A week ago the Sunday piece said the AI trade is two trades, and the one that survives is compute and power. This week the reason showed up in three places at once: a factory contraction with crude at 103, a BOJ hike the yen ignored, and an IPO pushed to November for 'safety.' The thread through all three is an input cost. The AI race with China is not a race to build a smarter model. Capability converged months ago. It is a war over what it costs to run one, and the two inputs that matter are oil and electricity. The US can weaponize one of them. It just did. The other, China owns, and the Fed's own inflation number is built to look through it. That gap is what the long end is pricing, and this note puts a number and a date on it.
The Dispatch publishes every trading morning. Subscribe on LinkedIn or at dispatch.thebondbro.com.
THE BOARD
Friday closed where Friday opened: twos 4.75, tens 5.00, thirties 5.33. WTI slipped back under 100. The yen recovered a big figure to 156.88 without anyone in Tokyo lifting a finger. The tell on the Board is the z-score column: MBB −2.00, GNMA −2.66, SUB −2.02, TIP −1.94. Every duration proxy on the tape is two standard deviations cheap and the front end is two standard deviations rich. That isn't credit. That's a market that will own anything with a coupon as long as it matures before November.
THE SPINE
Capability is not the scarce input. China ships a frontier-adjacent model within months of every US release, and the gap is not growing. Whatever moat exists in model quality is measured in quarters. The moat that lasts is measured in gigawatts.
On gigawatts, the asymmetry runs the wrong way for the US. China consumed roughly 10.4 trillion kilowatt-hours in 2025, two and a half times the US, and added more generating capacity in one year than the entire US data-center pipeline needs through 2030. It has around thirty reactors under construction. The US has none of size, a four-to-five-year interconnection queue, and gas turbines sold out to 2029. Yuekai's chief economist put it in Caixin last month without hedging: the end of AI is computing power, and the end of computing power is electricity.
Oil is the one the US can hit, and it did. China imports eleven million barrels a day, roughly half of it through Hormuz. The US is a net exporter. A $100 handle on crude is a terms-of-trade tax on Chinese manufacturing and a windfall for the US producer complex. That's an input-cost war the US can afford and China can't. Whether the Iran posture was designed as one or just functions as one doesn't matter to the price.
So the two legs pull apart. The US wins on oil and loses on watts. The AI buildout needs watts. Which means the US is fighting an input-cost war on the leg that doesn't power the thing it's trying to win, and paying for it at the long end.
THE NUMBER THE FED WATCHES
Here is the raw input. PJM's capacity auction cleared at $29 per megawatt-day for 2024/25. It cleared at $270 for 2025/26, $329 for 2026/27, $333 for 2027/28, and $325 for 2028/29 — at the price cap three years running. Ten-fold in two auction cycles. The market monitor attributed 63% of the 2025/26 charge to data-center load. For 2028/29 it's $6.3 billion of $16.4 billion.
Here is what the household sees. Residential retail up about 5% this year to 18.2 cents, per EIA. Here is what the Fed sees: CPI electricity +3.8% year over year in August, down 0.2% on the month, rolling over. Goldman's Abecasis and Wei put the pass-through at +0.1 on core PCE in each of 2026 and 2027.
The gap between $325 at the cap and 3.8% and falling is the rate-case lag. Regulated rates are set 12–24 months after the capacity gets contracted. What's in the CPI today reflects 2024 decisions. The 2027/28 and 2028/29 charges have not touched a residential bill yet. And when they do, they don't round-trip — rate cases ratchet. This is the one energy component that behaves like core and gets stripped out as if it were gasoline.
Electricity is 2.5% of headline CPI and zero percent of core. The Fed targets core PCE, which weights utilities lighter still. The Fed's public model of inflation isn't wrong. It's built to look through a cost that has stopped being transitory.
WHAT WARSH WON'T SAY
Warsh's public position, the one that got him the chair, is that AI is disinflationary: productivity lowers the neutral rate and the Fed can ease into it. Then he hiked 25 into a factory contraction with zero guidance and one sentence — relative prices must not broaden. His stated model says AI cuts inflation. His action says the input costs of building AI arrive first. He didn't reconcile those on the podium, and he could have.
