Dispatch Sunday Set-up

Two of the most-read AI notes of the week disagree about everything except the facts. Jon Gray, on Bloomberg Thursday, called demand for compute “almost parabolic”: model spend up twenty-fold in a year, neocloud pricing up half, H100s re-leasing near their original rates, a shortage he called “indisputable.” Torsten Slok, in Saturday’s Daily Spark, found no payoff at all: revenue per worker and margins rising for the seven, flat for the other 493, no productivity signal in the data, and corporate America beginning to book its token spend as a cost center. “The AI story will be decided in the S&P 493.”

Both are right, and the bond market has already priced the distance between them. Demand for the build is parabolic because the labs and hyperscalers are buying compute on fixed, long-dated obligations. Payoff outside tech is zero on average because it arrives in cost ratios at a few scale adopters before it arrives in anybody’s revenue line. The gap between a parabolic input and a zero output is a timing gap, and a timing gap financed with long fixed obligations is a credit cycle. This week the market put a number on it: an A+ rated, amortizing, senior data-center bond at 241 over Treasuries, 165 wide of the A index, while every developed long end but France’s was bought. The senior tranche got sorted with the junior. Sunday’s job is to say why, before a Fed that is about to tighten into the demand the build itself creates.

The Dispatch publishes every trading morning. Subscribe on LinkedIn or at dispatch.thebondbro.com.

———————————————————————————

THE BOARD · Friday Oct 23, 2026 close · Koyfin, CME, Freddie Mac · green = yield down / spread tighter / price up. Treasury one-day changes against Thursday’s close as recorded Friday morning, one-week against October 16; credit is Friday’s close against Thursday’s, approximate; ETFs Friday’s close.

The curve on the week. Twos 4.692, fives 4.816, tens 5.001, thirties 5.323. Call it two lower at the front, six in the belly, eight in tens and ten at the back from a week ago: a bull flattener again, the second in a row, and this one ran through 5% on the 10-year Thursday before Friday gave back about four basis points across the curve and closed the 10-year on the figure. 2s10s 31, 5s30s 51, 10s30s 32. The one-year at 4.40 sits 30 under twos. Nothing is priced for Wednesday; the question is the dissents.

The Six Tracker, week four close. Thirties 5.323, −10 on the week, the first weekly close under 5.35 and the eighth session under the 5.50 first rung. Freddie Mac 30-year 7.32%, −8 on the October 22 print, the first one to carry the rally. 10s30s 32, −2. Dollar-yen 156.83, −1.1% on the week. All four moved the right way for the refunding, and the one that matters most did it on a flight, not on supply.

Credit, Friday close. Koyfin’s Sunday export has IG at 89 (+1), BBB 111, HY 340 (+13), BB 210, B 346, CCC 1340 (+33). Treat the one-day changes as approximate; the feed’s timing on credit is loose. What is not loose: CCC closed through 1300 for a fourth session and this was the widest of the four. HY is ten from the 350 score line. Euro HY 354, EM HY 365. On an equity day of −0.5%, the senior tranche widened one and the junior widened thirty-three.

Elsewhere. Brent settled 116.72 (+3.1%), WTI 103.27 (+3.8%), on reported Houthi strikes on Saudi infrastructure; Brent is +6.5% on the week, a six-month high, and oil services closed +5.4% at z +2.9. France 10Y 5.037, the Bund 3.343, OAT–Bund 169, in from 178 at Thursday’s wide; Moody’s held Aa3 with a negative outlook Friday night. Gilts 5.366. JGB 10Y 2.796 after a 9 bp rally. India’s equity ETF −2.4% at z −2.3; Korea +1.5%, Brazil +1.8%. Staples −1.9% at z −2.0 and insurance −2.3% were Friday’s sellers; semis +0.7% and software +1.7% bounced.

———————————————————————————

Gray’s case is demand. Blackstone sees it from every seat in the stack: investment-grade credit, junior credit, participating equity, common. The shortage is real because the labs’ spend is contractual. Anthropic’s draft prospectus, as viewed by the Wall Street Journal, carries more than $600 billion of compute commitments, about $450 billion of it noncancelable. Meta’s commitments run past $800 billion. The demand curve looks parabolic because it is a schedule of take-or-pay obligations, not a forecast.

