The S&P 500 rose 0.49% on the week and the number is close to meaningless: three of eleven sectors advanced, the broad NYSE Composite fell 0.58%, and small caps lost 1.51%. The dominant driver arrived in the last few hours of trading, when Kevin Warsh’s first Jackson Hole keynote as Fed Chair moved September from a hold to a live hike meeting and repriced the two-year 12.2 basis points while the ten-year finished the week unchanged. That leaves markets carrying a tightening path into a week in which a leading manufacturing gauge fell into contraction and a payroll benchmark revision landed roughly 262,000 jobs below consensus — policy and data no longer pointing the same way.

1. Warsh’s Jackson Hole Debut Turns September Into a Live Hike Meeting — a Four-Day Hawkish Build That Detonated in One Friday Morning

The core facts:The week built the case in four instalments before the Chair delivered it. Tuesday, the Board published its July discount-rate minutes: on July 20 two Reserve Bank boards sought 4.00%; nine days later four did, Kansas City and Dallas having joined Cleveland and Minneapolis — one district wider than the 9–3 FOMC dissent disclosed. Wednesday, July PCE printed 3.7% year-over-year against a 3.6% consensus, with core in line at 3.3%. Thursday from Jackson Hole, Cleveland’s Beth Hammack said “now is the time to act,” Kansas City’s Jeff Schmid said he did not see current policy as restricting anything, and Chicago’s Austan Goolsbee said “everybody should be on edge.” Friday at 10:00 ET, Kevin Warsh said the Fed “must be confident that underlying inflation is moving to our objective… Otherwise, we have work to do,” and that the summer’s better readings “do not tell me that underlying trends have meaningfully improved.” The two-year closed at 4.354%, up 12.2 basis points, its largest single-day rise since June 17. CME FedWatch September hike odds moved from roughly 35% to between 46% and 57%; Polymarket’s 2026 hike contract jumped eleven points on the day and twelve on the week, to 68%. The target range is unchanged at 3.50%–3.75%.

Why it matters:The week’s real information is not that the Fed turned hawkish — four district boards and three dissenters had already said so. It is that the market ignored every one of them and then repriced the entire week in ninety minutes. Hammack is a sitting voter who called publicly for immediate action on Thursday and the two-year moved eight tenths of a basis point; the Chair spoke on Friday and it moved 12.2. A market that prices only the Chair is a specific and newly dangerous configuration, because the Chair has just told it he intends to say less (see #6). The transmission was clean and highly selective: gold −3.50% on the week, the Russell 2000 −1.51%, and banks bid on margin expectations, with Financials the third-best sector at +0.95% — see the sector rotation table in Section B. The curve did the rest of the talking, flattening from 49.9 to 37.5 basis points as the front end moved and the ten-year finished the week unchanged. That is tightening priced without any corresponding lift to long-run growth, which is the shape of a policy error being contemplated rather than a recovery being financed.

What to watch:The September 15–16 FOMC, which carries a Summary of Economic Projections, and the August employment report on Friday, September 4 (consensus +45K, unemployment 4.2%). One complication is unpriced: Governor Lisa Cook’s counsel wrote to the White House on Wednesday that there is “no legally cognizable cause” for her removal, so the September vote count is not yet settled.

↑ back to summary

2. The AI Trade Ran a Full Cycle in Five Sessions — Chip Bear Market Monday, One-Sector Melt-Up Thursday, Duration Unwind Friday

The core facts:Monday, Samsung’s record KRW 90–110trn shareholder-return plan still disappointed and took the US memory complex with it — SanDisk −6.49%, Micron −5.85% — with the Philadelphia Semiconductor Index down about 3.5% and roughly 23% below its 52-week high, compounded by reports the administration may let Apple source DRAM and NAND from Chinese suppliers. Tuesday it round-tripped: AMD +4.91% on a Raymond James upgrade to Strong Buy at a $641 target, Marvell +4.84%, while Mizuho quietly cut four memory and equipment targets with every rating maintained. Wednesday, Nvidia reported $96.22bn of revenue against $92.27bn expected, guided Q3 to $108bn against $104.2bn, and disclosed supply commitments more than doubling to $279bn from $119bn — then traded down 1.3% after hours. Thursday it reversed to +8.74%, Technology gained 3.09% and supplied effectively the entire S&P 500 advance of 0.72%, while the NYSE Composite fell 0.38% and eight of eleven sectors closed red. Friday, four of the five largest mega-cap declines came from one group: Lam Research −5.24%, Nvidia −4.56%, KLA −4.48%, Applied Materials −4.34%. Marvell fell 10.28% having beaten on both lines and raised two fiscal years of outlook, on a gross-margin guide of 57.5%–58.5%.

Why it matters:The discovery this week was not that AI capital expenditure is slowing. Nvidia’s $279bn supply commitment is management’s own balance sheet betting the opposite, and it converts forward revenue from a demand forecast into a procurement schedule. The discovery is that AI equities carry a rates beta nobody had been charging for. A 12.2 basis-point move in the two-year transmitted into a 4–5% loss across semiconductor capital equipment on a day the S&P fell 0.25% — the longest-duration cash flows in the index, sold first and hardest. Marvell supplies the second half of the same lesson from a different direction: it grew revenue 37%, raised fiscal 2027 and 2028, and lost a tenth of its value because 90 basis points of gross margin went to custom-silicon mix. Against Nvidia’s 75.0% margin, that gap is the price of building to a hyperscaler’s specification. The market has begun charging for the composition of AI growth, not merely its rate. Note what the weekly closes conceal: Technology finished +1.22% and the Nasdaq 100 +0.43% — see the sector rotation table in Section B — which describes none of the week that actually happened.

