1September is priced for a hike at 27.5% and a cut at 1.25%, six weeks after payrolls fell 23,000
Polymarket's September FOMC event is the deepest book checked today — $41.0m of volume against $3.5m of liquidity — and it prices no change at 71.5%, up a point in 24 hours. The live alternative is not a cut. A 25bp hike trades at 27.5%; a 25bp cut trades at 1.25%, and a 50bp cut at 0.25%. The market's distribution is one-sided in the direction most macro commentary is not.
Set that against what the labour data actually printed. July non-farm payrolls came in at 158.858m against 158.881m in June — a fall of 23,000 jobs. Unemployment nonetheless dropped to 4.1%, from 4.2% in June and 4.3% in May. Initial jobless claims for the week ending 15 August were 206,000, down from 212,000, which is historically low; continuing claims sat at 1.799m. Firing is not happening. Hiring is not happening either, and unemployment is falling because the labour force is shrinking rather than because anyone is being hired.
Andy Constan of Damped Spring Advisors, speaking to Justin Carbonneau and Jack Forehand on Excess Returns, put the puzzle plainly: long-term interest rates have been rising "while every major piece of data has been on the cool side — retail sales most recently, inflation, two months of benign to soft prints, NFPs that were weak, and yet the bond market is selling off."
His explanation is a change at the Fed rather than a change in the data. Constan says Kevin Warsh stated five or six times at the July meeting that he is happy the bond market is "doing the work of the Fed" — that is, that higher long-term yields are a tightening he welcomes — and has signalled interest in shrinking the balance sheet and changing its composition. Both would remove long-end suppression rather than add it. Constan is careful to call this a "drum beat" and not a decision: the balance-sheet work sits with a task force, nothing has been done, and he expects Jackson Hole to defer rather than announce. He notes three FOMC members already want to hike and that four more would be needed to override the chair.
He also names the two things that would falsify it. Scott Bessent could mute balance-sheet tightening the way Janet Yellen muted quantitative tightening by financing the government with bills, and Constan sees no evidence he is on board. And Constan thinks the market read that Warsh lacks credibility — the read he says explains both gold and bonds — is wrong, while declining to trade against it: "they could be right. I don't think they are, and I think they'll be surprised, but that's the narrative that's playing through markets. And so, I'm just not going to run in front of that."
The tape agreed with the hawkish half today. fell 0.90% while the S&P 500 fell 0.60%, so long bonds and equities dropped together rather than one hedging the other. The 10-year yield was 4.71% on 18 August, up from 4.68% on the 14th. The 2s10s spread narrowed to 0.46 on 19 August from 0.53 on the 17th — seven basis points flatter in two sessions.
The consequence is that a soft data print no longer buys a cut. It buys a flatter curve, because the front end is anchored by a committee that is not being asked to ease and the long end is being left to trade.
2The long-end sell-off is global: Japan's 10-year sits at a 1996 high and Australia's above 5%
Two channels reached this independently on the same day, with the same number, which is worth stating because it is the kind of agreement that usually turns out to be one source counted twice.
Felix Prins, on Felix & Friends, describes Japanese 10-year yields jumping to almost 3%, the highest since 1996, with the 30-year above 4% and the yen at 163 to the dollar. Bryce Leske and Alec Renehan, on Equity Mates, give the Japanese 10-year as 2.93% and call it the highest since 1996 — arrived at from an Australian desk, on a different show, with no shared guest. Equity Mates adds the rest of the curve: the US 30-year at about 5.31%, the highest since 2007, and the Australian 10-year above 5%, off a 5.16% peak in March 2026 that was itself the highest since 2011.
Their readings of it diverge sharply, and that is the more useful part. Equity Mates treat it as an inflation-persistence signal — "a sign that the market thinks inflation is stickier than a lot of politicians are willing to admit" — and reach for inflation-protected bonds, pricing power and infrastructure. Prins treats it as currency debasement, citing a Bank of America note from Michael Hartnett titled "strife begins at 40", a play on US debt approaching $40tn, whose recommendation he reports as long gold and "anything but bonds, anything but the dollar". He also points to US Vice President JD Vance describing the dollar's reserve-currency status on camera as a "resource curse" — a subsidy to the American consumer and a tax on the American producer.
Prins's other claims this week are harder to check and should be held more loosely. He describes a joint US-Japan intervention of about $85bn over two days, a Federal Reserve balance sheet growing again through repo lending and "reserve management purchases" at $40bn a month, and a 10% single-session fall in the Korean market that tripped a circuit breaker, with (SK Hynix, US OTC line; primary listing Korea) down 9% and Samsung down 6%. None of that appears in today's news fetch, which instead carries SK Hynix rising 4% on 19 August and a failed buyback pressuring , and pre-market. The two accounts may simply be describing different days of the same week; the transit of a circuit breaker is a matter of exchange record, so it is settleable, and until it is settled the Korea material carries less weight than the yield numbers, which three separate sources agree on.
