1The September contract flipped to a hike, and it moved twenty-two points in one session
Kevin Warsh gave a first Jackson Hole speech as Federal Reserve Chairman this morning and said the summer's inflation readings "do not tell me that underlying inflation trends have meaningfully improved." The supporting number was a distribution, not an average: 54% of the components of PCE inflation ran above 3% annualised over the past twelve months, and 49% did over the past six. That is the argument for why a cooling headline is not a cooling trend.
The rate market repriced immediately and heavily. On Polymarket's September contract, a 25bp increase now trades at 52.5%, up 22 points on the day, while no change fell to 46.5%, down 21. Every cut leg together is under 1%. This is the deepest book in the day's set — about $62m of lifetime volume, $3.6m of resting liquidity, and $6.9m traded in the last 24 hours, roughly six times the previous day's turnover. Yesterday the same contract had no change at 67.5% and an increase at 30.5%. One session inverted it.
A second venue says the same thing independently. CNBC reports fed funds futures moved the September hike probability to 55.7%, about 20 points higher than a day earlier. Two differently-constructed markets moving the same direction by the same magnitude is a stronger signal than either alone. The separate full-year contract — any hike in 2026 — sits at 68%, up 11.5 points, on $8.1m of volume, though only $79k of that traded today, so treat the annual leg as the thinner read.
The cross-asset tape agrees. Gold fell hard, with down 3.2% — a higher expected real policy rate raises the cost of holding an asset that pays nothing, and that is the cleanest read-through on the board. Semiconductors fell 3.2% via and utilities 1.3% via , both long-duration cash-flow profiles that discount worse when the front end reprices upward. Short and intermediate Treasuries fell, which is yields rising where a hike bites. The index itself barely moved, down 0.3%. The repricing was violent inside the market and quiet at the top line.
One thing this does not settle. Warsh is now openly opposed to the administration's demand for lower rates, and Reuters reported yesterday that Scott Bessent and Warsh diverge on who should set the price of money. A 52.5% probability is a coin flip that has been handed a political dimension, and nothing in today's data resolves it.
2One capital-spending forecast was bought and sold on the same day
Nvidia's results on 26 August produced a single number that both of today's arguments are built on: roughly $800bn of hyperscaler capital spending this year and $1.3trn in 2027, against $96.2bn of quarterly revenue, up 106% year on year, and cloud-industry backlog above $2trn.
Ross Givens bought the picks-and-shovels read, on camera, with $50,000. The reasoning is a bottleneck argument: memory can be engineered around and power cannot. Bank of America reported this month that Nvidia is testing its Rubin Ultra chip at 192–288GB of high-bandwidth memory against an original one-terabyte specification, and Google published a memory-compression technique in June that knocked memory names lower. Meanwhile the same roadmap puts Rubin Ultra at 600 kilowatts per rack with no fans at all. Givens' line is the compressed version: "Nobody has figured out how to compress a gigawatt." The purchase was — Vertiv, data-centre power distribution and liquid cooling, about $11.5bn of revenue, organic growth guided to 30–32%, deferred revenue doubled in six months from $1.8bn to $3.6bn — at roughly $266, 188 shares, with a stop at $248 and a target above the $380 prior high against a $388 average analyst price target. Two objections were volunteered rather than hidden: Vertiv stopped disclosing backlog in the second quarter, and Eaton, Ecolab and Trane have each bought their way into thermal management.
Ed Zitron, on The Compound, took the same forecast as evidence of the opposite. The demand behind it, on this reading, is concentrated to the point of being circular: 16% of Nvidia's latest-quarter revenue came from one customer, 44% of the first half of fiscal 2027 from three, and five customers make up 70% of accounts. Beneath that sit two loss-making buyers. Microsoft booked $34.33bn of AI revenue in fiscal 2026 and $24.1bn of it — 70% — was OpenAI, which means everything Microsoft sold to every other customer is a single-digit-billion business set against $260bn of capital spending. OpenAI's last raise was $122bn with only $12bn from venture capital; Nvidia put in $30bn, SoftBank $30bn and Amazon $50bn. Zitron's summary of the demand signal is that OpenAI and Anthropic take roughly 90% of AI infrastructure, so scarcity is being read as breadth.
Both cannot be right, and the market did not pick a side today — it sold both. fell 4.0% to $218.95, giving back nearly half of the post-earnings move. fell 3.9% to $258.83, which puts it below Givens' $266 entry within a day of the purchase and 4.2% above the $248 stop — the stated test is whether it holds the week's low, and that test has not been asked yet. The hyperscalers went the other way: +2.3%, +3.6%, +1.7%, +0.8%. Money moved from the suppliers to the spenders on a day when a Federal Reserve Chairman made long-duration cash flows more expensive. That ordering is hard to reconcile with either thesis and is worth naming as unresolved rather than explained.
The two also disagree about memory in opposite directions from the same fact. Both accept that memory costs are rising and that Nvidia's gross margin is guided down from 75% to about 71% by the fourth quarter. Givens reads that as a reason to avoid — memory is paid on price times bits, and the bit count is being cut. Zitron reads rising memory cost as one more thing stacked against Nvidia's ability to reach the roughly $674bn that consensus wants from fiscal 2028. closed the argument unimpressed, down 0.5%.
3The mania is not in the equity multiples — it is in the credit
The most useful exchange in the Compound episode was the host's objection rather than the guest's thesis. Josh Brown put it directly: Nvidia trades near 18 times forward earnings with 70% margins and Meta near 16 times, so where exactly is the mania? Zitron conceded the framing and relocated the answer — into data-centre construction, the neoclouds and private credit, none of which carries an equity multiple you can look up.