What to watch:Broadcom’s fiscal Q3 after the close on Wednesday, September 2, at roughly a $1.75trn market capitalisation — the definitive test of whether custom-silicon margin dilution is a Marvell problem or a category problem. Separately, Politico reported Thursday, citing eight people, that the administration is weighing extending semiconductor duties to laptops, data-center servers and gaming hardware, with January’s data-center exemption possibly scrapped. No rate, no legal authority and no instrument exists yet, which is a reason to size the risk rather than dismiss it.

↑ back to summary

3. The US–Canada Tariff Wall Becomes Two-Way and Dated — C$27.6 Billion of Counter-Duties From September 8

The core facts:Monday, President Trump posted that from January 1, 2027 tariffs on all cars, trucks, automotive parts and steel “will be increased to 50%”; Ford fell about 4%, Stellantis about 4%, General Motors about 2%, and Tesla 3.81% on a stacked recall catalyst. No proclamation or Federal Register action accompanied it. The same morning USTR Jamieson Greer disclosed that the collapsed bilateral package would have “cut tariffs in half on steel, on aluminum” — and US steel equities rallied precisely because that cut was now off the table, Cleveland-Cliffs up roughly 7% and Nucor and Steel Dynamics roughly 4% each. Tuesday, Finance Minister François-Philippe Champagne announced Canada will match “dollar for dollar, rate for rate”: duties of 15%, 25% and 50% across roughly 700 tariff lines covering C$27.6 billion of US imports, effective September 8, with steel and aluminium doubled to 50%, alongside a C$7.5 billion relief package on top of nearly C$25 billion already provided. That answers the 50% US duties on roughly C$28 billion of Canadian goods that took effect August 22. Separately, Bloomberg reported a 7.5% China overcapacity tariff being prepared ahead of the September 24 Xi–Trump summit, which Reuters explicitly could not verify.

Why it matters:Monday was rhetoric aimed at 2027. Tuesday was a foreign government’s signed instrument with a published line schedule taking effect in fourteen days, and the difference between the two is the entire story. What the week then established is how little of it is in the price. Equities rose 0.32% the day Ottawa published the schedule; Industrials closed flat that session and ended the week down 1.49% for reasons that had nothing to do with trade. The only participants visibly marking the cost are the Canadian banks: Royal Bank raised its provision for credit losses 13.5% year over year to C$1.00bn in a quarter earnings rose 11%, and the stock fell 1.29% on a beat. That is a lender provisioning ahead of September 8 rather than after it, and it is the single cleanest evidence anyone has actually priced a two-way 50% wall between the world’s two largest trading partners. Note also the inversion Greer’s disclosure created: US steel now trades as a short position on a US–Canada deal, so every constructive negotiating headline from here is a headwind for the sector rather than a tailwind.

What to watch:September 8, and any CBP implementation guidance in the interim — exclusions, quotas and de minimis carve-outs are the mechanism by which a headline rate becomes an actual cost. Also whether the Section 301 excess-capacity report on China publishes before the September 24 summit, which would convert a single-outlet report into a rule.

↑ back to summary

4. Every Bullish Catalyst Crude Could Get, and Brent Still Fell 5.99% — the Constraint Moved From Barrels to Refining

The core facts:Monday, Treasury launched “Operation Economic Outcast” — five OFAC sectoral determinations, roughly 25 individuals and more than 60 entities and vessels — while designating no major Chinese bank and penalising no third country; Bessent called it a “warning shot.” WTI fell 2.39%. Iran’s new Persian Gulf Strait Authority blacklisted 45 tankers including ADNOC and Bahri vessels, and a Houthi missile set the Bahri VLCC Amzan alight off Yanbu. Tuesday, WTI fell 4.92% and Brent 5.26% as Oman brokered a phased Hormuz corridor framework in Tehran; Ukrainian drones took roughly 290,000 b/d of Russian refining offline overnight. Wednesday, Iran confirmed a temporary seven-mile corridor agreed with Muscat, crude broke 3% lower, then round-tripped the entire move on a Bloomberg report that Moscow is weighing intensified strikes on Kyiv; Kpler counted five Hormuz transits against a ten-day average of fifteen. Thursday, Qatar entered as a third mediator and crude rose 1.61% anyway. Friday, three refineries were hit in one night — Rosneft’s Kuibyshev, Afipsky and Slavneft-YANOS — taking Reuters’ cumulative estimate to at least 17% of Russian refining capacity, with Bloomberg counting 21 strikes in August alone. WTI closed the session down 0.06%. Meanwhile the EIA put the national gasoline average at $4.085 on August 24 and on-highway diesel at $5.652, up 52% on the year, with distillate stocks 13–14% below the five-year seasonal average and refineries running at 97.4% of capacity.

Why it matters:Run four separate supply shocks through a market in five sessions and watch the price fall 4–6%, and the market has told you where the binding constraint is not. Two independent facts explain it. Goldman put Persian Gulf crude and product exports back at 15–16 million b/d, roughly two-thirds of pre-war levels and far above March’s trough, partly through dark crossings and ship-to-ship transfers — which caps flat price even while disruption persists. And drone strikes destroy conversion capacity rather than production, leaving more crude looking for a buyer and less refined product reaching one. The dislocation has migrated to the two places flat price cannot show it: freight, where Saudi-to-China supertanker earnings ran near $647,000 a day against roughly a tenth of that a year ago, and products, where diesel is 52% higher year over year with the US refining system already at 97.4% utilisation and no slack to absorb an outage of its own. For a US portfolio that is the uncomfortable configuration — a headline oil price that flatters the inflation print while the fuel that actually moves freight, agriculture and industry keeps rising. Energy was the second-worst sector on the week at −2.12% and ExxonMobil the third-largest weekly mega-cap decline at −5.09%; see the sector rotation and weekly movers tables in Section B.

What to watch:Diesel and gasoline crack spreads rather than crude flat price — that is where a 17% refining outage transmits. The EIA weekly status report on Wednesday for another distillate draw from an already depleted 103.4 million barrels, and Kpler’s Hormuz transit count against the fifteen-vessel baseline, which is what would validate the corridor as more than paper.