Where the tape can adjudicate, it partly backs Prins. Gold sits 18.7% below its 29 January high, close to the "about 20%" he cites, having recovered 35.8% from last August's low. Silver is 43.9% below the same January peak — a drawdown 2.3 times gold's, from the same date, which is what his "wilder brother" description means in numbers. rose 2.62% today against 's 0.12%. Polymarket prices gold reaching $5,000 by end-December at 57%, up 2.5 points, though on $13.8k of liquidity that contract is too thin to lean on.
The consequence is that this is not a US fiscal story that happens to have spread. Long-dated debt is repricing in Tokyo, Sydney and Washington at once, which narrows the plausible causes to ones that are common to all three.
3The AI capex question that decides the next six months is who funds it and where it can be built
Constan's argument is that the return on AI investment — the question the Excess Returns hosts say they put to guest after guest — is not the binding question over the next few quarters. The binding question is whether the capital markets stay open long enough to fund it.
The mechanism is a change in who pays. For years the hyperscalers funded capex out of accumulated cash and operating cash flow, because they were capital-light businesses that had been accruing earnings for decades. About a year ago, Constan says, that ran out. Capex now runs $600–700bn this year against roughly $1tn next year, and the gap has to come from corporate bond and equity issuance. Equity Mates put third-party numbers of the same order beside it: Goldman Sachs at $920bn of data-centre spend next year, Morgan Stanley at $1.1–1.2tn.
Constan calls it the hamburger thesis, after Wimpy in Popeye: "we're going to build you a data centre, a hamburger. You just loan us the money today and we'll pay you back on Tuesday." What worries him is not the promise but the arithmetic underneath it. US corporations have net-retired roughly $1tn of shares a year for decades, a passive bid he credits for part of the equity rally, and he expects that to fall to zero or turn to net issuance next year. His concern is that the market closes before the trillion is through the door: "if they close at $500bn and there's $500bn of new funding that's needed to fund the capex, what happens?" The reaction would come not from the missing supply but from the forward earnings already booked against it.
He watches issuance marks rather than forecasts, and they are the most checkable thing he said. Alphabet sold $80bn of stock at $355; trades at $340.82, still underwater. priced at $135 and traded $143 when he recorded; it is $131.67 now, which is below the deal price and strengthens his point rather than weakening it. is back at its issue price after rallying $10 off it. Every corporate bond issued in the past year, he says, trades at a wider spread than it came at. His read on the $500bn structure announced is deliberately measured: he describes collateral dropped into a trust with an Nvidia credit guarantee and BlackRock and Apollo taking the tranches, calls it a CDO in structure, and says explicitly that it is not an Enron signal — only that Nvidia has run out of capacity for direct deals and needed another route, which he grades "a slight negative". is his precedent: a data-centre announcement that ran the stock to the mid-200s, then a disclosure of how it would be funded that halved it.
Equity Mates arrive at a second constraint that has nothing to do with financing. A Gallup poll found 71% of Americans oppose a data centre being built in their community — a higher share than oppose a local nuclear power plant. In the first quarter of 2026 alone, 75 projects representing $130bn of investment were blocked or delayed, more than double the $64bn blocked across 2023 to 2025 combined. New York has imposed a one-year moratorium on facilities drawing more than 50MW, and around 15 states are considering similar measures. The driver is electricity: data centres take 6% of US power, 5.8% in the UK, 9.5% in Germany, and about 23% of Ireland's in 2025, nearly as much as all its households.
So the trillion has to clear two gates before it becomes revenue, and neither is a question about returns. It has to be financed at a price someone will pay, and it has to be sited somewhere that will have it.
4Copper's spot premium went from $34 to $545 a tonne, and the miners already banked the metal's move
Ross Givens makes the physical case, and the supply side of it is specific enough to check. London Metal Exchange copper inventories are near 200,000 tonnes and have fallen for 42 consecutive trading days, which he calls the longest drawdown streak since 2014, with roughly half of what remains already spoken for. The premium for immediate metal over three-month delivery went from about $34 a tonne at the end of July to $434 by the middle of last week and $545 this week. Backwardation that wide means someone needs physical copper now and is paying to jump the queue.
