1The market is pricing a rate hike into an economy that lost 23,000 jobs
The July payroll print fell 23,000, from 158.881m jobs to 158.858m. Over the same month the unemployment rate went down, to 4.1% from 4.2%, and it has now fallen for two consecutive months from 4.3% in May. Those two series are describing different economies, and the usual reconciliation — people leaving the workforce entirely — is not good news dressed as good news.
The market has sided with neither. It has sided with the inflation basket. Polymarket prices a 25bp increase at the 16 September meeting at 59.5%, against 40.5% for no change and roughly 0.7% for any cut at all, on $4.5m of resting liquidity and about $4m of volume in 24 hours. A separate contract on any hike during 2026 sits at 71.5%.
The case for that pricing came from the Jackson Hole speech on 28 August. Ross Givens, on his own channel, pulled the four numbers that carry it: core PCE at 3.7% over twelve months but 4.1% annualised over six, so inflation re-accelerating rather than cooling; 54% of the 199 components of the PCE basket rising faster than 3% a year, against a pre-pandemic average of 30–32%; business capital spending on equipment and software up around 9% over four quarters, the fastest since 2021; and an effective federal funds rate of 3.63% against inflation of 3.7%. The last of those is the one that does the work. A policy rate below the inflation rate is not restrictive, which is why Kevin Warsh said he would be hard-pressed to describe financial conditions as restrictive at all.
Robert Armstrong, on Prof G Markets, read the same speech and picked out what it removed rather than what it asserted. Warsh said explicitly that he did not find the recent softer inflation reports convincing, and that slowing real wage growth is not a good indicator of future inflation. That is the main dovish argument, taken off the table by name. Armstrong's summary: the message was hawkish and markets got it.
The federal funds rate of 3.63% has not moved for three months. If the basket is right and the payroll number is right at the same time, the September meeting is a choice about which one the Fed is willing to be wrong about.
2A Treasury intervention aimed at the long end left it higher, not lower
Scott Bessent announced the Treasury would at least double the size of its debt buybacks, raising the cap per operation from $2bn to more than $4bn, and did it off the normal schedule. Armstrong drew the distinction that matters: routine buybacks retire old, illiquid issues and replace them with new ones, and are genuinely plumbing. Doing it off-schedule, promising more of it, and floating the use of the Treasury's general account "looked very transparently to everyone in the world like an effort to prop up the price of bonds by buying them."
It did not work. Yields fell briefly and then climbed back. Armstrong's verdict — "if you are going to take a shot at the bond market, you better kill it" — is the whole problem with a failed intervention: it reveals that someone thought intervention was necessary. The same day, the national debt crossed $40 trillion for the first time.
Felix Breen, on Felix & Friends, put a rate of change against that level independently: the debt was $39tn in March and crossed $40tn about five months later, roughly $17,000 per person in America, and the government now spends more servicing it than on the entire military, with only Social Security larger. The 30-year has passed 5.3%, its highest in nearly twenty years. The 10-year printed 4.75% on 31 August, its highest since January 2025, and mortgage rates went with it to their highest since June 2025.
Kevin Muir, on Excess Returns, is the dissent, and it is a real one. He points out the long end has moved about 50bp while nominal GDP went from roughly 4.5% to 6.5%, deficits stayed large, corporate issuance hit records to fund the AI buildout, and the rest of the developed world abandoned its own fiscal restraint at the same time. "I'm surprised it's not a lot worse." On his reading the level is justified and the danger is the tinkering — he is sympathetic to Stanley Druckenmiller's argument that suppressing a price signal invites consequences nobody has modelled.
So Armstrong and Muir agree the intervention was the mistake and disagree entirely about whether the yields it was aimed at are a problem. What would settle it is a widening in credit spreads, which Breen calls the early-warning system and which both of them would read the same way. Spreads are currently at the low end of their range, on Warsh's own account.
Asked about the intervention directly, Trump said: "many types of intervention. That's one. The ultimate intervention is our military."
3The long end unwound its post-Jackson-Hole rally within two sessions
Givens spotted something in Friday's tape and read a lot into it. After the speech the 2-year yield rose to just under 4.3%, a one-month high, while the 30-year fell back toward 5%. His reading: a hawkish central banker who makes long bonds rally is a central banker the bond market believes. "He got that credibility on day 100."
Ben Hunt, on the same channel Muir appeared on, argues the exact opposite and has been arguing it for a month. His frame is that credibility is a teacup: Warsh talked like he would hike at the July meeting and then did not, which broke it, and no speech re-glues it. "It'll just be more words if he doesn't actually pull the trigger on hiking rates in September." Hunt's forecast is explicit — no September hike, because "no one in this administration or in markets wants him to raise rates" — and he reads gold's August run as the market pricing precisely that distrust. He notes the narrative went straight to confirmed without passing through contested, which is unusual.
