1A September rate hike became the base case in a single session
The August employment report landed on Friday and moved the deepest contract we track across the 50% line. Polymarket's "Fed Decision in September" — $8.59m of liquidity, $96.1m of lifetime volume, roughly $9.2m traded across its legs in 24 hours — now prices a 25bp increase at 50.5% against no change at 48.5%. A session earlier those numbers were 41.5% and 58.5%. The hike leg gained nine points in a day and seven in a week.
The mechanism is straightforward. Payrolls rose 162,000 in August, to 159.075m from 158.913m in July, against a July gain of only 21,000 on the same vintage. Unemployment held at 4.1%, unchanged from July and down from 4.2% in June. A labour market adding jobs at that pace is not one that forces a cut, and it hands the argument to the Fed chair's Jackson Hole line that policy is not currently restrictive. The 2-year Treasury yield sat at 4.34% on 3 September, about 71bp above the 3.63% effective funds rate, and CNBC reported it rising to its highest since January 2025 on the print. The 10-year-minus-2-year spread narrowed to 41bp on 4 September from 43bp the day before — the front end sold off harder than the long end, which is the shape a hike is priced in, not a growth scare.
The wider Fed book moved with it. "Fed rate hike in 2026" is at 71.5%, up ten points on the day and five on the week, on $133k of liquidity against $8.74m of lifetime volume. "No Fed rate cuts in 2026" is at 93.2%, up four points. October, by contrast, barely moved: no change 67.5%, a hike 27.5%, up one point. So the market did not add a hike — it pulled the existing one forward into September.
This directly contradicts what the two economists on the Charles Payne Show said 24 hours earlier, and the disagreement is worth keeping rather than resolving. Greg Daco read Governor Waller's 3 September remarks alongside John Williams's and concluded the pair "puts the Fed very much close to a hold at the September meeting." Joseph Wang, publishing the clip on his own Fed Guy channel, called Waller's speech "pretty dovish" and surprising after a hawkish Jackson Hole, and set the condition explicitly: "unless we have a hotter than expected CPI next week, he's inclined to hold." Wang added a mechanical point in the doves' favour — the PCE calculation changes in a few weeks in how portfolio-management fees are treated, which pushes the Fed's own target measure down. Daco went further, saying he would "be very cautious if I were a Fed policy maker to tighten into this environment," which he reads as disinflationary from the labour-market and interest-rate side.
Both were recorded before the payroll number. Neither has been shown wrong yet — their condition was CPI, and CPI has not printed. What settles it is the 11 September release, two trading days before the 15–16 September meeting.
One caution on reading any of these numbers. The prediction market that priced August payrolls put the actual outcome at 6%. On $19k of total liquidity across six legs it had 0–50,000 jobs as the modal bucket at 43.5% and 150,000-or-more at 6.0%; the print was 162,000. Liquidity is not decoration on a probability. The September Fed book has 450 times that depth, which is why it earns the top of this page and the payroll book earned a warning.
2The oil market expects a ceasefire and a closed Strait at the same time
Two books that ought to move together moved in opposite directions this week, and the split is the most interesting thing in the energy complex.
Ceasefire odds firmed hard. Polymarket's US–Iran effective-ceasefire ladder has a deal by 11 September at 62.0%, up five and a half points on the day, by 18 September at 69.5%, up six, and by 30 September at 75.5%. Every leg is thin — $12k to $40k of liquidity apiece — so treat the direction as better evidence than the level.
Shipping went the other way. "Strait of Hormuz traffic returns to normal by September 15" prices at 0.45% on $125k of liquidity and $81k of 24-hour volume: effectively resolved to no. The December contract, on $434k of liquidity and $10.4m of lifetime volume, sits at 26.5%, down five points on the week. Reuters reported Gulf transit holding below its 10-day average and Iran blacklisting further vessels. So the market prices a truce inside a month and a Strait still not functioning normally at the end of the year.