Slok’s case is supply, in the economic sense: what the build produces for the companies that are supposed to pay for it. His chart has revenue per worker and margins rising for the Mag 7 and flat for the 493. The San Francisco Fed’s utilization-adjusted productivity series shows no boost. Companies are beginning to treat token spend as a cost line to be managed down. On average, outside tech, the payoff is zero.

The reconciliation is in the companies that are not average. Bank of America has about 130 AI projects running and guides to roughly $250–300 million of benefit this year and about $700 million next, against two to three times that in cost; its headcount is down about 3,500 on the year. Citi has cut technology vendor spend by about 30% and shortened launch cycles by a third. UnitedHealth is spending $1.5 billion on AI this year against a 2:1 return target; claims that took days now take seconds, and its operating cost ratio fell to 12.8% from 13.5%, which the company does not attribute to AI directly. JPMorgan has said its spend and its measured value are each about $2 billion. Treat all of those as reported, not audited. The pattern is the point: the payoff shows up first in cost ratios and headcount at data-rich scale adopters, not in revenue, and not at the median company. That is why the 493 average looks flat while the adopters’ margins do not.

Gray conceded the risk in the same interview: “you’ve got to deliver on those commitments,” and “if the end market was weaker then it would be highly problematic.” That is the whole thing. The build’s obligations are fixed and long. The build’s revenue is uncertain in timing and concentrated in a few customers. An asset-liability mismatch of that shape has produced a credit cycle every time a real technology was financed this way, from railroads to fiber. The technology was real both times. The timing was wrong both times. Credit paid.

———————————————————————————

Here is what the bond market charges for the gap. Beignet is the name on the Blue Owl joint venture that financed Meta’s Hyperion data center in Louisiana: about $27 billion of 144A paper, a 6.58% coupon due 2049, amortizing, rated A+. It is the senior tranche of the AI build in its purest form. The tenant is Meta. The rating is two notches inside the index average.

It issued at 170 over Treasuries, at par. It traded up to 109, about 135 over. Thursday it closed at 92, about 241 over, the widest it has ever been, per Weinand’s chart. The A-rated corporate index closed Friday at 75. Beignet is 165 basis points wide of its own rating bucket, 130 wide of BBB, and about 70 wider than where it was issued while the index has gone roughly nowhere.

Three reasons, and only one is about Meta. First, the structure: a long lease to a single tenant whose initial term is reported to be short with renewal options and a residual value guarantee, which means the bondholder’s recovery depends on what a 2049 data center is worth if the tenant walks. Second, the market: 144A, $27 billion in a single name, and a buyer base that is now sorting every piece of AI paper by how much it depends on the build reaching cash flow rather than by its letter rating. Third, the week: on the same five sessions that bought every developed long end, the only AI bond with a public spread history got sold to its wides. The senior tranche was repriced with the junior because the market has stopped believing the rating and started pricing the timing.

That is Slok’s zero, priced. If the 493 are not paying for the output, the hyperscalers’ capex is the only demand, and the hyperscalers’ willingness to keep paying is the only thing standing between an A+ rating and a 92 handle. Gray’s parabola is what the market is lending against. Beignet is what the market thinks the parabola is worth.

———————————————————————————

Richard Clarida put the Fed’s problem in one chart Friday. PCE inflation has been above 2% for six years and is back above 3%. The San Francisco Fed’s decomposition splits it into demand-driven, supply-driven and ambiguous. Since 2024 the demand contribution has been rising while the supply contribution, tariffs and wages, has been falling. If the overshoot is supply, there is a case for looking through it; in March no FOMC participant projected a hike this year and twelve projected at least one cut. If the overshoot is demand, “strong technology investment, resilient consumption, and rising equity prices” in his words, then September’s hike and the tightening bias make sense.

The Dispatch read: the AI build is itself demand. Hundreds of billions a year of capital expenditure into power, construction, equipment and labor, all of it spent before any of Slok’s productivity payoff arrives. The build adds to aggregate demand now and adds to supply later, if at all. In the meantime the Fed tightens into the demand the build creates, which raises the cost of capital for the build, which hits the junior tranche first and, as of Thursday, the senior tranche too. The build is feeding the inflation that raises its own funding costs.

Oil closes the loop from the other side. Brent at a six-month high is a supply shock in Clarida’s framework, the kind the Fed can look through. But Christian Scherrmann’s point is that oil has been too high for too long, and the only thing stopping companies from passing it on has been moderate wages. Rick Rieder’s is that the structural pressures, energy, insurance, healthcare, education, do not respond to rates at all. A Fed facing demand-driven inflation from the build, supply-driven inflation from the Gulf, and a residual it cannot classify will not cut into a parabola. It will hold with a bias, and the long end will keep pricing the bias.

