By Lily Caruso · Thursday, September 3, 2026 — The Morning Brief, filed mid-morning** --- Good morning. Pour the coffee, and let's start with the sentence. At some point before the bell, Federal Reserve Governor Christopher Waller — the last policymaker who could legally speak before Saturday's blackout — said he would be "inclined to support" holding rates steady if inflation data continues improving.
That's it. That's the whole event. No decision, no vote, no data.
One governor, one conditional clause, and by mid-morning September hike odds had fallen roughly fifteen points to about 48%, the ten-year had retreated to 4.75%, and gold had ripped nearly two percent. I want to sit with how strange that is, because it tells you what kind of market this is. We spent all week watching companies report tens of billions of dollars of actual revenue — Dell's $47 billion, Broadcom's $29.6 billion, HPE's $12.2 billion — and the tape mostly punished them.
Then one man attached a conditional to a hypothetical, and every asset class moved. **This is a rates market wearing an earnings market's clothes**, and until the folder on the Fed's desk closes, it will stay one. **The gold reversal, and why I care more about it than the price.** For three straight sessions this page printed a claim in various outfits: gold is falling *through* war headlines because the market fears the Fed's response to inflation more than it fears the fire itself. A claim like that is only worth anything if it's testable in reverse — and this morning it got tested under laboratory conditions. Oil went *up* about two percent on renewed Hormuz risk.
Brent pressed $97. And gold, which had been sold on every one of those headlines, rallied **1.81% to $4,494.70** the moment a Fed governor sounded dovish. Same war.
Different Fed. Opposite gold. That's not a trading recommendation; it's a confirmation that the transmission line we've been drawing — war premium → inflation expectations → rate path → real yields → gold — is the actual wiring of this market.
And it cuts both ways. If Friday's payrolls come in hot enough to un-say what Waller said, that 1.81% is on loan. **Room two: the banks led, and this time it wasn't the steepener.** Here's the wrinkle I find genuinely interesting. On Wednesday I wrote that financials were being bid because a *bear steepener* — the long end selling off faster than the front — is the one curve shape that pays banks.
This morning the mechanism reversed and the bid got stronger: **XLF +0.92% at $58.19**, the best major sector on the board, on a morning when the long end *rallied* and yields fell. So which is it? Both, and that's the point.
Wednesday's bid was about the curve. This morning's is about credit and volume: a Fed that holds instead of hiking means cheaper funding, a friendlier borrower, fewer loan-loss headaches, and — the part nobody says out loud — a live deal calendar. Nvidia announcing a $12.9 billion acquisition on the same morning is not a coincidence in that sense; it's the kind of thing that happens when financing conditions look survivable.
Banks get paid by the curve when rates rise and by the calendar when they stop. This morning they're getting the second version. **Room three: health care broke its own streak, and the rotation ran backwards.** Two days running I told you health care was the market's last hiding spot — up 0.76% Tuesday, up 0.77% Wednesday, leading both sessions, with the warning attached that a hiding spot everyone finds stops being one. Today: **XLV −0.47%**, the weakest of the major sectors I stamped, on a morning when the broad market rose.
The rotation ran in reverse the moment rates relief arrived, which is exactly what a *defensive* bid is supposed to do — you don't need the fortress when the siege lifts. And within health care, the morning delivered its own brutal reminder of the rule this page repeats: **Ultragenyx fell about 47% on a failed Phase 3.** Not a guidance miss. Not a multiple compression.
A trial that didn't work. In a week where we've watched two profitable companies get fined single digits for beating, a biotech lost half its value in one morning because the data said no. *Biotech's only currency is data* — and the exchange rate is unforgiving in both directions. **Room four, the one I think is mispriced: Nvidia bought the front door.** Buried under Waller and the ISM print: **Nvidia confirmed it is acquiring Hugging Face for $12.93 billion.** Most coverage is filing this as "chipmaker buys AI startup." I'd file it somewhere else entirely. Look at what's actually being bought.
Hugging Face hosts roughly **three million models, one million applications, and 500,000 datasets, serving more than 18 million developers** — on about $150 million of annualized revenue. So the multiple is absurd on revenue and irrelevant as a metric, because revenue isn't the asset. **Distribution is.** Hugging Face is where the world's AI developers go first: it is the default repository, the default download, the habit. Nvidia already owns the compute layer and, through CUDA, the software layer.
It just bought the *discovery* layer — the place where a developer chooses which model to run before they've decided what to run it on. For the frameworks this desk uses, that's the data-moat editorial's three-question test answered in a single transaction: does the company own data others can't replicate (three million community-contributed models), is that data load-bearing in someone's workflow (it's the default import statement in half the AI code written this year), and is it the system of record for actions (the download, the fine-tune, the deploy)? Yes, yes, and increasingly yes.