↑ back to summary

5. The Software Disintermediation Question Got Answered Both Ways in 48 Hours — and the Answer Sorted by Who Owns the System of Record

The core facts:Tuesday evening Intuit beat fiscal Q4 EPS by roughly 13% and then guided fiscal 2027 revenue to $23.28–23.51bn against consensus near $23.72bn — 9–10% growth against the 14% delivered in fiscal 2026 — with TurboTax guided to 2–3% against 7%. Wednesday it fell 3.24% and drew eleven price-target cuts in a morning, JPMorgan downgrading and taking its target from $605 to $331. The same day UBS cut SAP from Buy to Neutral while raising its target 23% to EUR 201, on 17 AI agents generally available and 15 ramping against a stated goal of 200 by year-end; the ADR fell 2.42%. Also Wednesday, Salesforce launched Claudeforce, embedding 37 pre-built sales skills directly inside Anthropic’s Claude — the first time it has applied its own suffix to another company’s product. Wednesday evening Salesforce and CrowdStrike both reported. Thursday, Salesforce rose 22.58%, CrowdStrike 20.50%, and Palo Alto Networks 12.83% on the read-through plus reported approaches to Cribl and ClickHouse.

Why it matters:The week ran a controlled experiment on the thesis that conversational AI collapses the value of application front-ends, and the results sorted with unusual cleanliness. What got marked down was software whose value sits in a workflow a model can reproduce — tax preparation, a configuration layer, an agent count that is 16% delivered with four months to run. What got re-rated was software that owns a system of record or a control point: Salesforce’s data, workflow and governance layer, and CrowdStrike’s Falcon Flex consumption vehicle at $2.29bn of ARR growing 101%. Claudeforce is the strategy stated out loud, and read carefully it is a concession as much as a product: Salesforce is not putting Claude into Salesforce, it is putting Salesforce into Claude, accepting that the seat a salesperson occupies may belong to Anthropic and defending the layer beneath it instead. That is a coherent answer to the question and an admission that the question is real. The weekly leaderboard is the receipt — four of the five largest weekly mega-cap gainers are software names, and Technology and Communication Services were the only two sectors to gain more than a point; see the sector rotation and weekly movers tables in Section B.

What to watch:Palo Alto Networks on Tuesday, September 1 and Snowflake on Wednesday, September 2 — the first tests of whether Thursday’s re-rating extends beyond the two companies that reported it. Also whether Microsoft, Workday or ServiceNow announce comparable stack-inside-someone-else’s-assistant integrations, which would make interface concession the sector’s default posture rather than one company’s bet.

↑ back to summary

6. Warsh Retires Forward Guidance as Standing Fed Practice: “It Has Overstayed Its Welcome”

The core facts:In the same Friday keynote, and separately from the inflation message, Warsh announced a change to how the Federal Reserve communicates. Forward guidance “as a regular practice,” he said, “has overstayed its welcome.” He added that “transparency in communications about future policy decisions is not a virtue unto itself,” that “oversharing policy deliberations and overcommitting to future decisions can lead markets, businesses, and households astray,” and that in normal times its role “should be limited and circumscribed.” He declined to name a replacement: “you can call it an outline… you can call it a trail map… just don’t call it forward guidance.” He also rejected mechanical reaction functions, saying he wished the economy were understood precisely enough that a Taylor rule could be relied upon, and reasserted that “money matters” and that central banks should monitor monetary aggregates. The Board’s own 2026 archive contains exactly one Board of Governors speech in all of August — this one — after six clustered between July 13 and July 16.

Why it matters:This outlives the rate decision it shared a podium with. Forward guidance has been a core Fed instrument since 2008, and retiring it as standing practice removes the mechanism through which the committee has smoothed policy surprises for the better part of two decades. The mechanical consequence is that more of the information content of policy now arrives on decision days: realised volatility should rise around FOMC dates and fall between them, which is a specific and tradeable change in the term structure of rate and equity volatility rather than a rhetorical one. The week supplied the demonstration before the announcement explained it — three Fed officials warned on inflation on Thursday, one of them a sitting voter calling for immediate action, and the front end moved eight tenths of a basis point; the Chair spoke on Friday and it moved 12.2. Fewer, larger repricings is the regime that follows. It is graded uncertain rather than bearish because the trade cuts both ways: a committee that commits less can also change course faster without paying a credibility cost, which is precisely the flexibility Warsh claimed. The detail worth holding is the August speech drought — the practice changed six weeks before the speech that named it, so the market has already been operating in this regime without labelling it.

What to watch:The September 16 statement and press conference — specifically whether the forward-looking language is shortened or dropped, and whether the SEP dot plot survives in its current form. Option pricing around the meeting date is where this gets expressed before it gets confirmed.

↑ back to summary

7. Healthcare Was the Week’s Worst Sector, and Four Separate Attacks Each Hit a Different Leg of the Same Valuation

The core facts:Healthcare closed the week down 2.18%, the worst of eleven sectors, against +12.35% over three months and +23.77% over twelve. Wednesday, Eli Lilly fell 3.59% on employer coverage rollbacks for obesity treatments at PepsiCo and Starbucks, compounded by disclosures indicating Q2 outperformance was aided by non-recurring US rebate and discount adjustments; Merck fell 2.14% on the same session despite a Merck–Moderna Phase 3 melanoma win and an FDA sBLA acceptance for ENFLONSIA, and Moderna fell 5.46%. Thursday, Deutsche Bank cut Novo Nordisk from Hold to Sell citing the 2032 US patent cliff on Ozempic and Wegovy, with no price target disclosed, and Moderna launched a $2bn zero-coupon convertible — later upsized to $2.6bn — falling 4.60%. Friday, BioNTech dropped 8.37% after the Genentech-partnered adjuvant colorectal arm of autogene cevumeran was stopped for futility on a numerical overall-survival imbalance; the European Society of Cardiology’s rewritten guidelines gave semaglutide and tirzepatide a Class IIa recommendation in preserved-ejection-fraction heart failure with obesity, and not one affected name moved on it. Eli Lilly finished the week −6.44% and AbbVie −3.58%, two of the ten largest weekly mega-cap declines.