Three reasons he won't. Naming it makes the Fed a combatant — 'we're tightening because the power buildout and $100 crude are structural' reads in Beijing as the US pain threshold, dated. Naming it hands the White House the attack: a Fed hiking against the administration's two signature projects gets fired in public; a Fed hiking on relative prices is doing its job. And naming it moves the long end — the thirty-year at 5.32 is already pricing what he won't say, and saying it makes 6.5 happen faster.
The simpler read is also live: he's a headline hawk with crude at 103 and doesn't need a China theory. The way to tell them apart is October 28. If he hikes again with oil flat, it's the watts.
WHERE THAT PUTS THE LONG END
A thirty-year is expected average policy rate plus term premium. What's in 5.32 today: funds at 4.00, one more hike priced, then cuts back to a longer-run rate the Fed still publishes around 3.25, plus 60–80 of premium. A market that believes the hike is a cyclical mistake the Fed unwinds.
What the watts change, all additive. Structural inflation: Goldman's +0.1 on core is the retail pass-through of a wholesale input at a price cap; call the true run-rate +0.5 on headline when the 2027–29 rate cases land, and longer-run policy goes from 3.25 to 3.75–4.00. Neutral real rate: a capex boom — $500 billion a year in AI plus another leg in generation and transmission — raises r*; the last investment surge of that scale was the 1990s telecom build, when thirties averaged 6–7 on a 5.50 funds rate. Add 25–50. Term premium: the buyer who caps the long end today is a Japanese official holder that's a seller at 158 yen, and a Treasury that finances the buildout, the war, and the interest on both. Take premium from 70 to 120–150, the pre-2008 norm, not a crisis number.
Sum: 3.75–4.00 plus 0.25–0.50 plus 1.20–1.50 is 5.75–6.00 fair value, 6.25–6.50 if it overshoots, which long ends do. Gundlach's 6.5 is the full-pricing number with nobody standing in the way. Shape: 10s30s from 33 to 70–90. A watts regime is a steepener from the back, the opposite of what The Yen Didn't Buy It (https://dispatch.thebondbro.com/p/the-yen-didnt-buy-it) described last week.
Two clocks. The pricing clock is six to eight weeks and runs through three dates: October 28–29, when a second hike with oil flat confirms the read and puts thirties at 5.50; November 3–4, when the midterms hand the refunding to a Treasury that either shortens issuance, Miran-style, and caps the long end at 5.50–5.75, or holds coupon sizes and lets 5.75 print that week; and the yen at 160, which pulls all of it forward on any day it happens. The fundamental clock is 12–18 months, when Goldman's +6% prints into CPI and the longer-run dot has to move. Position for the first. The second is why you don't fade it.
What stops a 6-handle is not the Fed. A Fed hiking the front end into this flattens the curve until something breaks, then it's a bull steepener for the wrong reason. What stops it is November 4. That is the long-end test, and it's the reason the Vault's first file is the refunding playbook.
THE STATES ARE THE PACE
The precedent already exists: Georgia's PSC seats flipped in 2025 on power bills alone — the first election decided by a data-center rate case. There will be a dozen in November, PUC races and governors in Virginia, Ohio, Arizona, Georgia, and every one runs the same ad. The state response arrives in order of how fast it hits supply: large-load tariffs that make hyperscalers pay for their own interconnection and capacity, already live in Ohio, Virginia and Indiana; county-level moratoria and siting fights in the Loudoun-style markets; abatement rollbacks on data-center equipment.
Paced supply raises the price, not lowers it. Demand for compute doesn't change; less capacity built means the existing electrons clear higher for longer. The wedge issue makes the watts inflation stickier. And it's the Kill Switch by another door — the federal regulatory lag was the six-month version, state PUCs are the two-year version, and the GPU-backed paper that needs a 2027 energization date is the paper that eats the delay. Every large-load tariff turns a hyperscaler's power supply into a contracted revenue stream with a public-body counterparty. Project finance. Same desk.
NOVEMBER, FOR SAFETY
Anthropic moved the IPO from October to November and said safety. OpenAI used the same word to move to 2027. The regulatory lag from The Kill Switch (https://dispatch.thebondbro.com/p/dispatch-sunday-set-up-74cb) is now the disclosed reason for a capital-markets delay, twice in a month, and the $1 billion Accenture evaluator venture is the industry writing its own audit standard before someone else does.