———————————————————————————

Wednesday’s meeting is a hold at 3.75–4.00%. The question is the shape of it. Gregory Daco expects the “troika,” Warsh, Jefferson and Williams, to show no urgency, at least two dissents for a 25 basis point hike from Hammack and Logan, and a statement tweak from September’s “timelier return” to a “timely return” of inflation to 2%: a hawkish hold with no forward guidance. The press conference is the risk. Warsh cannot say the Fed wants more time to assess the data without undermining September’s move, and he cannot sound political a week before the midterms. Tiffany Wilding has the same hold with a clear tightening bias and one or two more hikes to come, reversing last year’s cuts; she sees the labor market as looser than the unemployment rate suggests, which gives the Fed room to wait. Scherrmann has December as “highly likely.”

The tape goes in with twos at 4.692, six inside the 4.75 dial, after Thursday’s flight took them to 4.646. Nothing is priced for Wednesday. Two dissents and a tightened statement reprice the front end back toward the dial and then toward the 4.85 reversion line; a Warsh presser that leans on patience does the opposite and extends the long-end bid. Either way the long end’s path is set by the thing the flight was hedging. Thursday brings Q3 GDP and September PCE, the first look at whether the build’s demand shows up in the quarter.

———————————————————————————

Moody’s kept France at Aa3 Friday night with the outlook negative, one notch above S&P and Fitch at A+. The hold assumes a deficit-reduction budget eventually passes; the agency said political instability will keep constraining consolidation and that the real test comes after the April–May 2027 presidential election. The deficit runs about 5.4% of GDP this year, debt near 120%. The National Assembly has already rejected the revenue side of the 2027 social-security budget. S&P reviews November 27.

The market had priced worse. OAT–Bund closed 169 Friday, in from 178 at Thursday’s wide, and the French 10-year at 5.037 is still above the US 10-year at 5.001, which it crossed Thursday for the first time since 2002. Exhibit 2 shows the week: every developed long end was bought except France’s, which rose 25 basis points while Bunds fell seven. That is the European version of the sort in Exhibit 1. Germany is the senior tranche; France is being priced as the junior one, and a rating hold does not change which side of the line it is on.

———————————————————————————

Thursday’s flight took the 10-year to 4.961; Friday gave back four and closed it at 5.001. The giveback was uniform across the curve, about four basis points at every tenor, which is what a consolidation looks like rather than a reversal. On the week the long end is ten lower, the front end two, and 5s30s is inside 51 from 55. Japan’s 10-year fell 14, Bunds 7, gilts rose 2. The Six Tracker moved the right way on all four.

The rule set two Sundays ago for the Term Premium dial was two weekly closes under 5.40 with HY inside 325, then amended last week to a Friday close under 5.35 with the giveback stalled. The level condition fired: 5.323. The dial stays red anyway, below, because the bid that got it there was a flight into the senior tranche against the junior tranche being marked, with crude up 6.5% on the week and a hawkish hold three sessions out. The long end has rallied 25 basis points from its high on a hedge. It has not yet rallied on supply, and the refunding on November 4 is the first day it can.

———————————————————————————

No public S-1 on EDGAR as of Sunday. The mid-November listing that has been reported requires the public filing by late October under the fifteen-day rule, so this week is the window. The draft prospectus figures as viewed by the Journal stand: first-half revenue $16.3 billion, third-quarter $16.8 billion, up 45% on the quarter, a first-half net loss of $80.7 billion of which about $72 billion was non-cash, and the $450 billion of noncancelable commitments above. Gray’s path for it is the one to watch: IPO, market cap, investment-grade rating, lower cost of capital, less reliance on the hyperscalers. The Dispatch question for the rating agencies is how $450 billion of noncancelable compute reads next to $67 billion of annualized revenue. Beignet suggests the market will not wait for the agencies to answer.

The capital-stack model for the filing, which reads the commitments as debt-like obligations for credit and as a required growth path for equity, is built and goes into the Vault with the S-1.

———————————————————————————

———————————————————————————

Front End / Policy: GROWING, held. Twos 4.692, six inside the dial, with September’s CPI at 3.7% and 2.6% core and crude at a six-month high. The hawkish hold is the base case and the dissents are the upside risk. Reverts to PRESSURE on a close through 4.85.

