1The rate-hike bet is unwinding on a jobs number nobody has seen yet
Polymarket's September Federal Reserve contract moved hard in one session. No change is now priced at 58.5%, up eleven points in twenty-four hours; a 25bp increase sits at 41.5%, down nine. That book carries $5.29m of liquidity against $86.9m of lifetime volume, so it is the deepest thing on this page. The broader "rate hike at some point in 2026" contract fell the same nine points, to 61.5%.
What moved it was a speech, not a statistic. Reuters reported bond yields falling and stocks rallying as remarks from Christopher Waller curbed hike bets. The 10-year Treasury yield had printed 4.79% on both 1 and 2 September, up from 4.75% on 31 August, and the 2s10s spread widened to 0.43 from 0.40 — a steepening, not a flattening.
Dan Niles, on Excess Returns, published the week's most emphatic hike call the same morning the bet came off. His reasoning starts with Fed chair Kevin Warsh at Jackson Hole: sixty-five months of sustained elevated inflation, and that "sits squarely with the central bank." Niles reads that as a commitment. "I firmly believe there's a rate hike coming on September 16," he said, adding that October is the meeting Warsh will skip, because hiking six days before the 3 November midterms would antagonise the White House. Polymarket agrees on the shape and not the timing: October no-change is 69.5%, a hike 26.5%.
The disagreement is inside the same institution, and neither video acknowledges the other side. Ed Elson, on Prof G Markets, cited a 77% chance of a 2026 hike and treated hawkishness as settled; that contract is now 61.5%. John Murray, chief investment officer at NFJ Investment Group, put the case against on the same episode: the 2-year yield already sits roughly 60 basis points above the fed funds rate, so the bond market has done the tightening, and raising rates "won't necessarily fix the Strait of Hormuz" or the reordering of supply chains that is actually driving the price level.
The August employment report is what settles it, and it lands this morning. That release date is confirmed on the Federal Reserve Economic Data calendar for today. Polymarket's payroll book moved even harder than the Fed book: 0–50k jobs added is priced at 43.5%, up fifteen points in a day, while 50–100k fell to 21% and anything at or above 150k is down to 6%. Add the two loss buckets and roughly a fifth of the book expects an outright contraction. That is a $19k order book, so treat it as a sketch rather than a survey — but the direction of travel over twenty-four hours is unambiguous, and it points the same way Waller does. A hike priced into a payroll print that has not happened is the single most reversible thing in this digest.
2The buyback bid that steadied the tape has been replaced by an equity raise
Two managers reached the same structural conclusion from different data, neither citing the other.
Michael Frazis, chief investment officer at Frazis Capital Partners, described the mechanism on Equity Mates. For more than a decade the largest technology companies stopped buybacks during earnings season, the tape got shaky, and then the buyback window reopened and "those multi-billion dollar bids would come back in and settle the market." That has inverted. The group is now borrowing to fund capital expenditure, and and are issuing equity, which he called "almost unheard of." His conclusion is about variance, not direction: "no matter what, it's going to be a much less stable environment going forward."
Niles arrives there through credit. He checks credit default swaps every morning on , and , and says they are at the highest levels they have ever reached. 's five-year swap, on his read that morning, trades above the average North American investment-grade contract. His argument for why this matters is a risk-asymmetry one: an equity investor can lose everything on one name and still win on the book, so a credit investor has to do more work and is therefore the better-informed marginal buyer. He also flags that 's free cash flow has turned negative for the first time since it went public.
Murray supplies the counterweight and does not dispute the facts. The cost of capital is up, but earnings resilience is beating it, and it is beating it because "we're probably sitting inside the biggest capex cycle in history." The US 30-year yield recently touched its highest level since before the financial crisis, at roughly 5.3%, and the 10-year is above 4.8% — and the Nikkei and the S&P are still at highs. Japan's 10-year crossed 3% for the first time in three decades; Germany's is at its highest since 2011, France's since 2008. His question is when cost of capital starts eroding earnings durability, and he will not name the level.
Niles has a number for the downside if it breaks. trades near 15 times calendar-2027 earnings against the S&P at 19; Cisco in 2000 was above 100 times. So he expects a semiconductor drawdown of 30–50% when the cycle turns, not the 78% the Nasdaq took between March 2000 and October 2002. That is a claim about depth, and it rests entirely on the multiple, not on the cash flows.
The consequence is that the flows that used to absorb a repricing now amplify it. That is a statement about how the next drawdown behaves, not about when one starts.
3Memory's shortage-to-2030 story has a China problem and a MacBook problem
Niles said something he had not said before on the record: he now wonders whether the semiconductor cycle breaks before the AI cycle does, and the reason is China. CXMT, China's DRAM producer, has gone public with about 8% of the global market and plans to lift wafer starts from 300,000 to 500,000 by the end of next year. YMTC, its NAND counterpart, is expected to list soon and has said it intends to be larger than Samsung or SK Hynix in NAND by the end of next year. CXMT also disclosed a high-bandwidth memory part yielding well and roughly one generation behind the leaders.
The historical frame he uses is specific. Intel once held 75% of DRAM; Japan's state-backed entry pushed Intel to the brink of bankruptcy and into microprocessors by the early 1980s. Korea's state-backed entry in the 1990s then drove the Japanese producers out. His point about China is that its motive is different: cut off from US chips, "having their own semiconductor supply is as important as having an aircraft carrier." That makes the capacity insensitive to price. He names the position he is arguing against — shortages running through 2030, memory no longer cyclical — and says flatly that it fails if China meets goals it has already published.
Frazis reaches a compatible conclusion from demand. Memory pricing hit a ceiling the end market would not clear: Apple, he says, pulled its highest-specification machines off the market rather than sell them at the implied price. And a 27-billion-parameter open-weights model from Alibaba, released in the days before the recording, runs on a laptop with no data-centre call at all, uncensored variants included. "That's why the last two days were very bad for semiconductors."
