. One flag on the top rung is worth the sentence: Amazon’s capex is not covered by its operating cash flow over the trailing four quarters and its interest coverage has fallen from 37 to 21 times in five quarters — the board tags it as drifting toward debt-funded behaviour. Nothing on the ladder has climbed this week. What has happened is that every rung’s cost of carry rose by the same real-rate move.
Our signal: three boards, one reading
Read together, the boards give a specific answer to the question the market has been asking since Saturday’s pacing essay. The physical layer of the AI trade — chips, memory, optical, power, the data-centre backlog — is the longest-duration asset in the market: its cash flows sit years out, in backlogs like Oracle’s $664 billion and in 2050 and 2055 bonds. A 28-basis-point rise in the ten-year real yield lowers the present value of those cash flows more than it lowers the value of a software subscription collected this quarter. Monday’s split — the semiconductor ETF −5.6% to $497.40, the software ETF +5.0%, Microsoft, Alphabet and Meta up 2% to 3% — is what a real-rate shock does to a trade sorted by duration. The pacing essay gave the market a reason; the real-yield series gave it the discount rate.
The factor gauge agrees. The momentum-over-low-volatility ratio (SPMO over SPLV) sits at the 88th percentile of its window but has fallen 2.7% in twenty days and 7.9% in sixty; the board’s regime label is “defensive rotation — low-vol leading”. The cointegration monitor has one pair locked and six breaking, with the dominoes at the top of the cascade — bitcoin against the Nasdaq-100, the Nasdaq-100 against the S&P 500 — already fallen and the S&P against gold breaking: relationships that held while the discount rate was stable are loosening as it moves. And the Rubin build-out index, our basket of the physical layer, lost 5.25% on Monday with all eight layers red and is down 11.9% on the month while the Agentic Winners, the basket of the buyers of compute, gained 4.0% and is up 4.0% on the month. Our two US baskets are pointing in opposite directions on the same day, and the real-rate series is the variable that separates them.
What the signal is not: it is not a credit signal. The tape is explicit that spreads are at multi-year tights and the AI residual is a few basis points. The doubt about the build-out that shows up in a 7.6% Oracle bond is a doubt about how long the money is tied up at a 2.6% real rate, not a doubt that Oracle pays. That distinction decides what to watch. A credit event would show first in the ladder climbing — T3 stress reaching T2, Oracle’s residual widening past its June range — and in high-yield spreads leaving the 4th percentile. A rates event shows in the real yield, and it is already showing. We are in the second, not the first.
Into the Fed: the levels that decide the week
The decision comes at 18:00 UTC tomorrow with a quarter-point increase priced at about four in five, taken into an energy shock — oil above $100 for a week, WTI $103.6 this morning — a −8.7 Empire State survey, and a ten-year that touched 5% on Monday. The outcome that helps both halves of the AI trade is a hike the long end takes as the last one: the ten-year closing below 5.00% on the decision, the real yield backing off 2.60%, breakevens unchanged. The outcome that turns Monday’s chip sale into an index sale is a hike that pushes the ten-year through 5% with the real yield making the move, because that is the same discount-rate shock applied to the whole market rather than to its longest-duration corner.
The levels we hold into it. On the equity side: the S&P 500 tracker’s $757.83 and the Nasdaq-100 tracker’s $708.69, Thursday’s closes — pre-market this morning has them at $756.83 and $705.00, both below, with 704 and 694 the next shelves on the Nasdaq-100; the semiconductor ETF’s $505, lost on Monday at $497.40, with $517.43 and the 50-day at $531.52 to recover; the Magnificent Seven fund’s 69.5 weekly line, held at $70.10. On the rates side: 5.00% on the ten-year, touched and not closed above; 2.60% on the ten-year real yield, the number this edition is about; the two-to-ten spread at about 30 basis points, which a hike could push toward zero. On the credit side: high-yield spreads leaving the 4th percentile, and Oracle’s residual over Nvidia widening past 1.36 points — neither has happened.
Tonight, before any of it, the print record scores Oracle and Adobe at the close: Oracle at $144.79 sits below its $148.35 sold line from a $152.94 entry, Adobe at $265.60 above its $256.29 paid line from $248.83. Oracle is the hinge of the credit ladder and the seller of compute that borrows to build; Adobe is the buyer of compute at falling prices. If the record scores them sold and paid on the same evening, it will have measured, in two closes, the same thing the real-yield series measured in twelve sessions.
This is a diary, not advice. The boards are public — the credit stress tape, the sovereign-pressure board, the Money Temperature and the factor gauge — and the point of reading them together before the decision is that we cannot rewrite the reading after it.
The signals behind thisEach line links to the tool it comes from
LabAI Credit Stress Tape — IG 0.80%, HY 2.65% (4th percentile of 3 years); ladder reading T3; Oracle 2055 at 7.62%, Nvidia 2050 at 6.26%, Oracle’s residual over Nvidia +5 bp in 20 days→LabSovereign Pressure — US 2y 4.68% / 10y 4.99% / 30y 5.35% on Monday; SPI 0.45 from 0.56 with the slope component −2.8; G7 ten-year 4.24%, +30 bp in 21 days→LabMoney Temperature 46 (from 50) — equities 52/49, Treasuries 31, gold 32; Duration Rotation −21: no flight to quality in this move→LabFactor regime — SPMO/SPLV at the 88th percentile but −2.7% in 20 days, −7.9% in 60: defensive rotation, low-vol leading→LabCointegration monitor — 1 pair locked, 6 breaking; BTC–QQQ and QQQ–SPY fallen, SPY–GLD breaking→IndicesRubin Build-Out 100 — −5.25% Monday with all eight layers red, −11.9% on the month: the longest-duration basket in the trade→IndicesAgentic Winners 40 — +4.0% Monday, +4.0% on the month: the buyers of compute on the other side of the same real-rate move→SignalsPrint record — Oracle $144.79 vs its $148.35 sold line and Adobe $265.60 vs its $256.29 paid line, both scored at tonight’s close→