Term Premium / Long End: PRESSURE, held. The step-down rule fired on level: 5.323, under 5.35, with the giveback stalled. It stays red because the rally is a hedge against the junior tranche, not a change in the supply the refunding has to clear, and because a hawkish Fed and $117 Brent are three sessions out. Re-scored Thursday after the FOMC and PCE. Steps down on a Thursday close under 5.35 with twos inside the dial through the presser.

Credit: PRESSURE. CCC 1340, a fourth close through 1300 and the widest. HY 340, ten from 350. Beignet 241 over. The 350 print is the next score change.

Inflation, Growth / Labor, Cross-Asset: PRESSURE. Brent +6.5% on the week. Payrolls slowing, staples and insurance sold Friday. Every FI ETF back inside ±1.2 z after Thursday’s +3 z day, which is a consolidation, not a reversal.

Composite: PRESSURE. Reverts to ELEVATED on CCC inside 1125 and EMB inside plus or minus one standard deviation for two sessions. Neither is close.

———————————————————————————

Gray and Slok are describing the same build from opposite ends of the pipe. Demand is parabolic because it is contracted. Payoff is zero because it is early and concentrated. What sits between those two is time, and time financed with long fixed obligations is what credit markets price. This week they priced it at 241 over on an A+ bond while they bought every long end but France’s. The Fed meets Wednesday facing inflation the build helps create and oil the build cannot control, and it will hold with a bias. Watch three numbers into it: twos against 4.75, HY against 350, and Beignet against the A index. The first tells you what the Fed did to the senior tranche. The other two tell you what it did to the junior one.

Earlier this week: The Crossing (Oct 23) · Paris at Five (Oct 22) · The Other Junior Tranche (Oct 21) · Three Basis Points (Oct 20) · The Senior Bid (Oct 18, last Sunday’s Set-Up)

The Dispatch runs every trading morning at dispatch.thebondbro.com. The Vault takes the same tape a step further: the analysis behind the levels, and now the Toolkit, three Excel calculators that run on your own holdings, brackets and cash.

———————————————————————————

Sources: Koyfin global yields, yield curves, corporate credit OAS, FX, FI ETF, US sector, factor, country and world-economics dashboards, Friday Oct 23, 2026 close (exported Sunday Oct 25, 02:04 ET); Treasury one-day changes are Dispatch calculations against Thursday’s close as recorded in the Oct 23 edition, one-week against the Oct 16 close as recorded in the Oct 18 Set-Up; credit one-day changes likewise, and approximate. CME crude settles Oct 23 via FinanceFeeds. Freddie Mac Primary Mortgage Market Survey, Oct 22, 2026 (30-year 7.32%). BLS CPI, September 2026 (3.7% headline, 2.6% core). Jon Gray, Bloomberg Television, Oct 22. Torsten Slok, Apollo Daily Spark, “No Signs of AI Payoff Outside the Tech Sector,” Oct 24. Skyler Weinand, Regan Capital, Bloomberg chart of Beignet (Meta Hyperion / Blue Owl JV 6.58% 2049), Oct 22; structure terms as reported. Bank of America, Citigroup, UnitedHealth and JPMorgan AI figures as reported in third-quarter earnings and public remarks; approximate. Richard Clarida, LinkedIn, Oct 23, citing the San Francisco Fed inflation decomposition. Gregory Daco (EY-Parthenon), Tiffany Wilding (PIMCO), Christian Scherrmann (DWS), Rick Rieder (BlackRock), LinkedIn, Oct 21–24. Moody’s on France, Oct 23, via CNBC and others. Anthropic draft-prospectus figures as reported by the Wall Street Journal, Financial Times and Benzinga; S-1 status per EDGAR, Oct 25. Prior Dispatch editions as linked.

The Bond Bro Dispatch is published by Positive Carry LLC. All content is general market commentary provided for informational and educational purposes only and does not constitute investment advice, a recommendation, an offer, or a solicitation to buy or sell any security. Nothing herein is tailored to the circumstances of any recipient. Data are drawn from sources believed reliable; accuracy and completeness are not guaranteed. [email protected] · © 2026 Positive Carry LLC. All rights reserved.

Produced with AI assistance. All data selection, analysis, conclusions, and final editorial judgment are the author’s. All content is reviewed and approved by Positive Carry LLC prior to publication.