Three separate supply breaks sit behind it. The Democratic Republic of Congo banned exports of copper and cobalt concentrate outright on 6 August, affecting roughly a fifth of national output. Chile is expected to produce 2.6% less than last year, with Codelco pushing its El Teniente expansion to 2029 after a seismic event and Antofagasta cutting guidance. And about 17% of world copper is produced using sulfuric acid, whose supply was hit twice — China banned acid exports in May, and the Hormuz closure removed roughly half of seaborne sulfur shipments. Global mine production fell 1.6% over the first five months of the year; the International Copper Study Group expects refined production to grow 0.4% for the full year.
The tariff question is the near-term swing factor and it is genuinely undecided. April's 50% Section 232 tariff covers copper pipe, wire, rod and sheet but not refined cathode, the benchmark grade. Commerce recommended a phased cathode tariff starting at 15% in January and stepping to 30% the following year, and the White House ruling is roughly seven weeks overdue. The US imported more than 200,000 tonnes of refined copper in July, the largest month in twelve years, which is what front-running that decision looks like. Polymarket does price this — Section 232 tariffs on copper cable by end-2026 at 15%, up 3 points, and the 2027 leg at 31%, down 6.5 — but on $13,909 of lifetime volume and $488 of liquidity, which is not a market so much as a quote.
Givens's demand argument connects this section back to the last one: a data centre needs roughly 27 to 33 tonnes of copper per megawatt, BHP measured a 180MW build consuming more than 2,000 tonnes before grid reinforcement, and the International Energy Agency reckons existing mines cover about 70% of projected 2035 needs.
The part that deserves a counterweight is the claim that the equities are early. His own best evidence is operating leverage — produced 3.5% less copper last quarter and reported net income up 71.6% as cash costs collapsed — but that leverage is precisely why the shares have already run. Over twelve months the metal proxy is up 40.7%, close to the 46% he quotes for futures, while is up 66.5%, 89.4%, 105.5% and 161.1%. Today made the same point in miniature: rose 2.09%, 1.29%, 1.24% and 1.00% while fell 0.41% and the S&P 500 fell 0.60%. The squeeze is documented and the miners are the geared expression of it; the geared expression has been paid once already.
Two disclosures belong with the call. Givens says he is long copper futures again and booked roughly $30–35,000 closing a July position, and he promotes a paid subscription service several times inside the segment.
What the sources recommended
Nine single-name calls came out of four videos, and they cluster in two places rather than spreading across the tape. Six are the copper complex, all from one presenter in one video, so they are one call expressed six ways and not six independent votes.
The buys. Ross Givens named (Freeport-McMoRan), citing Grasberg throughput doubling between April and June and cost guidance cut to $1.90 a pound against copper near $6; (Southern Copper) on record revenue and margins and a $3.23 dividend; and (Taseko Mines), his preferred pure North American exposure, where price collars at $5.40 a pound expired in June so the company captures the full price for the first time — though its cost per pound is the highest of the three at $2.41. For anyone not wanting single names he offered for the metal and for the basket. The caption garbled Taseko's stated June rebrand, so the company name is reported here from the ticker he gave, which is correct.
From Equity Mates, the invited expert Owen Rask called (Southern Cross Electrical Engineering, ASX) "top of the list" for Australian exposure to the data-centre build, with (Maas Group, ASX) named in the same breath as a different business in the same position. His caveat is attached and load-bearing: this is a tender-driven industrial services business whose margins have not risen with revenue, and in his experience the de-rating in such names arrives before management discloses that the tender book has slowed, because investors are forecasting what others will forecast. The stock is up 125% in the past year. Separately (Scorpio Tankers) was the pick of the community member whose portfolio was reviewed, not of the show or its expert, and his stated reason for owning it is that Michael Burry bought it — Burry sold within a few weeks and he held on.
The sell. (SpaceX), from Andy Constan, who calls it "incredibly overvalued" and says the post-lockup supply is the real story. He discloses a small personal short from about $145 in an account he calls his degenerate account, describes it as trading for lunch money, and notes it went against him. Treat it as a view held in size zero.
The next 24 hours test the funding claim, not the rates one
Walmart is the day's evidence that the consumer half of the pie argument is not hypothetical: fell 9.05% to $104.05 on subdued US sales growth, the largest single-name move on the tape, and it dragged the Dow to a 1.04% loss against the S&P 500's 0.60%. Small caps fell hardest, with down 1.37%.
The bigger test is dated. reports after the close on 26 August, with , , and the same day and , and on the 27th — the week that marks the AI capex forecasts to market. Constan's expectation is that Nvidia's numbers and guidance are "through the roof", which is why his argument does not rest on them; it rests on what the issuance calendar looks like afterwards. has already risen 13.2% in three sessions on a Google custom-silicon deal, and gained 0.51% today rather than falling on the same news, so the market is not yet treating that contract as zero-sum.
Before then, Jackson Hole is where the balance-sheet question either advances or defers, and the source closest to it expects deferral.