That combination changes which part of the barrel matters. Crude itself is near $91 and WTI's September high-of-$90 contract has already resolved yes, having gained 35 points to settlement in a day; $100 prices at 27.5%, up thirteen and a half points on the week, while a pullback to $80 has fallen to 36.5%, down forty-two and a half. But the sharper move is downstream: Reuters and CNBC both report US diesel at a record high, with Ukrainian and Iranian strikes knocking out refining capacity rather than wellheads. A refining-capacity shock passes into freight, food and every delivered good, which is exactly the kind of inflation a central bank cannot look through — so this reinforces the hawkish case in the first claim rather than sitting beside it.
Dan Skelly, who runs Morgan Stanley Wealth Management's equity model portfolios, described on The Compound and Friends the rule his team has used all year: "every time we've had a ceasefire is to buy more energy in the book, because every ceasefire deal has been pretty short-lasting." He also read the divergence between flat high-yield spreads and widening CCC spreads as partly energy-composition — the high-yield index's underlyings in energy, materials and chemicals are the names benefiting from the Middle East disruption, and post-2015 energy management teams are paid for capital return rather than production growth, so higher prices do not automatically summon supply. The consequence, not the recommendation: a ceasefire alone stops the escalation premium building; only a functioning Strait releases the freight and refining premium, and no contract we can find prices that inside 2026.
3Record AI earnings have stopped moving AI stocks
reported roughly 90% growth and gross margins 200bp lower, and CNBC reported the stock down 5% on the session as weak guidance overshadowed the beat; it closed 3 September at $357.19, 2.8% below the prior close, and went nowhere on Friday. beat and raised into a capacity-constrained backdrop, worked on the day, then failed to follow through. Michael Batnick put the puzzle plainly on The Compound and Friends: investors worry estimates are too high, and when companies destroy those estimates the stocks still do not work.
Skelly's answer is that this is a feature. "I like the fact that it's in slow motion" — a name that does not rocket 70% on the print keeps its upside available, and the muted reaction signals the market is applying rationality and the law of large numbers rather than euphoria. He put the scale of the dependency on the table honestly: 60% of the S&P's 28% headline earnings growth in Q2 came from AI infrastructure, with the median company doing 14% and revenue up 15% on 6% nominal GDP. Morgan Stanley's internet analyst Brian Nowak models 2027 AI capex at $1.5tn against a street number nearer $1.2tn, so the estimate revisions have been running one way.
His own bear points are more specific than the usual bubble talk. He published a note on 1 June cautioning on semiconductor momentum, and the argument was deployment constraints, not demand constraints — servers and chips are being ordered against 40 to 50 gigawatts of projected data-centre construction, and if those buildings arrive late the supply chain has double-ordered. Not a risk today, he said; possibly a risk a year out. The credit side carries the same tension: and are the last two AAA-rated names in the market, CDS trades at notably wide levels, and Michael Cembalest's note puts hyperscaler and investment-grade debt and SPV issuance at $320bn in 2026. Positioning has already leaned in — Goldman data showed leveraged funds' short positions up 35% since mid-June and near record levels, and median S&P 500 short interest as a share of market cap is climbing even in the least-shorted decile, where it sits at an all-time high of just 1.2% against 9.8% in the most-shorted decile.
Tyler Cowen, the George Mason economist, spent an hour on Prof G Markets arguing the framing itself is wrong. "I don't like the word bubble. Were automobiles a bubble in the 1920s?" His distinction: companies failing does not make a thing a bubble; the South Sea Company and tulip bulbs made no sense, and this does. He notes revenue growth at the AI companies compares well with the pre-2000 cohort, the sector is capitalised by equity rather than debt, and token prices have fallen more than 100-fold since GPT-4 even as usage climbs. He expects volatility and says so repeatedly. His actual worry is elsewhere: cybersecurity, where he expects a lot more hacks and a long, expensive remediation, and the political question of how much change society will absorb — the data-centre backlash being the visible edge of it.
Cowen and Skelly disagree about how load-bearing AI is, and neither is obviously right. Cowen: "it's wrong to think that AI is currently this massive net contribution to US GDP and that we would have nothing else if it weren't there" — the capital would have flowed somewhere. Skelly: 60% of the index's earnings growth came from one place, and the question that matters is whether the productivity boost at AI adopters arrives before spending at AI infrastructure decelerates. What would settle it is the next two earnings seasons, in the margin lines of companies that buy AI rather than sell it. Both also agree on one uncomfortable fact — the two largest buyers in the ecosystem are private, unprofitable and unaudited, with Skelly noting their combined annual recurring revenue went from about $10bn late last year to $60–70bn last month.