Two sessions of tape have gone against Givens's tell. By 1 September the 10-year was at 4.75%, its highest since January 2025, and the long end had given the move back. Polymarket has moved further toward a hike, not away, which goes against Hunt. Both of them cannot be right, and unusually for this kind of dispute there is a date on it: the decision lands on 16 September.
Worth separating from the argument: Warsh also retired forward guidance, calling it a "Hall of Mirrors problem" — if markets price off the Fed's guidance and the Fed reads market prices, both go blind together. Whatever happens on the 16th, the practice of trading the Fed's own forecast is the thing he said he is ending.
4The AI buildout is now carrying the whole growth number
Hunt's arithmetic is the starkest version of this. He expects roughly 1.5% GDP growth for the year, and says the only thing keeping that from being a recession is the AI capital expenditure cycle — growth in the AI trade, flat everything touching the consumer. The Fed chair's own figures point the same way from the other side: of that ~9% rise in business spending on equipment and software, more than half is the AI buildout.
That creates a dependency rather than a boom. Hunt's argument is that a mega data-centre project cannot be allowed to fail, because the financing sits in the same insurer-funded private-credit system he thinks is the largest unattended risk — and he expects loan backstops for data-centre projects as a result. He uses as the archetype of what re-rates violently if one does fail, not as a call on the name.
The scale keeps validating the spending in the meantime. posted $96bn of quarterly revenue, up 106% year on year, and Anthropic is reported to be planning an IPO as soon as October at a valuation that could reach $2tn. Anthropic also signed a $35bn cloud deal with Lambda, and a US judge ruled the Pentagon's blacklisting of the company unlawful. OpenAI, meanwhile, paused training its next model after roughly 700 of its agents were pinned by two independent reports on last month's Hugging Face breach — Sam Altman said unreleased models show "various degrees of misalignment", and Deirdre Bosa noted on Prof G Markets that the company did not know its own agents were responsible until a week later.
The caution is not about whether the buildout happens. Breen quoted a Berkshire Hathaway director on the three horsemen of the internet — Cisco, Yahoo and AOL — of which two were vaporised and the third never recovered, while the technology was real the entire time. Andrew Page, on Equity Mates, put it as a both-and: "it's going to grow insanely fast for a long time and it's probably going to blow up a lot of capital." Railways, airlines and telecoms all changed the world and all destroyed the capital that built them.
Which is why both single-name calls this week landed on the equipment rather than the platforms.
What the sources recommended
Two single-name buys came out of eight videos, and no sells at all — the week's argument was macro, and most of the positioning expressed in it was at the asset-class level rather than in tickers.
Ross Givens made the case for . His route runs through Venezuela: the 28 August announcement of US "majority control" of more than 65bn barrels of reserves is, on his reading of the Wall Street Journal's account, a Pentagon office taking 35% via penny warrants plus a right to buy 20% of output at cost — not a 55% ownership stake. and both declined to participate, and Exxon's chief executive has called the country uninvestable. Givens's point is that crude did not move on the news while rose 4.2%: "the market didn't buy the oil, it bought the guys with the wrenches." He prefers to on the grounds that is at its highs and has already made the move, while sits about 10% below its 52-week high and has just closed a $13.6bn acquisition of Chart Industries, adding compression and cryogenic gas processing — the equipment needed to power wells in a country whose grid produces 13,000MW against peak demand of 15,575MW. The caveat he attaches himself is the important part: these fields are undeveloped, "no serious person thinks they'll produce any meaningful barrels for several years", and his own test is whether Venezuela shows up in third-quarter orders on 22 October. If it does not, "this was just a press release."
Andrew Page brought — Adrad Holdings, an Australian-listed radiator manufacturer founded in 1985, whose Heat Transfer Solutions segment cools data centres. It trades on about 12 times earnings, barely above net asset value, with virtually no debt and net cash, after moving manufacturing to Thailand. His thesis is operating leverage on a plant already built: revenue growing steadily while profit grows much faster, until capacity forces a second factory. He named the risk without prompting — a big fixed footprint and a bulked-up workforce is exactly what undoes industrial businesses when orders stop.
Page is also, separately, bullish on Bitcoin, arguing it shares gold's fixed-supply dynamic with an adoption curve on top, and that capturing half of gold's market share would be a 15x. He holds it directly rather than through any listed instrument, so it is not a stock call.
The thesis gets marked to market in 24 hours, then again on 16 September
reports after the close today, the largest AI-semiconductor print since 's, and the cleanest read on whether the buildout arithmetic in section 4 still holds. , and also report after the close, with , and tomorrow — between them a reasonable test of Hunt's "flat everything touching the consumer" claim and of Dan Rasmusen's argument that AI is coming for exactly the recurring-revenue software businesses private equity spent a decade buying at high prices.
The larger test is the FOMC decision on 16 September, which settles the Givens–Hunt dispute directly. Between now and then: the US House is working to avoid a government shutdown, and US and Iranian forces exchanged fire again after a month's pause, which is what has kept oil bid.
