The week posed two questions that had been hanging over the market all summer, and it answered both — in opposite directions.
Can AI demand justify the spending? Nvidia answered emphatically yes on Wednesday evening, with a quarter and a forecast that exceeded expectations on every line that mattered. Is the Federal Reserve done raising rates? Kevin Warsh’s answer from Jackson Hole on Friday morning sounded much closer to: not yet.
That tension is what carried into the weekend. Major indices finished the week higher but faded at the end, with the Russell 2000 diverging sharply lower. The S&P 500 closed Friday at 7,711.76, down 0.25% on the day but up roughly 0.5% for the week. The Nasdaq Composite ended at 26,402.42, off 0.52% Friday but up about 0.85% across the five sessions. The Dow finished essentially flat Friday at 53,559.99, up around 0.5% on the week. The Russell 2000 fell 1.37% Friday to 2,973.09 and was the clear weekly laggard.
The market bent. It didn’t break.
Wednesday Morning: Disinflation Stalls
July’s PCE report landed before Wednesday’s open and delivered a message the market has been reluctant to internalize: the disinflation trend has plateaued rather than resumed.
Headline PCE rose 0.2% for the month and held at 3.7% year over year, unchanged from June and slightly above the 3.6% economists had forecast. Core PCE rose 0.2% monthly and 3.3% annually, matching both June’s pace and consensus expectations.
The trajectory is what matters. Before the Iran conflict escalated in late February, headline PCE ran at 2.9%. The energy shock pushed it to a three-year peak of 4.1% by May, with core reaching roughly 3.4%. Since then it has essentially flatlined — 3.7% headline and 3.3% core for two consecutive months. The acute shock stopped worsening; the retreat toward 2% stalled well above it. Headline sits 1.7 percentage points above target, core 1.3 points above.
The composition offered some encouragement and one familiar distortion. Goods prices actually fell 0.1%, led by a 2.7% decline in gasoline and energy-related goods and a 0.9% drop in furnishings. Services rose 0.3% — and within that, financial services and insurance jumped 1.2%, precisely the portfolio-management-fee effect this column flagged earlier in the month. That category tracks the level of the stock market almost mechanically, and the BEA is revising its treatment on September 30 for exactly that reason.
Personal income rose 0.4% and spending 0.2%, both stronger than expected. But adjusted for inflation, real spending was essentially flat, rising less than 0.1%, and the saving rate held at 3.0%. Income growing twice as fast as nominal spending is a household choosing caution — the same picture Walmart described a week earlier.
Wednesday Night: Nvidia’s Answer
Then came the report the entire week had been built around, and it cleared every bar.
Revenue reached $96.22 billion, up 106% year over year and 18% sequentially, against the $92.17 billion consensus. Data Center revenue hit $89.0 billion, up 117% and ahead of the $86.33 billion expected, now representing 92% of the company’s total. Adjusted earnings came in at $2.22 per share versus $2.10 estimated. Gross margin held at 75.0%. It was the fourteenth consecutive quarter Nvidia has exceeded its own guidance.
The forward numbers were stronger still. Third-quarter revenue guidance of $108 billion, plus or minus 2%, beat the $104.2 billion analysts anticipated. And management projected fiscal 2028 revenue growth of approximately 70% — far above street estimates, and explicitly characterized as a supply-constrained outlook rather than a demand-limited one.
Jensen Huang’s framing captured why this mattered beyond the numbers: “AI has reached its inflection point. It’s doing useful work. Its tokens are productive and profitable. Now, compute is revenue.” He added a point that directly addresses the concentration risk investors have been debating: “This time last year, one lab alone was driving the buildout; today, we have a golden age of new AI labs and startups, multiple frontier labs scaling in parallel.”
The segment detail supports that claim. Hyperscale revenue reached $49.0 billion, up 13% sequentially, while the ACIE segment — covering neoclouds, AI natives, enterprises, and sovereign customers — grew to $40.0 billion, up 25% sequentially and 138% year over year. Separately, Amazon announced that AWS will purchase two million Nvidia GPUs and adopt the company’s new Vera CPU.
The Memory Confession
The most consequential disclosure wasn’t the beat. It was what Nvidia said about its own costs.
The company guided gross margin to decline and bottom at 71–72% in the fourth quarter, and named the reason directly. CFO Colette Kress: “We want to be direct about this, rather than let it linger as an open question. Memory scarcity today is being driven in large part by the AI buildout itself.”
More striking, Nvidia disclosed that its supply commitments more than doubled from $119 billion to $279 billion, described as “primarily related to the procurement of memory.”
