1The September hike bet came off nine points in a day, with no new data behind it
Polymarket's September Federal Reserve contract now prices a 25bp increase at 50.5%, against a 47.5% chance of no change. That is a fall of ten points in twenty-four hours, on a market carrying $4.88m of liquidity and $3.0m of volume in the same day — deep enough that the move is not one participant's opinion. A day earlier the same contract read 59.5%.
The direction of the week runs the other way. Over seven days the increase leg is up twenty points and the no-change leg down twenty. So the last day did not reverse the trend; it gave back roughly half of what the week had built. The 2026 contract, which asks only whether a hike lands at any meeting this year, is steadier at 70.5%, down 1.5 points on the day across $8.44m of lifetime volume. Ed Elson, on Prof G Markets, put the equivalent Kalshi number at 77% as of Tuesday's close.
Nothing printed in between. The only macroeconomic release inside the window was the ADP employment report on 2 September, and ADP is not showing a break: its employment level is up 763,000 year over year in August, a 0.58% pace that is essentially identical to June's 0.61% and July's 0.59%. Reuters reported stocks climbing and US yields easing on Tuesday evening, then shares and bonds rallying overnight while "markets await Fed signals". The repricing is positioning, not information.
That matters because of what sits on the other side of it. The July payroll report showed employment falling 23,000, and core PCE inflation running at 3.34% year over year with headline CPI at 3.3%. Those two numbers pull in opposite directions for a central bank, and the market has spent a week deciding the inflation number wins, then spent a day partially un-deciding. A 50.5/47.5 split is not a view. It is the absence of one, held twelve hours before the August employment report.
2The oil premium is de-pricing faster than the war is de-escalating
Polymarket's September WTI ladder has $90 already resolved yes. Above that, the $100 leg is at 25%, down 12.5 points on the day — though still 13.5 points higher than a week ago — and the $95 leg at 55.5%, down eight. Downside is firming in the mirror image: a September low of $85 now prices at 73%, up 6.5 points. The book carries $786k of liquidity and $202k of volume in the day, which is thin next to the Fed contracts and should be weighted accordingly.
The de-pricing is not matched by de-escalation. Ian Bremmer, of Eurasia Group, told Prof G Markets his baseline is that the conflict runs "certainly through the midterms" and beyond, because neither side accepts the other's account of who holds leverage. He was explicit that this does not become Iraq or Afghanistan — the US is unwilling to put troops on the ground or to strike Iranian critical infrastructure, and Iran's leverage over the Strait is, in his phrase, "a wasting asset" as Gulf states build defences and export routes around it. Polymarket agrees on the tail: a US invasion before 2027 prices at 15.5%, down a point on the day, on $64.3m of lifetime volume.
Two numbers for Hormuz throughput are in the record this week, and they cannot both be right. Reuters reported the US energy secretary saying 17 million barrels transited the Strait on Monday — roughly a normal day. Bremmer, recording for Tuesday, said five to seven million barrels are getting through, "a lot less than was before the war, but more than a couple months ago." The gap is not a rounding error; it is the difference between a blockade that is biting and one that is mostly rhetorical. What would settle it is whether the 17m figure counts crude only or all cargoes including refined products and LNG, and whether Monday was a one-day catch-up after the weekend strikes rather than a run rate.
Bremmer added two supply details that cut against the softening premium: US stockpiles are at their lowest in decades, diesel refining capacity is constrained with that price at record highs, and Chinese crude imports have ticked back up from five million barrels a day to seven. He read the administration's Venezuela arrangement — an American ownership stake in Venezuelan fields — as a response to exactly that pressure. If the 17m figure is the run rate, the barrels are moving and what has come out of the price is a headline premium. If Bremmer's five to seven is closer, the current price is understating the shortfall.
3Whether 5% long yields are a crisis or a repricing depends on growth, and growth is the number that has not printed yet
Japan's 30-year government bond yield reached 4.18%, the highest in the instrument's history, and Felix Breen of Felix & Friends led with it: the "widow maker" trade that destroyed three decades of short sellers finally paid, and Japanese equities lost more than $200bn in a day, led by technology. He reads Japan as roughly ten years ahead of the United States on the same road — debt accumulates, the central bank becomes the buyer of last resort, and the currency pays the bill. UK gilt yields hit a fresh 19-year high on Tuesday, per Reuters.
The US long end is in the same move. Ed Elson described the bond market's worst stretch since 2006, with the 30-year "back near 5.3%, a level it hasn't held this consistently in nearly two decades". The official series lags the move: the last Federal Reserve print for the 30-year is 5.22% on 28 August, and the 10-year 4.79% on 1 September, up from 4.73% five sessions earlier. The 10s-2s spread sits at 0.40, barely changed.
Matthew Klein of The Overshoot reads the identical facts as benign, and Ben Carlson and Michael Batnick put his argument on Animal Spirits without much resistance. Incomes and spending are rising about 7% a year in dollar terms; against that, a 10-year yielding under 5% is "the opposite of a disorderly market". Klein's condition is precise: today's rates are obviously too high only if inflation and growth are both about to slow sharply. Batnick's supporting note was that real GDP per American rose less between 2007 and 2019 than it did between 1929 and 1941 — the decade people are nostalgic for was the slow one.