Both flag 's vendor financing, and both read it the same way. Frazis calls it effectively a price cut — margin given away through lending guarantees and equity stakes in customers — which makes real demand impossible to separate from financed demand. He also expects the hyperscalers to keep going vertical, taking 's TPUs, 's Trainium and 's Maia straight to , and treats that as a genuine two-to-three-year threat to 's share.
The Korean retail unwind sits underneath all of this in both videos: leveraged single-stock products on Samsung and SK Hynix, bought on margin, liquidated. Frazis notes that a triple-leveraged product on a single stock has a terminal value of zero, because every stock eventually falls 30–40%.
So the memory debate is now empirical rather than narrative. If CXMT's wafer starts land where it has said they will, the shortage case dies on supply, independent of whatever demand does.
4The day's single-name conviction sits in drug development, and one call is two weeks stale
The technology videos produced no single-name buys at all. Every explicit call today came from biology.
Ross Givens made the most specific one: , Veradermics, an oral time-release formulation of minoxidil. The phase-3 result he cites, published 27 April, grew 33 new hairs per square centimetre against 7 on placebo, with both primary endpoints hit. The confirming phase-3 in men reads out before year end; a women's trial follows in the first half of 2027, and there has never been an approved pill for the 30 million American women with pattern hair loss. He puts the company at roughly $4bn with $820m of cash, funded to 2030, against an average analyst target near $163. The stock closed at $97.76 on 3 September. His explanation for the recent weakness is mechanical rather than clinical: the 180-day lockup from the 3 February listing expired on 2 August, freeing 20 million shares.
He spends as long steering around the crowded alternative. , Absci, is up more than 300% this year on hair loss, has an Eli Lilly investment and ten covering analysts — and, he says, has never grown a hair on a human being, with everything published so far being safety data from 32 volunteers. He quotes the company's own annual report saying the drug's effects "have not been demonstrated in any clinical studies to date." He stops short of telling anyone to sell it, and says he hopes it works, so it is a caution rather than a call; the caveat is the whole content.
Frazis contributed and . On Eli Lilly he is brief and unhedged — the GLP-1s "are clearly just going to continue to power on", with oral formulations still to arrive and retatrutide not yet on the market — and allows that orals might give a way back in. On Tempus AI his logic is second-order: personalised cancer vaccines require tumour genotyping, so the testing companies expand their market whether or not any single vaccine wins.
That call is older than its publication date, and the arithmetic proves it. The episode went up on 3 September, but Frazis discusses being "up 177% in a day" on melanoma vaccine data and being up 25% "yesterday". rose 176.9% on 19 August, from $62.93 to $174.27, and fell 23.5% the next session; it closed at $148.81 on 3 September. rose 23.9% on 19 August, peaked at $72.70 on 21 August, and closed at $64.58. So the conversation was recorded around 19–20 August, and the beneficiary trade has already round-tripped most of the pop the thesis was built on. Frazis is candid that his own process would have sold into it: asked what his model would have done with on the day, he said it "would almost certainly have hit our profit targets and got out."
Two independent sources also reached for the same broken compounder without prompting. Niles named as "one of my big disasters this year"; Frazis asked "have you seen Nike stock lately?" and answered "down 75%." Neither was making a call — both were using it to argue that buy-and-hold fails, alongside for Niles. closed at $38.74. When two managers who agree on nothing else pick the same example on the same day, the example is doing work.
On software, both are constructive and neither will name a favourite. Niles has three niches he thinks are structurally safe — security, systems of record, and video games — and points at rising 35% on results, a reported Silver Lake approach to at around $50bn, and up more than 20% on its Anthropic deal, with the sector up about 4% year-to-date against a semiconductor index up more than 60%. Frazis independently cites accelerating to 28% growth and at new highs, and says cybersecurity names growing 30–40% are the easy part of the argument. Both stop at the sector.
What the sources recommended
Three single-name buys came out of four videos, and all three are healthcare. There were no sell recommendations at all, which is worth holding next to the fact that two of the four videos are explicitly bearish on the next two months.
is Ross Givens's, and it is the only one with a dated binary attached — the confirming phase-3 reads out before year end, and he is explicit that he wants to own it before that rather than after. He acknowledges the generic version of the molecule is cheap and calls that "a real competitive question."
and are Michael Frazis's, and both carry the same two-week timestamp problem described above. The Eli Lilly view is a continuation rather than a fresh thesis. The Tempus AI view is a derived beneficiary call rather than a company-specific one, and the price has already moved and given most of it back.
is named at length and steered around, not sold. Whether that counts as a call is a judgement, and the answer here is no: he says he hopes it works.
Today's payroll print marks claim one to market before lunch
The August employment report lands this morning, and it is the test the first claim asks for. A print in the 0–50k band that the market has spent a day moving toward would confirm the de-pricing of the September hike; anything at or above 150k, currently a 6% bet, reverses a week of positioning and vindicates the call Niles published yesterday.
The corporate calendar is nearly empty by comparison — six listings report, none of consequence, and the next dense session is 8 September. So the macro number has the day to itself.
Oil is the other live test. Reuters has crude set for its steepest weekly gain since mid-July on US–Iran clashes, with US diesel at a record high, and Polymarket's September WTI book has already resolved its "$90 high" leg while the "falls back to $80" leg dropped nine points in a day. Against that, the same venue prices a US–Iran ceasefire by 30 September at 76%, up seven and a half points. Those two moves point in opposite directions and both happened in the same twenty-four hours. The ceasefire book is thin enough that the oil book is the one to believe until the ceasefire one deepens.