The cynical read is the better one. November 4 is the single best pricing environment an equity duration asset can get: a Treasury-engineered rally in the discount rate, one week after a midterm removes the deal as a wedge target. A $100 billion revenue book does not leave that to chance. Two things it tells you. The deal is a duration asset and its bankers know it — a company that needs the long end capped to price is telling you its multiple is a function of the thirty-year. And the safety story and the cynical story produce the same date, which means the safety story is free. Watch the pricing date: the week of November 9 means it was the refunding.
The company that runs on the electrons prices the week the market prices the cost of them.
THE OTHER SIDE
The bull case is the inflow. Bloomberg counted record money into bonds this week and the short end as the popular bet on a Fed inflation win; SHV at a +2.33 z-score says the same thing. If the factory contraction is the first of several, the front end is 50 basis points too high and 4.75 twos are the trade of the quarter, the long end rallies on the growth scare, and the watts are a 2028 problem for a 2026 market.
Three problems. The inflows are going to bills, not duration — that's the z-score column. Growth scares in a supply-shock regime don't cut long yields; 2022 is the template. And the yen: a weaker yen at these levels is a forced seller of the long end, not a buyer. The bull case needs Tokyo to stay home and PJM to clear below the cap. Neither is on the calendar.
PRESSURE GAUGE
Scores carried into the week. Cross-Asset stays GROWING with the yen back at 156.88 — the 160 trigger didn't fire. Inflation stays PRESSURE and gets a new note: the electricity input at the PJM cap is the structural leg under oil. Credit stays CALM: the −2 z-scores on MBB, GNMA and SUB are duration, not spread; IG 15+ closed the week at 98, flat. Growth stays STABLE on one factory print; flash PMIs Wednesday are the second.
THE WEEK AHEAD
Monday. Goolsbee. Japan closed. Chinese loan prime rate.
Tuesday. Williams, Jefferson, Barkin — the first post-hike Fed speakers. Anyone who says 'October' out loud moves the two-year. Two-year auction.
Wednesday. Flash PMIs, the second read on whether the factory number was noise. Barr. Five-year auction.
Thursday. Xi's state visit — the Trump–Xi summit, with trade talks already running in New York. The war on watts, in the room, the same week the US sells seven-years. Claims. Williams, Hammack, and a full BOE slate after last week's gilt decision.
Friday. Durable goods, Atlanta Fed nowcast. Moody's on Italy and the EU. Two weeks to PCE.
Standing. USDJPY 160. WTI 100 — it closed under it Friday. The Six Tracker.
THE KICKER
The US is fighting an input-cost war with China on the leg it can win, and financing it at 5.32% on the thirty-year. The leg it can't win is the one that powers the thing it's fighting for. The Fed's number is built to look through it, the chair won't name it, the states are about to pace it, and the biggest IPO of the cycle just moved to the week the Treasury has to answer for it.
The Dispatch shows you the signal. The Vault shows you what it means.
EARLIER THIS WEEK
The Kill Switch · Sunday Set-Up — https://dispatch.thebondbro.com/p/dispatch-sunday-set-up-74cb The Hike Nobody Asked For · Tuesday — https://dispatch.thebondbro.com/p/dispatch-morning-brief-e2b4 One Hundred Basis Points Apart · Wednesday — https://dispatch.thebondbro.com/p/dispatch-morning-brief-6d3d October, Not December · Thursday — https://dispatch.thebondbro.com/p/october-not-december The Yen Didn't Buy It · Friday — https://dispatch.thebondbro.com/p/the-yen-didnt-buy-it
If this was useful, the daily lands every trading morning at dispatch.thebondbro.com. The Vault — the desk-level work underneath it — opens shortly. First file: the refunding playbook.
Sources: Koyfin, Friday Sept 18 close (pulled Sunday 6:18am ET). BLS CPI, August 2026. PJM Base Residual Auction results 2023/24–2028/29; PJM Independent Market Monitor. EIA Short-Term Energy Outlook (May 2026) via Utility Dive. Goldman Sachs Research, Abecasis and Wei (Feb 2026) via Fortune. Caixin Global, Luo Zhiheng (Aug 14, 2026). Bloomberg, Sept 19–20: Qatari premier on Iran; Qatar Energy on Hormuz; US–China trade talks ahead of Trump–Xi summit; AI risk and billion-dollar funds; short-term Treasuries as the Fed-inflation-win bet; IMF on record global debt. WSJ/NYT, Sept 18–19: Anthropic IPO to November; Accenture evaluator venture. Newsquawk calendar, Sept 21–25. Georgia PSC election results, November 2025.
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.
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.