1Oil settling above $100 turned a geopolitical story into an inflation story
Brent settled above $101 a barrel, its first close over $100 since July, as US-Iran fighting widened beyond the two combatants. Reuters reported Houthi attacks on southern Saudi cities for a second consecutive day, an oil tanker struck in Iraqi waters, and roughly one third of Gulf crude still unaccounted for despite so-called dark crossings. This is a supply story with a widening blast radius, not a headline risk premium.
US equities took it as an inflation signal. The S&P 500 closed at 7,636.36, down 0.48%; the Dow fell 405 points to 52,380.66, down 0.77%; the Nasdaq lost 0.64% and the Russell 2000 fell 1.32%, the widest loss of the four. Small caps leading the decline is the tell — they carry more floating-rate debt and less pricing power, so a crude shock that lifts input costs and yields at once hits them hardest. That makes the move a read-through to rate-sensitive domestic earnings, not an energy-sector rotation.
The prediction markets moved with the news rather than against it. Polymarket's US-Iran effective-ceasefire event drew $605,000 of 24-hour volume against $90,800 a day earlier, a near sevenfold jump, with total liquidity rising from $144,500 to $221,000. The contract on a ceasefire beginning by 11 September trades at 49%, on $48,500 of liquidity in that leg; the 30 September leg sits at 68%. The de-rating is steady rather than sudden: the same contract was quoted at 70.5% on 7 September, 60% in the early hours of 9 September, and 49% now. Near-term peace has lost more than twenty points of implied probability across three readings, and each step came with fresh escalation news rather than drift.
President Trump said the Iran war will end right after the US midterm elections, which sets a political horizon of early November rather than a military one. Against that, Mike Moffatt, speaking on Prof G Markets, noted Western Canadian Select has gone from about $50 a barrel before the Iran action to about $80 — the escalation is already paying producers outside the conflict zone.
The consequence: an escalation premium is now inside the crude price rather than hovering above it, and the ceasefire market has stopped treating near-term de-escalation as the base case. Only a genuine settlement returns barrels; a pause merely stops the premium compounding.
2The September Fed meeting is priced as a hike, not a cut
The most liquid contract on the question has moved decisively. Polymarket's Fed Decision in September event, resolving 16 September, carries $4.5m of liquidity and $4.2m of 24-hour volume — deep enough to take seriously. It prices a 25 basis point increase at 52.5%, down two points on the day; no change at 45.5%; a 25 basis point cut at 0.35%; and a larger increase at 0.8%. A cut is not merely disfavoured, it is priced out. The live argument is hold versus hike.
The hard data supports the hawkish read. Federal funds stood at 3.63% in August, unchanged since June, per the St. Louis Fed. Unemployment printed 4.1% in August, down from 4.2% in June, and payrolls rose to 159.08 million from 158.91 million in July — about 162,000 jobs added. The 10-year Treasury yield reached 4.80% on 8 September, up from 4.78% on 4 September, while the 10-year-minus-2-year spread narrowed to 0.40 from 0.41. A labour market that is firming while crude breaks $100 gives a committee inclined to tighten the cover to do it.
Brian, on Business with Brian, reports that Fed chair Kevin Warsh used Jackson Hole to warn that he might have to raise rates rather than cut them, and cites a figure of 54% of goods and services in the basket showing price increases above 3% over the past twelve months. Felix Prehn, on Felix & Friends, describes the same repricing from the other end: the bond market has stopped believing the new chair will cut, and is pricing rates higher instead.
The disagreement worth keeping is not between these two — it is inside the government. Both describe Washington easing and tightening simultaneously. The Treasury is buying back its own long-dated debt while the Fed signals higher rates; Prehn calls it one foot on the accelerator and one on the brake, and Brian frames the resulting yield as the price of not knowing which wins. Neither can say which force dominates, and neither pretends to. Nothing in today's evidence settles it.
Polymarket's dissent market suggests the decision will be contested whichever way it lands: three dissents and four-or-more dissents are each priced at 30.5%, against 10.5% for a unanimous vote. That contract holds only $20,800 of liquidity, so treat it as a weak signal rather than a forecast.
The consequence: for the 16 September meeting the question has narrowed from how fast to ease to whether to tighten, and the committee is expected to split either way.
3The long end is repricing for reasons the Fed does not control
Three separate lines of evidence converged on the same point today, and none of them runs through the policy rate.
The first is supply from an unexpected competitor. A chart shown on The Compound's Animal Spirits put hyperscaler and debt issuance at roughly 70% of Treasury bond issuance — AI capex is now bidding against the US government for the same pool of lenders. Michael Batnick and Ben Carlson pair it with a Wall Street Journal chart showing data-centre construction rising since December 2023 while all other private construction falls away, and argue that the macro explanation and the crowding-out explanation are both operating rather than competing. Their own caveat is worth carrying: strip out AI spending and, on their reading, the economy is already in recession, so the borrowing that lifts yields is also what is holding growth up.
