Three Names Carried the Tape

Monday did not feel like a victory lap. Chip stocks wobbled on tariff talk. Treasury Secretary Scott Bessent’s Iran sanctions — billed as an economic D-Day — sat on the calendar beside a strait that is still a wartime waterway. Asia had opened the week treating Nvidia’s print and the sanctions briefing as one price: the discount rate on already-booked compute. Brent was still a $90-class problem at the open. The 30-year had spent the prior week near a 2007 high.

The middle of the week was a waiting room. Yields eased after Treasury’s plan to at least double long-end buybacks to $4 billion per operation took some heat out of the long bond. Wednesday brought the number the Fed actually targets: July PCE held at 3.7 percent year over year, a 65th straight month above 2 percent, with core at 3.3 percent. Real consumer spending was essentially flat. Then, after the close, the megacaps reported.

Nvidia printed $96.2 billion of second-quarter revenue, up 106 percent from a year earlier, against a Street number near $92 billion. Data center was $89.0 billion. The third-quarter guide is $108.0 billion, plus or minus 2 percent, still assuming no data-center compute from China. The line that reset the tape came on the call: fiscal 2028 revenue growth of about 70 percent, against a 44 percent analyst average. Salesforce reported $11.35 billion and said Agentforce annualized revenue had crossed $1.5 billion, up 240 percent. CrowdStrike called the July quarter the best in its history — $1.47 billion of revenue, a record $333 million of net new annual recurring revenue — and said the “Mythos moment” had become mass-market demand for AI security.

Thursday, Nvidia rose 8.7 percent, its first up-day after an earnings release since May 2024. Salesforce gained about 18 percent. CrowdStrike added a double-digit session. Okta, another AI-security print, jumped as well. The Nasdaq closed up 1.57 percent at 26,541.35. The S&P 500 rose 0.72 percent to 7,730.99. That is the week as a headline.

It is not the week as a market.

The Wires Called It Proof the Boom Is Not a Bubble

The dominant explanation wrote itself. Three AI-exposed companies beat, guided higher, and the indexes followed. Nvidia’s 70 percent call, on a base that consensus already puts near $396 billion for fiscal 2027, implies something like $670 billion of revenue the year after — a run-rate that would put the firm ahead of Apple and Alphabet on Wall Street’s own numbers. Salesforce’s print was treated as a rebuttal to the fear that agents would hollow out seat-based software. CrowdStrike’s was treated as proof that the same agents create a security bill large enough to replace the old endpoint story.

Those prints are real. They are also incomplete as a description of Thursday, and they leave Friday’s speech looking like an afterthought. The residual is simpler. If the AI boom had healed the tape, the rest of the S&P would have been in the room. It was not. If the Treasury market had digested last week’s fiscal scare, foreign demand at this week’s 7-year auction would have been ordinary-to-strong, and gold would have faded. Neither happened. If the cut-cycle story were still the regime, Chair Warsh’s refusal to offer forward guidance would have been noise. Instead, September hike odds moved from roughly one-in-three to a coin flip in a morning.

The Index Was an Invitation List of Three

Information technology rose 3.4 percent on Thursday. It was the only one of the S&P 500’s eleven sectors in the green. More than two-thirds of index members fell. The equal-weight S&P 500 declined about 0.3 percent while the cap-weighted index rose 0.72 percent. Consumer staples, energy, and industrials took the worst of it. Charlie Ripley at Allianz said the quiet part: Nvidia restarted the party, and not every sector was invited.

That is concentration, not confirmation. Nvidia alone accounted for nearly all of the S&P’s gain on some calculations. Amazon, Meta, and Alphabet — the other pillars of the same AI-capex story — finished mixed to lower. The software surge was real and narrow: Salesforce, CrowdStrike, Okta. It does not make a factory-floor recovery, a consumer-staples bid, or an industrial restocking cycle. It makes an index that can print green on three tickers and still leave most portfolios looking at red screens.

Thursday’s S&P print was an Nvidia event. Friday’s hike odds are a Warsh event. Neither is breadth.

The Salesforce move deserves a second look for a different reason. For most of this year the stock had been a SaaSpocalypse casualty — the fear that coding agents would pull software production back inside the enterprise and leave seat-based vendors on the wrong side of the ledger. Agentforce ARR above $1.5 billion, and a Claudeforce tie-up with Anthropic announced the same afternoon, is the company arguing that it owns the harness those agents run in, not the seats they replace. A $2.6 billion mark-to-market gain on the Anthropic stake padded net income. That is a genuine business argument. It is also one stock, on one Thursday, in a market that otherwise declined.

CrowdStrike’s argument is more specific, and more checkable. In the April quarter George Kurtz had called the collision of frontier models and cybersecurity the “Mythos moment,” and said AI detection and response would have to cover seven surfaces — data, models, prompts, agents, identities, infrastructure, and the interaction layer — where endpoint detection covered one. This week’s Q2 numbers are the conversion: AIDR ending ARR nearly tripled versus the first quarter alone, and Kurtz said the line can exceed the old EDR business because each person may run on the order of ninety agents. That is not “AI stock go up.” It is a claim about unit economics. It still does not populate the other ten sectors.

Foreign Demand Thinned. Gold Kept the Debt Story.

Last week’s long-bond break was not a hallucination. The 30-year had closed at 5.31 percent, a finish last seen in 2007, even as hike odds were falling on weak sales. Bessent’s buyback enlargement bought an afternoon. It did not recruit a new foreign bid.