That relocation is checkable, and the named tests are specific. Amazon's recent bond sale was only 1.6 times oversubscribed where a subsequent Google issue drew four to five times; a hyperscaler deal pricing near flat would be the visible sign that appetite for data-centre paper is thinning. CoreWeave has raised billions three times this year, at one point paying above 9%, and needs to keep doing it; fell 3.6% today. Oracle is down about 65% from its high, rated a notch above junk, and — on Zitron's account — has revenue that has been flat for fifteen years in real terms; fell 1.0% to $150.47 on a day the software complex barely moved. Roughly a sixth of insurance annuities are private-credit funded, which is the channel by which a data-centre write-down reaches a balance sheet that never bought a GPU.
A same-day event sits awkwardly close to that argument. Bloomberg reported this morning that Advent and Stripe abandoned their pursuit of PayPal, a deal reported at $60.50 a share and more than $53bn that would have ranked among the largest leveraged buyouts ever attempted. fell 12.3% to $53.88, which leaves it 11% below the abandoned bid; the reporting says the situation stays fluid and the consortium could return. This is not evidence that credit markets closed — PayPal rejected the first offer as insufficient and the two sides were negotiating price, which is a valuation disagreement, not a financing failure. It is worth recording only because the largest buyout in the pipeline stopped in the same week the argument on the table is that leverage is the fragile part, and because a returning bid or a stated financing reason would tell you which it was.
Two caveats belong to this section rather than the appendix. Zitron holds no position at all — "I don't have cash in the market," with reputational rather than financial exposure — so the loudest version of this argument is being made by someone who is not short it. And this was the show's first explicitly AI-bearish guest in five years by the hosts' own account, on a programme whose guests they describe as mostly bullish. That makes the episode a deliberate counterweight, not a shift in the venue's house view, and it should be read as one.
4The Hormuz reopening is being priced further away for a second day
Six months into the Iran war, the news flow this week pointed at a settlement and the money moved the other way. Reuters reported Iran setting conditions for reopening the Strait of Hormuz, mediators focused on the strait, and Iran and Oman working on a revenue-sharing accord — while Donald Trump said the United States is not talking to Iran and rejected a return to earlier ceasefire terms.
The contracts fell on all of it. An Iran-Oman Hormuz management agreement by 30 September now prices at 35.5%, down 7 points today, after falling 14 points yesterday — two consecutive sessions of the deal being priced further away, on $1.3m of volume and $96k traded today, which is thin enough to weaken but not void the read. The August leg of the same contract is at 4.5%, down 4, and expires Monday. Traffic returning to normal by 31 August is at 0.25% on a $24m book and is effectively already resolved; the 30 September leg is 4.5%, down 2, on $722k of resting liquidity. The market's verdict on this week's diplomacy is that talks about talks are not a deal.
The physical picture supports the pricing more than the headlines do. A US Navy blockade has cut Iranian oil exports as the administration shifts to economic warfare; sanctions were extended to Egyptian bank branches; QatarEnergy cancelled gas deliveries to Italy's Edison until early November; and Japan is backing pipelines that bypass the strait while Gulf states fund new ports. That is capital being spent on routing around Hormuz, which is what you build when you do not expect it back soon.
Crude has still drifted lower — WTI printed $83.90 on 25 August against $89.75 on 20 August — and neither energy proxy moved much today, with down 0.2% and up 0.4%. So the near-term price of oil is being set by the demand side and the Fed, not by the strait, even while the strait's own contracts say the supply problem is not close to being fixed. A ceasefire stops the escalation premium from rebuilding; only a reopening returns the barrels.
What the sources recommended
Four single-name-or-fund calls came out of three videos, and three of the four carry a caveat that belongs with them. is the only conviction call: bought live with $50,000 at about $266, stop $248, on the argument that data-centre power and cooling is the bottleneck that cannot be engineered away — and it is already below the entry. The same video steers clear of and memory generally, for the mirror-image reason. That video also promotes a $5-a-year subscription service twice, with a QR code, which is a paid-promotion context readers should weigh.
The other two calls came from a sponsored segment and should be read as such. On an episode PIMCO paid for, PIMCO's head of ETF sales named two of PIMCO's own funds — , the global bond fund, quoted at about a 7.6% yield to maturity at end-July on a AA− average credit rating, and , a half-Australian half-global bond fund — both listed in Australia, not the United States. The general case made was that high-quality fixed income now offers equity-like returns of six-and-a-half to seven-and-a-half per cent, so a retirement portfolio can be funded from income rather than by selling capital into a down market. Two further disclosures matter: the episode was recorded on 13 August and published on 27 August, so its macro references are two weeks stale, and the podcast's parent is itself an ETF issuer.
The next four sessions test every claim above
The rate claim gets marked first and hardest. Warsh's argument was about the distribution of PCE components, so the next core PCE print is the direct test of it; the last one landed 26 August and read +3.34% over twelve months against an effective fed funds rate of 3.63%, a real policy rate of roughly a third of a point. Until then, the 52.5% is a market opinion about a speech.
The capex claim gets tested by earnings. reports after the close on 2 September and is the cleanest independent read on whether AI infrastructure orders match Nvidia's forecast rather than merely echoing it. reports 1 September, with and the same evening, and on 2 September, and and on 3 September — a dense week for the software half of the argument that ran hardest yesterday. reports 31 August before the open, which prices tanker rates against a strait the contracts say stays shut.
The Hormuz claim resolves mechanically. The 31 August legs of both Hormuz contracts expire Monday at what is now near zero, which converts today's probability into a fact and resets the question onto the September legs.