That is the single strongest confirmation yet of a thesis this column has tracked for weeks. Elon Musk endorsed the view that memory, not compute, is the rate limiter of the agentic era. DRAM spot prices have risen for eighteen consecutive weeks. Micron’s chief business officer told investors 2027 would be tighter than 2026. Nvidia itself raised prices 15% over the prior weekend, citing memory procurement difficulty.
Now the company at the center of the AI economy has committed $279 billion largely to securing memory, and is accepting four points of gross margin compression to do it. When the most profitable company in the supply chain pays that price to lock in supply, the scarcity is not a narrative — it is a line item.
The read-through cuts both ways, though. Nvidia’s memory costs are Micron’s and SK Hynix’s revenue, but they are also the hyperscalers’ costs. Every dollar of that $279 billion eventually appears in someone’s capital expenditure budget, at a moment when free cash flow at several of those buyers has already turned negative.
Technology stocks rallied Thursday on the report.
Friday: Jackson Hole and What the Curve Said
Then Warsh spoke, and gave back much of it.
Delivering his first Jackson Hole keynote as chair at 10 a.m. Friday, Warsh emphasized that inflation remains too high and that restoring price stability will be the Federal Reserve’s predominant focus. The symposium’s official theme was financial innovation and payments. The only sentence anyone traded was that one.
The immediate reaction was a spike across asset classes, followed within minutes by a full reversal and then some. The Nasdaq went from up 0.5% to down 0.5% by the close. September rate-hike odds rose considerably. Gold, which had broken above $4,600 to a three-month high on the debasement trade, fell sharply.
The most informative reaction, though, was in the shape of the yield curve. The 10-year rose roughly 12 basis points. The 30-year initially fell, then finished up about 1.5. That is a bear flattening — and it is worth understanding why it matters.
If markets had disbelieved Warsh, the opposite would have happened. Lost central bank credibility shows up as bear steepening: long-term inflation expectations un-anchor and the 30-year rises faster than the 10-year. That is precisely the pattern that drove the long bond to a nineteen-year high earlier in August, and precisely why the Treasury felt compelled to double its buyback program.
Friday inverted it. The market priced higher near-term policy rates in exchange for better-controlled long-run inflation. Gold’s sharp decline says the same thing from another angle — it is the purest hedge against a central bank losing the plot, and positions in that trade were being unwound.
In other words, Warsh got what his predecessors’ ambiguity had cost the Treasury real money to chase: credibility. He just charged equities for it.
The September Setup
The market now faces a specific configuration heading into a thinner-catalyst stretch.
Both tails have been trimmed. Nvidia removed the “AI demand is a mirage” scenario. Warsh removed the “inflation expectations are un-anchoring” scenario. Citadel’s near-complete liquidation of the Situational Awareness portfolio has largely cleared the forced-selling overhang that distorted August. A repeat of August’s mechanical air pockets looks less likely.
But the base case got modestly worse. Higher September hike odds mean a higher discount rate, and that cost falls most heavily on the longest-duration, highest-multiple assets — which is exactly what has been leading this market. The Russell 2000’s 1.37% Friday decline, against a nearly flat Dow, is that arithmetic in miniature.
The genuine risk is neither hawkishness nor AI disappointment. It is the divergence in the data. Business activity hit a four-year high in August while Walmart posted its weakest comparable sales in six years. The corporate economy is accelerating; the household economy is decelerating. A Fed tightening into that split is not fighting an overheating economy — it is fighting an energy-driven price shock while the consumer absorbs it. Whether that constitutes appropriate vigilance or a policy error depends on data that hasn’t arrived yet.
Which makes September 4’s employment report the most important scheduled event on the horizon. July payrolls fell 23,000 with unemployment at 4.1%. Another weak print, arriving with hike odds elevated, revives the least comfortable narrative available: a central bank tightening into a slowdown.
Add unresolved trade friction — tariffs on roughly $20 billion of Canadian goods, with Canadian retaliation scheduled to begin September 8 — and the inflation picture stays cluttered for at least another month.
Bottom Line
This was a week of genuine answers, which is rarer than it sounds. Nvidia’s $279 billion in supply commitments and 70% growth forecast settled the demand question about as definitively as a single company can. Warsh’s insistence on price stability, validated by a flattening curve and a falling gold price, settled the credibility question.
The problem is that the two answers point in opposite directions for equity prices. Corporate earnings remain exceptional. AI investment remains extraordinary. But the cost of money remains high and could go higher, and hundreds of billions of dollars of data center construction has to be financed at whatever rate the bond market sets.
Nvidia can deliver phenomenal results. AI spending can keep growing. Neither happens in a vacuum. That is the tension that will define September — and the first real test comes Friday morning with the jobs report.
Leave a Reply