1A rate hike repriced to nearly 90% in twenty-four hours, on both venues at once
The September decision lands tomorrow, and the two deepest prediction markets moved together overnight. Polymarket prices a 25bp increase at 87.5%, up eight points in a day, on $3.5m of volume in that leg alone and $12.3m of book liquidity. Kalshi's matching contract last traded at 88 cents against a previous 79, on roughly $2.0m of 24-hour volume and $8.5m of open interest. The hold leg fell on both: 10.5% on Polymarket, 12 cents on Kalshi from 20. A cut prices at one cent. Two venues, two order books, the same nine-point move — that is corroboration rather than one crowded market talking to itself.
The bond market got there first. The ten-year Treasury yield topped 5% for the first time in three years, and Reuters put the thirty-year at its highest since 2007. The two-year/ten-year spread narrowed to 0.32 on 14 September from 0.39 on the 10th, which is the short end repricing faster than the long end — the shape a market makes when it expects tightening, not when it fears recession.
The hard print underneath is August CPI: 334.131 against 332.813 in July, a monthly gain of 0.40% after July's 0.07%, and 3.35% year over year. The monthly pace roughly quintupled. Oil is the mechanism most cited for it — Brent held around $105 with crude touching $107 last week, the first time since May, after the US destroyed five Iranian oil tankers. Scott Galloway and Ed Elson, on Prof G Markets, ran the pass-through: diesel is 15–30% of grocery cost, diesel prices are up 57% since the war began, and corn, wheat, cotton and rice are up 16%, 23%, 32% and 49%. A fuel price is not an isolated line item when almost everything arrives by truck.
Timing sharpens the point. That Prof G episode was recorded on 14 September and cited hike odds "above 60%", and 65% on Kalshi earlier in the show. Both numbers were accurate when spoken and are twenty-plus points stale now. So what: the hike has become the base case rather than the surprise, which moves the asymmetry to the other side — a hold is now the outcome that would force a repricing.
2The AI-slowdown selloff rests on a compute assumption the one operator asked says is backwards
Dario Amodei published an essay on Saturday titled "We must pace the frontier", warning that a swarm of rogue agents could take over the internet within six to twelve months. Sam Altman and Elon Musk each posted agreement within hours. Felix, on Felix & Friends, made the structural observation that the three are rivals who agree on nothing and are suing each other, so their agreeing at all is the information.
Monday's tape took it as a spending signal. closed down 3%, down 4%, down 7% and down 14%, with the Philadelphia semiconductor index off 5.86% and down 5.6%. Money moved to defensives and to security: rallied 14%.
Charlie O'Neal, co-head of model training at Baseten, argued on Prof G Markets that the inference is backwards. Pacing the frontier does not mean training smaller models. It means allocating a share of compute to monitoring and safety on top of the existing roadmap — he cites rumours of up to 20% of internal compute at OpenAI, and says the figure at Anthropic may be higher. On his read that raises total compute demand rather than cutting it, and he puts the price of a megawatt at roughly $15m now against $10m previously, heading to $20–25m next year. His words: not "a bearish sign for the amount of compute the world is going to need."
He is not disinterested — he works with open-source models, and the accusation circulating is that the frontier labs are pursuing regulatory capture against exactly that constituency. He rejects the accusation, which is the direction his own interest would push him. Worth weighing either way.
Two facts sit underneath and neither is settled. Altman told Fortune that OpenAI will not go public in 2026, calling it an "ill-advised moment". Anthropic has told investors, per the Financial Times, that it has been profitable for two straight quarters — but only on an adjusted operating basis, and its claimed gross margin above 80% is struck before revenue-sharing agreements and before the cost of training its models, which are its two largest expenses. Elson's position is that the adjustment is doing the work and the claim means nothing until the filing exists. Michael Burry, quoted in Monday's coverage, called the slowdown rhetoric IPO hype rather than safety. So what: whether Monday's selloff was correct turns on unit economics that exist in no public document, so the argument cannot be settled by anyone outside the two companies until a prospectus is filed.
3A trillion and a half dollars of AI spending has not produced a productivity gain, and two separate reads land on inflation
Two sources reached the same finding this week from different directions, and neither cites the other.
Felix, on Felix & Friends, put up Bank of America strategist Michael Hartnett's chart of productivity against consumer confidence. The two have tracked each other since 1978; this week both fell together, which Hartnett notes has not happened since 2008. Alongside it: $1.5trn spent on AI, and in Hartnett's words no evidence yet of an economy-wide productivity gain.