4A gross-margin miss is the most punished thing in this earnings season
Adam Parker's data, shown on The Compound and Friends, tracks the 12-month industry-group-relative return of companies whose year-over-year gross margins contracted by more than 1%. Those stocks underperform badly and consistently. is this week's illustration — blowout revenue, margins down 200bp, stock down.
Skelly's explanation is the most useful thing in the episode. Companies passed through 60–70% of tariff costs last year and never gave the price back, creating what he calls a synthetic operating-leverage boost for the average company. If you are not showing that gross-margin benefit now, you failed to take price. And if your peer group is capturing some AI productivity on top of an unchanged headcount and you are not, you get "triple dinged" on the same line. The named cohort is food service and staples: , whose earnings blow-up was blamed on paper goods and beef, , — "horrible, that's years" — and , whose Frito-Lay business has missed organic revenue growth targets for two to three years running. There is no offsetting AI story available to them, in Josh Brown's phrase, because Shake Shack is not supplying french fries to data centres.
This matters more than usual because dispersion is extreme. Three-month intra-stock correlations are near their lowest in almost 40 years, so an idiosyncratic miss is not absorbed by the index — each name is marked on its own numbers. It also sets up a hard 2027 comparison: Skelly's own forecast is that the average company struggles to comp the tariff-driven margin boost, and that next year rhymes with 2023, when size and quality outperformed the average stock.
What the sources recommended
Four buys and one sell came out of four videos, and the highest-conviction buy is an event bet the person making it explicitly invited you to decline.
Ross Givens spent his video on , whose Galleri multi-cancer blood test goes before an FDA advisory panel on 23 September. The setup: February's NHS trial missed its primary endpoint on combined late-stage cancers and the stock fell 45% overnight, but stage-4 diagnoses in a pre-specified group of 12 deadly cancers fell more than 20%, revenue rose 26% on more than 61,000 tests sold, and Samsung put in $110m. Shares recovered from about $50 to $80.28 at Friday's close, with the 52-week range spanning $30 to $118 on a $3.6bn market capitalisation. The caveat is his own and belongs with the call: the panel votes after the close, so the move happens overnight, "big upside, but it cuts both ways," and he ends by asking whether it is worth the risk at all. He named as the lower-volatility way to express the same theme and then argued against it — more than doubled in twelve months, near its 52-week high, and the analyst price target sits below the current price. "You're not early."
From The Compound and Friends, Skelly named and as the holdings he thinks the market misunderstands, both survivors of a February–March selloff that priced data-centric businesses for AI obsolescence. His case on is that the issuance business is undervalued in a world financing this much debt, and that ratings are a regulatory requirement rather than a service anyone can substitute. On , earnings compound in the mid-to-high teens while the multiple stays flat, and the exchange is levered to a returning IPO pipeline. Josh Brown offered as his own position and his own conviction — "I think the market is Uber wrong personally" — with no further work shown on air.
The one sell is a disclosed exit rather than a fresh call: Skelly's team took out of the portfolio in the last two or three months at roughly 70 times earnings, on the reasoning that it is a genuine AI winner whose valuation the market has now correctly discounted, and that a 50-stock portfolio cannot carry that tracking error. They exited over a year ago, which is history rather than a call, and it is worth noting reports on 10 September under a chief executive named eight days ago.
Every claim here gets marked to market inside a week
US markets are closed on Monday for Labor Day, so the next session is Tuesday 8 September, and then the calendar does the work.
PPI prints on Thursday 10 September and CPI on Friday 11 September — the release Wang set as the explicit condition for a hold, landing two trading days before the 15–16 September FOMC. A hotter number and the 50.5% hike leg from claim one stops being a coin flip; a cooler one and the doves' read survives its own test.
The same Thursday brings and , both after the close. is the name whose credit spreads were held up as the AI cycle's early-warning system, so its guidance and its balance-sheet commentary read on claim three directly. reports a week after naming Anil Chakravarthy chief executive. Earlier in the week, , , , and report on Wednesday 9 September — the consumer names among them are the gross-margin test in claim four, on a week when diesel is at a record.