Why it matters:No single item here was large enough to move a sector, and the sector moved anyway — which means the connection is worth naming. Each development attacked a different leg of the same valuation. Employer coverage rollbacks reduce covered lives. A rebate and discount true-up reduces realised net price per script, and does so retrospectively, which lowers the earnings power the last print implied. The Novo downgrade attacks terminal value by putting a date on exclusivity. The BioNTech futility stop attacks the platform premium every mRNA name carries, and it does so on the harder signal — a survival imbalance rather than a simple efficacy miss. Shrinking covered lives and lower net price compound rather than offset. Merck is the diagnostic: a company that delivered a Phase 3 win and a regulatory acceptance on the same day still fell, and when good news cannot lift a name the marginal seller is positioning, not fundamentals. The ESC guideline moving nothing at all confirms it from the other direction. This reads as a crowded trade unwinding at the edges after a strong twelve months, not as a thesis breaking — see the sector rotation table in Section B for the horizon context.

What to watch:The pivotal ESMO presentation on October 23–27 in Madrid, where the Merck–Moderna melanoma data gets full peer scrutiny; further employer or PBM announcements withdrawing GLP-1 coverage; and Lilly’s next disclosure of net price realisation against volume growth in the incretin franchise, which is where the rebate question is settled.

↑ back to summary

8. Meta Pays $16.7 Billion to 29 States and Accepts a Decade-Long Two-Hour Cap on Teen Accounts — and Finishes the Week Up 5.11%

The core facts:Disclosed in a court filing mid-trial on Wednesday. Meta will pay $16.7 billion to resolve claims by 29 state attorneys general that it designed Facebook and Instagram to addict minors, improperly captured data from children and misled the public about platform safety; reporting places aggregate exposure across related matters nearer $18 billion. The case was co-led by California’s Rob Bonta alongside Colorado, New Jersey and Kentucky, and Meta admitted no wrongdoing. The non-monetary terms are the substantive half and run for ten years: teen use of Facebook and Instagram restricted to two hours per day, access blocked entirely from midnight, expanded parental controls and raised age-verification standards. Shares traded a 6.5% intraday range on Wednesday — a $561.95 low against a $598.37 high — on 22.5 million shares against a 14.8 million average, and finished the week up 5.11%, the fourth-largest weekly mega-cap gain. Meta separately narrowed its 2026 capital spending outlook to $130–145 billion from $125–145 billion, with $10.8 billion reclassified into escrow under multi-year infrastructure purchase agreements not releasing until 2028–2030.

Why it matters:The cash is roughly a quarter’s free cash flow and close to beside the point. The decade-long engagement cap is not. A two-hour ceiling and a midnight blackout on the teen cohort is a permanent, court-supervised constraint on impressions in the demographic that seeds every subsequent cohort of users, and it is being imposed on the exact product surface Meta spent three years defending as a First Amendment and Section 230 matter. That defence has now been traded for finality. The market’s week-long verdict was that removing an uncapped legal tail is worth more than the constraint costs — but read the base before reading the reaction: Meta is −12.43% year to date and −23.04% over twelve months, so a 5.11% week is a discounted stock relieved of an overhang rather than a franchise being re-rated. The durable consequence is the template. Twenty-nine states have now established quantified time limits in a signed instrument, and TikTok, Snap and YouTube face substantially the same claims from substantially the same plaintiffs, which converts a company-specific settlement into an industry-wide design constraint with a known price.

What to watch:Whether Snap, Pinterest or Alphabet disclose settlement discussions with the same attorney-general coalition, and Meta’s next 10-Q for the first disclosed estimate of the revenue impact from the teen time limits — the number nobody currently has.

↑ back to summary

9. Advent and Stripe Walk From a $53 Billion PayPal Buyout — and the Stock Falls 12.71% From Above the Rejected Offer

The core facts:Bloomberg reported Friday, with Axios confirming, that the Advent International and Stripe consortium has abandoned its pursuit of PayPal. The offer was $60.50 per share, valuing the company at more than $53 billion. It was made in July, when PayPal traded near historic lows at roughly a $40 billion market capitalisation, and the board rejected it as undervaluing the company without sending a formal reply. PYPL closed at $53.66, down 12.71%, at a $45.90 billion market capitalisation and the session’s worst large-cap decline, having traded as much as 16% lower pre-market. The decisive number sits in the prior session: PayPal closed Thursday at $61.47 — above the offer it had turned down. Three brokers published the same day, all reacting to the collapse: Loop Capital cut its target to $50 from $62 and Mizuho to $51 from $60, both at Hold, while KBW maintained a Buy at $70. Mizuho’s Dan Dolev cited branded-checkout commoditisation, German market share loss and competition from X-Money. Bloomberg reported the buyers could return if circumstances change.

Why it matters:The board was vindicated on price and punished on outcome, and the distance between those two things is the lesson worth carrying. PayPal traded above the rejected offer on Thursday, so on the market’s own evidence the valuation judgement was correct — and the stock lost an eighth of its value the moment the bid disappeared, because what had been supporting the price was the bid rather than the business. Anyone holding a rumoured target should price that distinction explicitly. For the wider payments complex the read is unambiguous and negative: the most credible strategic and financial buyers in the sector examined a $53 billion take-private of the incumbent, and walked, at a moment when leveraged finance was demonstrably wide open — SoftBank launched a $10 billion term facility the same day (see #10). The constraint was not funding. It was the asset.