John Kerschner and Mike Keough of Janus Henderson, interviewed on Excess Returns, arrived at the same absence and pushed it one step further — AI is inflationary in this cycle, not deflationary, through four channels. Capital spending on data centres, semiconductors, power infrastructure and cooling runs to hundreds of billions a year. Energy demand is bidding up electricity. Labour is scarce with 12,000 baby boomers retiring daily against demand for AI engineers and infrastructure specialists. And a wealth effect from rising AI equity prices keeps upper-income spending elevated. Keough's framing: productivity "has actually gotten worse", so he cannot locate where the disinflation is meant to come from.
Host Jack Forehand flagged the weakest leg himself — the wealth effect is questionable, because a rich household's marginal propensity to spend is low. He also drew the boundary: this says AI is inflationary now, not permanently.
Where the two sources diverge is the conclusion each draws. Felix routes it to gold, noting central banks bought a record $40bn in a single quarter through a 22% drawdown, with gold near $4,300–4,350 against published targets of $4,900 at Goldman Sachs and $6,000 at JPMorgan. Janus Henderson routes it to fixed-income construction. So what: if the productivity dividend is genuinely absent rather than merely late, the bond repricing in the first claim is a level change rather than an overshoot.
4The same 5% yield splits four investors four ways
With the ten-year through 5%, four people looked at the same number this week and drew four conclusions.
Kerschner's objection is to the index, not the yield. The Bloomberg Aggregate is built by who issues the most debt, which is why it now carries roughly six years of duration against about a 5% yield — a 100bp rise produces a negative total return. Treasuries and near-Treasuries are about 50% of it, up from roughly 30% coming out of the financial crisis. His conclusion is that a core-plus allocation buys more rate risk and less yield than its owner believes, and that securitised credit outside agency mortgages, emerging-market debt and high yield are the parts most investors under-own.
Ed Elson, on Prof G Markets, went the other way: with 3–5% inflation possibly persistent, "bond investing as we know it" is finished, and the only remaining answer is to hold more equities. He notes long-dated Treasuries have posted their worst decade in more than a century. Galloway pushed back in the same conversation — investment-grade corporate credit at 6–8% is being paid for its risk for the first time in years, and his stated objective is "not to get rich, it's to not get poor". Same show, same segment, unresolved.
Todd Barlow, CEO of Washington H Soul Pattinson, told Equity Mates he is sitting on just under A$3bn of cash, about 20% of the portfolio and sharply higher than usual, having reduced public equities and sold a A$1.9bn industrial property portfolio to Goodman Group. His reason is the same rate level: "we're getting paid to wait", and fixed income is "actually a good place to deploy" for the first time in a long while. That is a CEO describing his own book on his own show, with no bear case put to him, so read it as a position rather than as analysis.
Barlow's market-structure observation stands on its own though: capital has fled to large caps, which now look well priced, leaving small and mid caps outside the indices behind — and private equity has responded with a wave of takeover approaches, including to companies he holds. His concern about US private credit is the sponsor-leveraged-finance book, much of it lent against SaaS businesses now facing AI disruption, where he notes that by the time credit takes a dollar of loss the equity beneath it is already gone. So what: four allocators reading one yield four ways is the honest state of the question, and none of them is waiting for a new data point to settle it.
What the sources recommended
One single-name call came out of six videos, and it is a negative one whose author concedes it has not worked. Ed Elson reiterated a bearish stance on on Prof G Markets, on valuation — 36 times earnings against about 26 for the S&P 500 — and on a foldable iPhone he considers an unimpressive use of the capital Apple saves by staying out of the AI buildout. He states plainly that he has been wrong so far: Apple is up 19% year to date and 40% over twelve months, and is the outperformer among big tech. His co-host disagrees on the same show, calling the product launch a likely hit. Treat it as a contested view, not a call the channel is united behind.
Ross Givens reiterated on Ross Givens as a two-month-old recommendation "up about 40% and counting", inside a tutorial rather than as a fresh entry, and named metals and mining as the leading theme — gold miners, rare earths and steel — pointing at the holdings list as the place to source names. The screen capture he used to demonstrate it is dated February, which makes the demonstration a re-cut rather than current work.
None of the six videos named a new long. That is not the harvest being thin — it is what these six chose to spend their time on instead: two on rates, two on whether AI pays, one on gold, one on portfolio construction.
The hike thesis gets marked to market in twenty-four hours
The September FOMC decision arrives Wednesday 16 September; Kalshi's contract closes at 13:59 ET. Both Prof G hosts predicted a hike, Galloway arguing that Kevin Warsh has a twelve-year appointment and cannot be removed, so the decision costs him nothing politically and would be well received. At 88 cents, the market has already taken that side, which means the informative outcome is the other one.
Two clean reads on the same thesis report the next morning. is a direct measurement of freight and diesel cost, the exact channel by which oil reached the CPI print above. is a direct measurement of what a 5% ten-year does to housing demand. reports today and carries the grocery-cost pass-through. And the Senate votes on cloture for the CLARITY Act, with , , and all lower pre-market ahead of it.