Note also the price history — Hugging Face **rejected a $500 million offer from Nvidia in 2023** and just sold for twenty-six times that. Three years of AI adoption, priced. Jensen Huang has pledged the platform stays open, that developers can choose their own models, frameworks, clouds and chips.
I believe he means it — for now — and I'd also point out that the most durable moats in technology have always been open on the surface and gravitational underneath. The question this desk will track: does model availability on that platform stay genuinely hardware-neutral eighteen months from now? That's the tripwire, and it goes on the watchlist without odds until we have a workup. **The undertone, and it's the sour one.** While all of this was happening, the ten o'clock data landed and disagreed with the entire morning. **ISM services printed 55.4 — a beat, with business activity at its highest since November 2022 — and the prices index hit 72.6, the hottest since August 2022, while services employment sat at 47.8, in contraction.** Read that carefully: the biggest part of the economy is *growing faster*, charging *more*, and *hiring less*.
Set it beside manufacturing's 71.1 prices line from Tuesday, and both ISM surveys now show prices above 70 in the same week. Waller's own condition was "if inflation data continues improving." Ninety minutes after he said it, the data declined. So the market this morning is pricing the dove it heard over the data it received.
That's a defensible trade for a day and an uncomfortable one into tomorrow at 8:30. **Housekeeping, at full size.** My colleague's desk graded six published rows this morning and printed two losses — one on a call that HPE's re-scaling would earn it applause (it raised both fiscal years, lifted its 2027 framework from 8–12% to 13–17%, disclosed a $3.5 billion hyperscaler inferencing award, and fell 7.4% anyway), and one on which piece of Broadcom's forward language would govern the reaction (the answer was gross margin guided to roughly 73%, not the FY27 AI target). Both losses are logged. The statute gained a clause worth more than the rows: **re-scaling gets paid only when the crowd hasn't already spent it** — HPE walked in up roughly 120% on the year; Dell walked in having been sold 6.5% that morning.
Same kind of print, opposite reception, and the difference was the audience. One image before the day gets away from us. There's a version of this morning where you read only the headlines — *Fed dove, gold rallies, stocks up* — and conclude the storm passed.
But walk the floor and you find a services survey charging the most for its services since 2022, a labor line in contraction, two of the best earnings reports of the quarter being fined, and the largest chipmaker on earth quietly buying the place where every AI developer starts their day. The storm didn't pass. Somebody just opened an umbrella and everyone decided to believe in it until tomorrow morning.
One sentence moved the board. One print contradicted it. One acquisition changed the map.
Payrolls at 8:30 tomorrow. See you at midday. — Lily **Tickers in play:** NVDA · AVGO · HPE · DELL · MRVL · SNOW · PLTR · NTAP · RARE · XLF · XLV · XLU · IWM · GLD · SOXX · SPY · QQQ --- *This is TrendyVest's analysis and opinion — for informational purposes only, not investment advice or a recommendation to buy or sell any security or commodity. Sources: sector and index stamps per stockanalysis.com, Sept 3 — XLF $58.19 +0.92% (9:40), range $58.00–58.52; XLU $42.85 +0.42% (10:22); IWM $294.81 +0.27% (11:07), range $293.43–296.18; XLV $172.13 −0.47% (9:54) vs. the $172.95 prior close, with Tuesday's +0.76% and Wednesday's +0.77% stamps per Episodes 27 and 28; the Waller "inclined to support" remark, the hike-odds move to ~48.4% (down ~15 points), the 10-year at ~4.75%, gold futures $4,494.70 +1.81%, silver +1.42%, WTI $92.94 +2.12%, Brent $97.45 +1.90%, and the index moves (S&P +0.57%, Dow +0.80%, Nasdaq +0.64%, Russell +1.13%) and movers (SNOW ~+24%, PLTR +8.7%, NTAP −9.4%, RARE −47%) per TheStreet's Sept 3 live blog; the August ISM Services report (PMI 55.4 from 54.1; business activity 61.7, highest since November 2022; new orders 60.9; employment 47.8; prices 72.6, highest since August 2022 and the 111th consecutive month of increases) per the institute's release via PR Newswire, Sept 3; ISM manufacturing prices 71.1 per Tuesday's release as previously cited; the Nvidia–Hugging Face acquisition ($12.93 billion; 3 million models, 1 million applications, 500,000 datasets, 18 million developers; ~$150 million annualized revenue; the rejected $500 million offer in 2023 and the $235 million 2023 round; the Huang open-platform pledge and the Delangue quote) per TechCrunch's Sept 3 report of Nvidia's confirmation; HPE's fiscal Q3 results and raised FY26/FY27 outlook, Broadcom's Q3 results and Q4 guidance, and the six graded rows per this desk's Sept 2–3 fact-checked columns and the master tracker.
Stamps were moving at publication; every number carries its stamp. Do your own research.* *Markets. Tech.
The Edge. Research with receipts.*