What to watch:Whether a revised approach materialises at a lower price now that the standalone case has been repriced, and whether PayPal announces a capital-return or strategic response before its next quarterly report.

↑ back to summary

10. The Cost of AI Capital Gets Quoted Twice in One Week — Alibaba Pays an 8.4% Discount, SoftBank Pays SOFR+275

The core facts:Priced into Monday’s session, Alibaba sold 710 million new shares at HK$112.70, raising HK$80 billion (about $10.2 billion) — the largest primary follow-on ever by a Hong Kong-listed issuer and the third-largest globally in 2026 — at an 8.4% discount to the prior close, with 100% of net proceeds earmarked for “full stack” AI: chips, infrastructure, model development and deployment. The Hong Kong line closed down 9.67%, and Michael Burry disclosed in a Sunday Substack post that he had exited entirely on a return-on-invested-capital argument. The same session, SoftBank set out a ¥1 trillion (about $6.3 billion) seven-year retail bond, the largest ever by a Japanese issuer, with an indicative coupon of 4.3%–4.9% and pricing on September 4. Then on Friday SoftBank launched a $10 billion two-year recourse term facility at 275 basis points over SOFR, stepping up 25 basis points if OpenAI has not completed an IPO within twelve months, carrying a parent guarantee and a cash-collateral account with mandatory prepayment tied to the OpenAI valuation; Apollo and Sumitomo Mitsui joined Goldman Sachs, JP Morgan and Mizuho as arrangers. IFR describes it as the first substantial takeout of the $40 billion bridge maturing March 2027.

Why it matters:Two of the largest AI spenders outside the US mega-caps funded themselves inside one week, at opposite ends of the capital structure, and both prices carry information a portfolio can use. Alibaba generates cash and could have issued debt; it chose dilution and the market charged nearly 10% for it, which is the clearest available quotation on what the AI build-out costs an existing shareholder. SoftBank’s terms run the other way and are notably conservative for AI-adjacent paper — full recourse, a parent guarantee, cash collateral, and an IPO step-up that converts a private-company milestone into a contractual cash-flow trigger. Lenders there are financing SoftBank’s balance sheet, not OpenAI’s equity story, and they priced the exit explicitly. Set both against Nvidia’s $279 billion of supply commitments in the same week (see #2) and the pattern is a build-out whose demand side keeps expanding while its funding side becomes visibly more structured and more expensive. Neither transaction is distress. Both mark the end of the cheap-capital phase, and Burry’s reasoning is the part that generalises — a returns-on-capital argument travels across the sector in a way a price target does not.

What to watch:Final pricing on SoftBank’s retail bond on September 4 and where in the 4.3%–4.9% range the coupon lands — the top of the range would say retail demand is thinner than the record size implies. Also whether the remaining roughly $30 billion of the bridge refinances on comparable terms before March 2027, and whether the next large AI-capex equity raise is forced to repeat Alibaba’s discount.

↑ back to summary

1. Headline PCE Accelerates to 3.7% While Core Holds In Line at 3.3%, and Q2 GDP Is Confirmed at 1.5% (BEA, Wed Aug 26)

What they’re saying:Real GDP grew at a 1.5% annualised rate in Q2’s second estimate, unchanged from the advance reading and down from 2.1% in Q1. The PCE price index rose 0.2% in July, lifting the year-over-year rate to 3.7% — a tenth above the 3.6% consensus — while core PCE held in line at 3.3%. Personal income (+0.4%) and spending (+0.2%) both beat estimates, and Q2 corporate profits jumped 8.2%.

The context:The print moved almost nothing on the day — the S&P closed −0.02%, the ten-year added a single basis point and the VIX actually fell 1.55%. That non-reaction was not calm; it was a market that had stopped trading releases directionally and was waiting for the person who decides. Two days later Warsh cited these exact readings as failing to show that underlying trends had meaningfully improved, and the two-year moved 12.2 basis points — see the Volatility & Treasuries table in Section B. The week’s lesson is that the data was necessary and not sufficient: headline inflation at nearly double target supplied the Chair his premise, and core landing in line supplied the market its excuse to ignore him for forty-eight hours.

What to watch:The August core PCE print in late September, against the Cleveland Fed’s 3.40% nowcast. A confirmed acceleration above July’s 3.3% would put the September hike beyond argument.

2. Chicago Business Barometer Craters to 47.1, the Steepest Monthly Drop Since COVID (MNI, Fri Aug 28)

What they’re saying:The MNI Chicago Business Barometer plunged to 47.1 in August from 57.6 in July, badly missing the 58.3 consensus — a 10.5-point month-over-month collapse, the largest single-month drop since the COVID shock and the lowest reading of 2026. A sub-50 print signals outright contraction in Chicago-area manufacturing and business activity.

The context:It landed on the same morning as the Jackson Hole keynote, and every instrument that repriced on Friday repriced for a tighter Fed rather than a weaker economy: the two-year rose 12.2 basis points, the dollar gained 0.51%, gold fell 3.34% and September hike odds went up. A market pricing a manufacturing contraction produces the opposite of all four. The only readings consistent with a growth scare — the Russell 2000’s 1.39% decline and Utilities falling 1.12% despite their defensive character — are equally well explained by the rate move itself. Chicago leads the national ISM, and the size of the miss is what makes the conflict live rather than academic: the Fed signalled tightening in the same week a leading manufacturing gauge went into contraction.

What to watch:ISM Manufacturing PMI on Tuesday, September 1, consensus 55.3. A confirming print anywhere near 50 would put the Fed’s hawkish turn and the manufacturing cycle into open conflict for the first time this cycle.