Thursday’s $44 billion 7-year note auction looked fine on the cover: a 2.50 bid-to-cover against a 2.49 six-month average, a high yield of 4.512 percent. The allocation is the number that matters. Indirect bidders — the proxy for foreign official and international accounts — took 60.8 percent, against a 65.1 percent average. Direct domestic accounts stepped up. Primary dealers absorbed a slightly larger residual. That is a successful sale with a thinner overseas book, not a squeeze. Similar duration repricing has already shown up in gilts, JGBs, OATs, and Bunds. The United States is not the only sovereign asking who, exactly, wants to own more of this paper at these coupons.

Gold has been running the same thesis in another room. Spot pushed through $4,630 this week and printed a three-month high near $4,670 on Monday, still well off the January record near $5,600, on a softer dollar and — in the dealers’ own language — concern about government debt levels. The metal rallied even as hotter PCE revived hike talk. That is not a classic real-yield trade. It is a fiscal-credibility trade. When bullion and a weak indirect bid agree, the bond-market worry is not a Treasury-desk local.

The rest of the world’s rate complex did not wait for Jackson Hole. The Bank of Korea raised its base rate a second consecutive time, to 3 percent, into a chip tape that Nvidia had just blessed. Brazil’s Selic still sits at 14 percent after a fourth quarter-point cut on August 5, against a federal-funds range of 3.50 to 3.75 percent. São Paulo printed a 5.3 percent jobless rate. That differential is the carry trade most U.S. readers never connect to a Fed speech: a September hike in Washington compresses the spread that has been paying people to own the real. It is the international channel for a domestic pivot.

Oil, for once, was not the week’s villain. Brent slid toward the high $80s and was on track for a roughly 5 percent weekly decline as Qatar’s prime minister pressed Tehran on Hormuz navigation and Iran said it would name conditions for a reopening. A tanker strike near Khasab on Thursday was a reminder that “diplomacy” and “cleared” are not the same as a peacetime strait. The inflation input eased anyway. That made Warsh’s hawkishness harder to blame on crude, and easier to read as a judgment on the underlying PCE trend.

Warsh Did Not Invent the Hike. He Confirmed the Pivot.

The cut-cycle story died earlier than Friday. In July the FOMC held at 3.50 to 3.75 percent on a 9–3 vote. Beth Hammack, Neel Kashkari, and Lorie Logan wanted 25 basis points more. The minutes, released last week, showed sympathy beyond those three: many participants saw tightening as likely if inflation did not decline. Warsh has also starved the old signal channel — statements near 130 words, forward guidance gone — so speeches and the dot plot do more of the work. Markets that spent the spring pricing cuts have spent the summer pricing whether the next move is up.

Friday’s Jackson Hole keynote did not name a date. It named a standard. “We must be confident that underlying inflation is moving to our objective, clearly and at sufficient speed. Otherwise, we have work to do.” He said summer’s better-than-expected readings “do not tell me that underlying trends have meaningfully improved.” He reminded the room that 65 months of elevated inflation sit with the central bank, that more than half of PCE components still run above 3 percent, and that credit markets show few signs of restraint at the current funds rate. He also said, explicitly, that the remarks were not forward guidance and not a reaction function. Traders treated them as both. CME FedWatch moved a September hike from about 36 percent on Thursday to the mid-50s, with some desks near 60 percent. The two-year yield jumped from 4.22 percent toward 4.30 percent.

Accounts that already disagreed on whether a hawkish Warsh lifts or drops yields were quoting different sessions. The July 29 press conference steepened the curve. Bessent’s buybacks produced the August slide. Friday’s move was a front-end reprice of the September meeting, not a new 30-year panic. That distinction matters. Term premium and hike odds can travel together. This week they did not have to.

One more Friday fact, because it punctures the “risk-on forever” read of Thursday: Bloomberg reported that Stripe and Advent International had abandoned a $53 billion pursuit of PayPal. The stock fell 12 to 16 percent. A deal that had been holding a turnaround narrative simply ended. That is CAPITAL reallocating in public, on the same morning the Fed chair told markets to stop looking to him for the next trade.

Payrolls Will Move the Odds He Planted

Labor Day is Monday, September 7 — the week after the data, not a shortened session before it. Next week is a full calendar aimed at the September 15–16 FOMC, a meeting that comes with a Summary of Economic Projections.

Tuesday, September 1: ISM Manufacturing and the July JOLTS openings report. Wednesday, September 2: the Beige Book. Thursday, September 3: ISM Services and weekly jobless claims. Friday, September 4: the August employment situation — nonfarm payrolls, the unemployment rate, wages. Claims already printed a soft 203,000 for the week ended August 22. July payrolls were a 23,000 decline at 4.1 percent unemployment; that print is why hike odds had faded into this week. August payrolls are the first labor test of Warsh’s “work to do” standard.

A firm jobs number with sticky wages would make the coin flip a base case and leave Thursday’s three-name rally looking like a pause in a tightening tape. A second soft payrolls print would argue that the July dissents over-read the labor market, and that Warsh’s rhetoric ran ahead of the incoming data. Watch the unemployment rate and average hourly earnings, not just the headline addition. Watch whether equal-weight starts to participate, or whether Nvidia has to keep carrying the index alone. Watch the next coupon auction’s indirect share, and whether gold fades if the dollar firms on hike odds. Nvidia’s 70 percent call is itself a supply cap — memory, not demand — so a multiple that assumes unconstrained growth is already the wrong multiple.

The week told a clean story at the index level and a split one underneath. Three companies converted AI into cash-flow prints large enough to lift a cap-weighted average. Nine sectors, a weak foreign bid for Treasuries, gold near a three-month high, and a Fed chair who will not bless a cut cycle all stayed on the page. September 4 is when those two stories have to occupy the same number.