The second is the buyer walking away. Brian reports Japan holds $1.1 trillion of Treasuries, more than Britain or China, and that the stack is down about $120 billion since February; Prehn puts a single month's selling at $88 billion and notes 70% of Japan's reserves sit in US debt. Japan's own 10-year crossed 3% for the first time since 1996, which is the mechanism — once domestic yields clear the currency-hedged return on Treasuries, the money stays home. The repricing is not confined to America: Britain's 30-year reached its highest since 1998 and Germany's its highest since 2011.
The third is the borrower's response. On 18 August the Treasury bought back 30-year bonds issued in February 2021 at 52 cents on the dollar — arithmetic, not distress, since a sub-2% coupon has to be discounted to compete with a new issue near 5.3%. Treasury Secretary Scott Bessent conceded on air that part of the exercise was signalling, and the size was about $2 billion against roughly $30 trillion of debt. Prehn reports the programme is being doubled and run for two months to 4 November. Brian's framing is that Washington has four levers to lower long rates — borrow shorter, have the Fed cut, have the Fed buy, or spend less — and is currently pulling none of them, where in 2023 two were available and the 30-year fell a full point in ten weeks.
The cost is already visible in corporate credit. Brian reports Google borrowed for 40 years at 5.75% in February and 6.5% in August — the same issuer, six months apart — with about half the increase coming from Treasury yields and half from lenders charging more for a $200 billion annual data-centre programme. That is the crowding-out story arriving as a spread, not a rate. On Prof G Markets, Ed Elson sizes the sovereign version: US annual interest payments are $1.25 trillion, roughly a fifth of federal revenue and more than defence spending, against total debt above $40 trillion, with the Congressional Budget Office projecting net interest costs double over the next decade.
Batnick and Carlson supply the counterweight, and it is the most interesting disagreement of the day. Citing Rob Anderson of Ned Davis Research, they note fewer than 5% of S&P 500 stocks now yield more than the 10-year Treasury, the fewest since May 2007, and that the index dividend yield is barely under 1%, the lowest on record. They read that as good news rather than a warning — investors are finally paid to own bonds — and Batnick says he is considering adding fixed income to a portfolio that has been all equities. But they explicitly refuse the stronger version: pushing back on a Bank of America chart circulating on social media, they call this the best entry point for bonds in twenty years and not a generational one, and note that 1959, the last comparable reading, preceded a decade of negative real bond returns because inflation was coming. Given claim 2, that caveat is the load-bearing half.
The consequence: a decision on 16 September moves the front end, but issuance volume and the identity of the marginal buyer are setting the 30-year. Those can pull in opposite directions, and today they did.
4The robotaxi lead has moved to China
On the operational numbers, Baidu's Apollo Go has passed Waymo. Bryce Leske and Alec Renehan, on Equity Mates, put 's service in 28 cities against Waymo's 14, with 23 million lifetime orders against Waymo's 20 million, 350 million kilometres driven autonomously and 240 million of those with no safety driver. The detail that matters most is geographic: Apollo Go operates in Dubai, Abu Dhabi, Hong Kong, London, Seoul, Switzerland and Kazakhstan, so this is not a protected home market producing flattering figures. Waymo still leads on intensity at about 4 million miles a week.
The American contest is meanwhile being fought with regulators rather than mileage. launched the Cyber Cab, a two-seater with no steering wheel, pedals or side mirrors, and the National Highway Traffic Safety Administration opened an investigation almost immediately — Tesla self-certified that the vehicle meets federal safety standards on the reasoning that rules written for steering wheels do not apply to a car without one. Amazon's Zoox took the opposite route, requesting and receiving an exemption capped at 2,500 vehicles a year. Tesla has driven under 400,000 cumulative miles across seven cities against Waymo's four million a week, so the gap being litigated is large. The launch itself was thin: Elon Musk did not appear, and a livestream two million people waited for on X never happened.
Leske and Renehan decline to call the outcome for , which owns no fleet and has hedged across Waymo, , , Wayve and Motional — either massively disrupted or a major beneficiary, and they say they cannot tell which. That refusal is the honest position and it is worth preserving.
The consequence: the US regulatory fight is about the form factor of the vehicle while the operational lead is being set abroad, and a favourable NHTSA outcome would resolve the former without touching the latter.
What gets tested in the next week
Three of today's claims have dated tests, and two of them land within 48 hours.
reports after the close on Wednesday 10 September, with consensus at $1.78 in earnings on $19.53 billion of revenue, and Yahoo Finance flagged an outsized expected move. It is the cleanest read available on whether AI infrastructure demand still justifies the borrowing described in claim 3 — the spending is the thing lifting yields, so a miss would pressure the capex narrative and the crowding-out mechanism at the same time. also reports after the close the same day, consensus $6.20.
Polymarket's ceasefire-by-11-September contract resolves on Friday, which marks claim 1 to market directly at 49%.
The Federal Open Market Committee decides on 16 September, and that settles claim 2 — hold or hike, with the dissent count as the secondary read.