3. Two Consumer Surveys, One Direction — Expectations at 68.2 Below the Conference Board’s Own Recession Line, Michigan at 51.7 (Conference Board, Tue Aug 25 / University of Michigan, Fri Aug 28)

What they’re saying:Conference Board Consumer Confidence fell to 89.4 in August from a downwardly revised 90.2, missing the 90.2 consensus and marking a second straight monthly decline to a seven-month low. The forward-looking Expectations Index dropped 5.8 points to 68.2 — below the level of 80 the Conference Board itself flags as historically preceding a recession within twelve months — while Present Situation rose 6.8 points to 121.2. On Friday the University of Michigan’s final August sentiment index came in at 51.7, above the 51.0 consensus but down from July’s 55.2: a roughly 6% monthly fall and an 11% drop from a year ago, which survey director Joanne Hsu attributed to persistent worries that inflation will stay elevated.

The context:Both surveys beat or missed on the headline in confusing ways and agree completely on the trend, which is the useful reading. A Present Situation gain against an Expectations collapse describes households who are managing now and expect not to be; Michigan’s beat-against-a-6%-monthly-fall says the same thing with different arithmetic. Corporate confirmation arrived inside the same week: Dick’s Sporting Goods cut every line of full-year guidance on its Foot Locker unit and fell 30.68%, dragging Nike down 3% and Lululemon 4% on no news of their own, and July new home sales fell 10.5% to a 607,000 pace, the weakest since January. What the market did with all of it is the tell — Consumer Cyclical closed the week −0.62% and Consumer Defensive −0.33%, neither a rout, and see the sector rotation table in Section B for how ordinary that ranks.

What to watch:Michigan’s September preliminary reading and its year-ahead inflation expectations, which feed directly into Fed deliberations, plus August retail sales for the first hard confirmation of what both surveys are forecasting.

4. The Labour Market’s Two Faces: Claims Fall to 203K, Then the Benchmark Revision Misses Consensus by Roughly 262,000 Jobs (Dept. of Labor, Thu Aug 27 / BLS, Fri Aug 28)

What they’re saying:Initial jobless claims fell to 203,000 for the week ended August 22, below the 208,000 expected and down from a revised 207,000, extending a run of historically low readings since mid-July’s 189,000; continuing claims eased to 1.778 million from 1.796 million. The next morning the BLS’s preliminary annual benchmark revision lowered the March 2026 nonfarm payroll count by 79,000 (−0.1%), with private-sector employment revised down a larger 178,000, against a Bloomberg consensus that had expected an upward revision of 183,000 — roughly a 262,000-job negative surprise. Last year’s revision subtracted about 911,000.

The context:A benchmark revision landing on the same morning as a hawkish Chair keynote is close to a natural experiment in what this market actually prices, and it returned an unusually clean answer: no rates instrument moved in the direction the data pointed. The two-year rose 12.2 basis points and September hike odds went up. For most of two years soft labour data reliably bid the front end; on Friday it did not move it at all, which says the employment leg of the dual mandate is not the binding constraint right now. That has a positioning consequence — labour disappointments are currently a weaker hedge against rate risk than they have been — and a measurement one. If the March level was overstated, every subsequent monthly print is being measured off a base that is too high, flattering the run rate at precisely the moment the Fed has stopped weighting it.

What to watch:The August employment report on Friday, September 4 — consensus roughly +45,000 payrolls with unemployment expected to rise to 4.2% — and specifically whether a miss moves the two-year at all, which is the direct test of whether Friday’s non-reaction was a one-off or the new regime.

5. The Atlanta Fed’s Q3 Nowcast Turns Back Up to 4.6% Against a Confirmed Q2 of 1.5% — and Cleveland’s Says Inflation Is Re-Accelerating Too (Atlanta Fed / Cleveland Fed, Wed Aug 26)

What they’re saying:GDPNow lifted its Q3 2026 real GDP estimate to 4.6%, from 4.0% on August 18, with the Q3 nowcast for real personal consumption growth rising from 2.5% to 3.1% and real gross private domestic investment from 13.7% to 14.5%. The series had fallen all month — 6.2% on August 3, 5.9% on the 4th, 4.3% on the 14th, 4.0% on the 18th — and then turned. Separately, the Cleveland Fed’s inflation nowcast has August core PCE running at 3.40% year-over-year and CPI at 3.37%, both above the July core PCE of 3.3% that printed the same morning.

The context:Two Reserve Banks’ own models now point at growth and inflation re-accelerating into the September FOMC, which is the single configuration that makes a hike defensible rather than merely feared — and it flatly contradicts the consumer and manufacturing evidence in the boxes above. The gap between a confirmed 1.5% for Q2 and a 4.6% nowcast for Q3 is far too wide to be noise. The honest reading is that GDPNow is being driven by an investment component growing 14.5%, overwhelmingly AI capital expenditure: the same spending Nvidia sized at $279bn of supply commitments that same evening, and the same physical flow that widened July’s goods trade deficit to $118.8bn on an 11.3% jump in capital-goods imports including a 68.7% surge in Korean semiconductor shipments. That is one economy running at two speeds, averaged by a statistic that describes neither — and a portfolio manager should read 4.6% as a statement about datacentre construction rather than about the household.

What to watch:The next GDPNow update and whether 4.6% holds or resumes falling, and whether August capital-goods imports stay near record pace in the advance trade report due mid-September.

1. Salesforce (CRM): +22.39% on the week | Guidance Raised and Agentforce ARR Up 210% — With a $2.6 Billion Mark Inside the EPS Line

The Numbers:Fiscal Q2 2027, released after the close Wednesday, August 26. Revenue $11.35bn against $11.32bn expected, up 11% year over year, with subscription and support revenue $10.8bn, up 12%. Non-GAAP diluted EPS $5.90 against a $3.27 consensus and GAAP EPS $4.29 — both including a $2.6bn gain on strategic investments arising from the company’s Anthropic stake. Full-year FY27 revenue guidance raised to $46.1bn–$46.4bn from $45.9bn–$46.2bn, with fiscal Q3 guided to $11.42bn–$11.50bn. Combined Agentforce and Data Cloud ARR reached nearly $3.9bn, up more than 210% year over year.