So the two readings are not really about bonds. Breen's case needs nominal growth to crack while the debt compounds; Klein's needs it to hold. The evidence available today splits: inflation is not slowing, with CPI at 3.3% and core PCE at 3.34%; payrolls fell 23,000 in July while the unemployment rate improved to 4.1%; ADP shows steady 0.58% year-over-year growth. Housing is the one clearly contracting piece, with the 30-year mortgage at 6.66% on Freddie Mac's last weekly print and Animal Spirits citing a daily read of 6.8%. Friday's employment report is the first hard test of whose condition holds.
4The AI build was financed on a rate assumption that has since inverted
Breen's number for the borrowing is $410bn raised this year by companies building AI capacity, taken on in the expectation that rates would fall. Robin Wigglesworth's New York Times piece, read out on Animal Spirits, sizes the same trend from the other end: AI companies tapped debt markets for $27bn last year, close to $450bn by mid-August, and roughly $600bn for 2026 as a whole, with planned capital expenditure running near 3% of GDP from 2027 to 2029 — about a trillion dollars a year. Barron's ran the capex question as a cover story in the same week. Rates went up.
The financing is also getting harder to read as arm's-length. Elson walked through Anthropic's $35bn compute agreement with Lambda: is among Lambda's largest investors and its largest GPU supplier, the data centre comes from and is leased by Nvidia rather than Lambda, and Nvidia is also an investor in Anthropic. His conclusion was that the deal says nothing about underlying demand in either direction — it establishes that Nvidia has money to deploy, and little else.
The counter-evidence is in the same episode, and it is not weak. Net margin for the 493 companies in the S&P 500 outside the largest technology names has gone from below 11% in the fourth quarter of 2023 to 13.6%, and is forecast near 15% by the end of 2027. Rich Bernstein's screen of global companies with projected earnings growth above 25% returns roughly 200 names, of which exactly one is a Mag 7 constituent and many are listed outside the United States; his own note says the rotation away from the momentum-driven AI theme has happened during 2026 without a bear market. Alex Tabarrok's read of census data — 2.3% of firms raised employment because of AI, 2% cut it, 96% reported no change — says the labour-market disruption is still a rounding error four years in.
Torsten Slok's framing, also quoted on the episode, is that bonds win either way: if AI succeeds, the deflation lowers rates; if it fails, the crash sends money into Treasuries. That is elegant, and this year has embarrassed it. AI has been succeeding, on revenue at least — Federal Reserve chair Kevin Warsh told Jackson Hole that annualised token sales at the two leading labs alone exceed $100bn, up more than 500% year over year — and long rates rose anyway. A framework that says rates fall in both branches has no branch left for what actually happened.
Software is the cautionary case for how fast the market changes its mind about all of this. , the software ETF, fell around 40% from its highs with individual names down 50-60%, and now sits less than 7% below them. Batnick, who traded it, described a slow fall, a crash, a rebound, new lows, a 30% rally in a fortnight, another crash, then new recent highs — and named , , , , and as names whose recoveries looked nothing like the index's. His point was not that software is cheap. It was that "software is dead forever", which is how he characterised the mood nine months ago, is already unrecognisable as a description of the sector.
What the sources recommended
Two single-name buys came out of five videos, both from the same person, both in precious-metals miners, and both were placed on a technical trigger rather than a catalyst. Ross Givens bought , Kinross Gold, and , Hecla Mining, on-camera during the episode, sizing 500 and 700 shares respectively with stops about 7% below entry.
His entry signal was breadth, not fundamentals: the share of Nasdaq composite constituents trading above their five-day moving average had fallen to the 20% line he treats as short-term oversold, a level he says last printed in late March. He was explicit that it is "not a guarantee" and used the word "indicator" with a disclaimer attached. The fundamental case he layered on top was operating leverage — Kinross at roughly 10 times forward earnings with an all-in sustaining cost near $1,800 an ounce and $520m of buybacks year to date against a commitment to return 40% of free cash flow; Hecla carrying no debt at all, with a silver all-in sustaining cost of $6.70 an ounce against a spot price he quoted at $65.
He named as the direct way to hold gold and then explicitly did not buy it, preferring the miners for the leverage. Treat that as a comparison, not a call.
No sell or avoid recommendations came out of any source this run. The closest was Bryce and Ren on Equity Mates walking through Star Entertainment's result — a A$307m loss, A$267m of cash left, and a possible A$400m regulatory fine against a company that told the Federal Court a fine above A$100m would make it difficult to keep operating — but they framed it as a going-concern question, not a position to exit.
Givens promoted a $5-a-year paid service twice inside an eleven-minute video, and Equity Mates is a subsidiary of BetaShares, whose products dominate the listener portfolio reviewed in the episode and who sponsor the segment. Both are disclosed on air; both are worth knowing when weighing the calls above.
Friday's payroll report settles more of the September question than this week's repricing did
Weekly jobless claims land this morning. The August employment report lands tomorrow, 4 September, and it is the last major labour reading before the Federal Open Market Committee meets on 15-16 September. A market split 50.5/47.5 on a hike has, by construction, no cushion for a surprise in either direction.
The CPI release follows on 11 September, four days before the meeting. In corporate news, , , , , and report after today's close, with , and before it — a consumer-plus-software cluster that will read directly on the two arguments above: whether nominal spending is holding up, and whether the software recovery has substance behind it.