The Problem/Win:Separate the two halves. The operating quarter is good rather than spectacular — 11% growth, a $30m revenue beat and a $200m raise to the full-year range — and the EPS headline should be read with the $2.6bn Anthropic mark stripped out, because a mark-to-market on a private position is not operating performance. The genuine win is $3.9bn of agent ARR growing above 210%, which is the first datapoint of real scale showing an incumbent converting AI product into recurring revenue rather than describing it.

The Ripple:Ten firms raised targets — Deutsche Bank to $275, Loop Capital $270, JP Morgan and Mizuho $265, BMO $260, UBS $240, Morgan Stanley $235, Citigroup $233, Wells Fargo $230, Bernstein $195. This was the largest weekly mega-cap gain on the board and the anchor of a software cohort re-rate that also carried CrowdStrike and Palo Alto Networks. Salesforce announced Claudeforce with Anthropic the same evening, with no commercial terms disclosed.

What It Means:An 11% revenue grower re-rating 22% in a week is the market repricing terminal growth rather than the quarter. Read the horizon stack before sizing it: the week leaves the shares still −3.36% year to date and +0.58% over twelve months, so this is a laggard restored to roughly flat, not a leader extending a run. The equity now rests on one disclosed number compounding from a $3.9bn base.

What to watch:Whether the next quarter separates Agentforce ARR from Data Cloud — the combined disclosure makes the growth rate impossible to attribute, and management will be asked — and how much of the fiscal Q3 guide assumes a Claudeforce contribution.

2. NVIDIA (NVDA): +1.32% on the week | A $96 Billion Quarter, a $108 Billion Guide, and Supply Commitments That More Than Doubled to $279 Billion

The Numbers:Fiscal Q2 2027, released after the close Wednesday, August 26. Revenue $96.22bn against $92.27bn expected, a 4.28% beat, up 18% sequentially and 106% year over year from $46.7bn. Adjusted EPS $2.22 versus $2.09; GAAP EPS $2.46 against $2.11. GAAP and non-GAAP gross margins both 75.0%. Data Center revenue $89.0bn, up 117% year over year and now 92% of total company revenue. Q3 guidance $108bn plus or minus 2%, against roughly $104.2bn consensus. Supply commitments more than doubled to $279bn from $119bn the prior quarter, primarily memory procurement. The shares fell about 1.3% in extended trade, rose 8.74% to $227.98 on Thursday, then lost 4.56% to $217.59 on Friday.

The Problem/Win:The line that matters is not on the income statement. A supply-commitment figure that doubles to $279bn is management putting its own balance sheet behind a demand curve it expects to keep bending upward, and it converts forward revenue from a forecast into a procurement schedule. The problem is where the bar now sits: a $3.9bn revenue beat and a $3.8bn guidance raise produced a negative after-hours print, which says expectations have caught up with delivery and the company must beat a trajectory set by itself rather than by analysts.

The Ripple:Thursday’s reversal carried the entire market: Technology closed +3.09% and supplied effectively the whole S&P 500 gain of 0.72%, with Broadcom +4.49% and Intel +4.36%. Twelve firms raised targets, Bernstein to $400 from $315 at the top. The $279bn memory commitment reads straight through to SK Hynix, Micron and Samsung, and a Q3 guide implying roughly $100bn of quarterly Data Center revenue flows into the grid and electrical complex behind it. It also puts a number on the investment component that lifted the Atlanta Fed’s Q3 nowcast to 4.6% — see Section D.

What It Means:Two sessions after the strongest print of the week the stock is up 1.32% on the week, and that is not a verdict on the quarter. It is Friday’s rate move taxing the longest-duration asset in the index. The quarter is unambiguous and the supply commitment is the most credible forward signal available on AI capital expenditure, precisely because it is the company’s own capital at risk against 2027 demand.

What to watch:Whether the 75.0% gross margin holds as the $279bn memory commitment converts into cost of goods, and Broadcom’s report on Wednesday, September 2 as the custom-XPU cross-check on the same demand.

3. CrowdStrike (CRWD): +13.78% on the week | Record Net New ARR of $333 Million, Accelerating to 51% Growth, and a 630 Basis-Point Guidance Raise

The Numbers:Fiscal Q2 2027, released after the close Wednesday, August 26. Revenue $1.47bn against $1.44bn expected, up 26% from $1.17bn a year earlier; adjusted EPS $0.31 versus $0.29. Annual recurring revenue $5.84bn at July 31, up 25% year over year. Net new ARR of $333m was a record and accelerated to 51% year-over-year growth. Operating cash flow $530.3m against $332.8m; free cash flow $377.4m against $283.6m. Falcon Flex ending ARR exceeded $2.29bn, up 101%. Full-year FY27 revenue guidance $5.99bn–$6.01bn against $5.94bn consensus, with FY27 net new ARR growth guidance raised 630 basis points to 34% at the midpoint. Shares rose 20.50% Thursday to $227.96 and gave back 4.19% on Friday.

The Problem/Win:Net new ARR governs this business and it accelerated rather than merely grew — 51% growth in the incremental number on a base already at $5.84bn, delivered alongside a 630 basis-point raise to the full-year growth guide. That is the strongest combination the company can produce. Falcon Flex is the delivery mechanism, and its ARR doubling to $2.29bn is what platform consolidation looks like when it works rather than when it is described.

The Ripple:Nine firms raised targets, Citigroup and RBC to $260 at the top. The read-through made this a sector event rather than a company one: Palo Alto Networks added 12.83% on Thursday partly on this print, and the whole cybersecurity cohort re-rated with it. It is also the week’s sharpest counterexample to the disintermediation case running through Intuit and SAP — security is a software category where AI adoption creates incremental demand instead of substituting for the vendor, and this quarter put a number on that at 101% growth in the flagship consumption vehicle.

What It Means:Unlike Salesforce, CrowdStrike’s week extends an existing run rather than repairing one — +86.36% year to date and +97.65% over twelve months, so 13.78% is momentum continuation, not a snap-back off a laggard. Two headline weeks of similar size, two entirely different positions, and anyone sizing them off the percentage alone will get it wrong.

What to watch:Palo Alto Networks on Tuesday, September 1 — the direct test of whether this is category growth or share taken from a competitor. Falcon Flex attach rates are the metric where the difference would show.

Q2 2026 earnings season is effectively complete at 97% reported. The coming week is a short, concentrated tail: no company above $100 billion reports on Monday, Thursday or Friday, and every name below lands across Tuesday and Wednesday.

Palo Alto Networks (PANW) — AMC, Tuesday, September 1 — consensus $0.98 EPS on $3.35B revenue, $302.85B market cap. Key focus: platformisation attach rates and next-generation security ARR, and management’s posture on the reported approaches to Cribl and ClickHouse that added 12.83% to the stock on Thursday. Jefferies raised its target to $450 from $335 on Friday; shares closed -2.93% today.

Dell Technologies (DELL) — AMC, Tuesday, September 1 — consensus $4.91 EPS on $44.93B revenue, $295.77B market cap. Key focus: AI server backlog conversion and, more pointedly after Marvell, the gross margin earned on it — Dell’s AI systems mix has been the clearest case study in whether AI infrastructure revenue is profitable revenue.

Medtronic (MDT) — BMO, Tuesday, September 1 — consensus $1.39 EPS on $9.55B revenue, $116.77B market cap. Key focus: diabetes and cardiac rhythm segment growth, plus any commentary on the ESC’s rewritten heart failure guidelines, which upgraded durable mechanical circulatory support and left ICD indication expansion unsupported by Friday’s CMR GUIDE result. TD Cowen raised its target to $110 from $100 on Friday; shares closed +1.38% today.

Broadcom (AVGO) — AMC, Wednesday, September 2 — consensus $3.22 EPS on $29.24B revenue, $1,754.55B market cap. Key focus: custom AI accelerator (XPU) revenue and the gross margin attached to it, read directly against Marvell’s 57.5%-58.5% guide. At $1.75 trillion this is the single most consequential print of the week and the definitive test of the custom-silicon margin question.

Snowflake (SNOW) — AMC, Wednesday, September 2 — consensus $0.45 EPS on $1.48B revenue, $113.68B market cap. Key focus: product revenue growth, net revenue retention and consumption trends — the cleanest available read on whether enterprise AI workloads are translating into incremental data platform spend.

Q3 2026 earnings season begins mid-to-late October.

1. Does Tuesday’s ISM confirm Chicago — and if it does, has the Fed already committed? Chicago fell 10.5 points into contraction on the same morning Warsh put September in play, and consensus for ISM sits at 55.3. There is no reading in which both surveys describe the same economy, and the September FOMC arrives carrying a Summary of Economic Projections that has to reconcile them.

2. Broadcom reports Wednesday: is custom-silicon margin dilution a Marvell problem or a category problem? Marvell beat on both lines, raised two fiscal years of outlook, and lost a tenth of its value on a 90 basis-point gross-margin guide. Broadcom is roughly eight times the market capitalisation running the same model, and the answer decides whether the market keeps charging AI names for the composition of their growth rather than its rate.

3. With forward guidance retired, where does the volatility go? If more of the policy signal now arrives on decision days, options struck around September 16 are cheap relative to the weeks either side. A VIX at 14.42 after a 12.2 basis-point move in the two-year suggests nobody has adjusted for that yet.

4. Does anything actually price September 8? Canada’s C$27.6bn counter-tariffs take effect in ten days across roughly 700 lines at 15%, 25% and 50%. The only participants visibly marking the cost this week were Canadian banks provisioning ahead of it; US industrial, agricultural-equipment and appliance names have not moved.

WHY THIS CHARTFour of the week’s five charts examined the Fed or the historical record; this one examined the gap between a headline number and the economy underneath it, which is the exact tension the week went on to resolve. It won because it was published the morning before Warsh spoke and had already named the configuration that made his keynote possible — a 1.5% GDP print concealing 4.2% growth in what American households and businesses actually bought, with core inflation at 3.3% and September priced for a hike rather than a cut.

ORIGINAL CHART ANALYSIS — FROM THURSDAY’S MIBThe headline decelerated. The economy did not. Real GDP grew 1.5% in the second quarter, unchanged in yesterday’s second estimate and down from 2.1% in the first. But GDP counts production, and it deducts imports and swings in inventories. Strip those out and what American households and businesses actually bought grew 4.2% — revised up from 3.9%. Net exports cost the headline 1.14 points, slower stock-building another 0.72. BEA said it plainly: stronger consumer spending, partly offset by an upward revision to imports. The same appetite that lifted demand pulled in the foreign goods that get deducted from it. Count the same economy from income rather than purchases — wages, profits and rents — and it grew 2.2%, against 1.2% in the first quarter. The two measures swapped places, and averaged, as BEA publishes them, the economy went slightly faster: 1.8% against 1.7%. Corporate profits rose $400.9bn, after $74.4bn. Which is why September is priced near 40% for a hike, not a cut, with core inflation at 3.3%, and why Warsh’s Jackson Hole debut tomorrow matters more than a 1.5% print suggests. The recession probability below sits at 7 against a trigger of 67. The one crack: durables, equipment and housing were marked down in the release that marked demand up. Not a slowing economy. A slowing measurement of a fast one. What it means: the number that made headlines yesterday understates what Americans are actually spending, and the Fed knows it. If you are positioned for rate cuts — long-dated bonds, or shares that do well when borrowing gets cheaper — this release argues the other way. Watch durable goods and business equipment: both shrinking in the same quarter, which last happened in 2021, is what would change it.