Four sessions, four consecutive declines, and then a rally that recovered most of the damage in the last one: the S&P closed the holiday-shortened week down 0.80%, the Dow 1.57% and the Russell 2000 2.41%, with every major index lower. One barrel drove all of it. A Gulf shipping war took Brent up 9.03% and WTI 9.66%, fed a hot August PPI and CPI, and lifted September FOMC hike odds from roughly 60% to 86% — the 2-year adding 25.3 bps against the 10-year’s 18.8 and flattening 2s10s to 34 bps. Friday inverted the mood on a single announcement that Iran would meet Gulf foreign ministers, and the market chose to read the whole week’s inflation as an oil impulse that is now reversing — a bet that the Fed will test on Wednesday and that Monday’s meeting in Oman could break before then.

1. Four Days of Escalation, One Announcement of a Meeting: Brent Runs to $108 and Closes the Week +9%

The core facts:The week opened with Houthi ballistic missiles and drones striking southern Saudi Arabia on Tuesday — Abha, Jazan, Najran and the King Khalid air base — halting operations at several southern energy facilities and wounding 73. That followed US Central Command’s destruction of three Iranian tankers the prior Saturday under an explicitly economic doctrine, and OFAC’s indefinite stay of three Iranian general licences including the bunkering and emergency-repair carve-out. CENTCOM destroyed five more tankers late Tuesday, one near Kharg Island, Iran’s principal export terminal. Iran answered Wednesday with ballistic missiles at the Al-Azraq base in Jordan — Jordan engaged 20 and intercepted 18 — and the IRGC claimed attacks on ten ships in the Strait of Hormuz. By Thursday WTI had settled at $103.06 and Brent at $108.37, both the highest since May 19. Then on Friday Iranian Foreign Ministry spokesman Esmaeil Baghaei said “Plans are underway for a meeting with Gulf neighbors on Monday” — the first GCC-Iran ministerial of the war, set for Salalah, Oman. WTI fell 2.43% to $99.99 and Brent 2.94% to $104.47, ending an eight-session Brent winning run stretching back to August 31. Brent still finished the week +9.03%, WTI +9.66%.

Why it matters:The asymmetry between what caused the rise and what caused the fall is the week’s central fact. Four sessions of confirmed military action — facilities burning, tankers destroyed, missiles crossing a third country’s airspace — added roughly 12% to Brent. One spokesman confirming that a meeting is being planned removed a third of it in a single session, with nothing agreed, no transit resumed, and the same session delivering a Saudi pipeline shutdown and a further Houthi advance in the Bab el-Mandeb. That tells you the premium was never priced on barrels lost; it was priced on the absence of a diplomatic channel, and the mere appearance of one was enough. It also means the week’s equity rally rests on a scheduling announcement. The cross-asset receipts are unambiguous and they point the same way all week: crude rose while equities fell in every escalation session, both yields rose rather than catching a duration bid, and the dollar finished unchanged — the signature of a cost shock, not a flight to safety (see Energy and Vol & Treasuries tables in Section B). Gasoline drove over a third of the August monthly CPI increase, which is how a Gulf shipping war became the binding input into Wednesday’s FOMC.

What to watch:Monday September 14 in Salalah — whether the ministerial convenes at all, and whether Iraq attends. A collapse or postponement puts the week’s 9% crude premium straight back on, hours before the FOMC convenes on Tuesday morning.

↑ back to summary

2. September Hike Odds Go From 60% to 86% in Four Sessions — and the Market Rallied on the Print That Sealed It

The core facts:The repricing ran the length of the week and accelerated at the end of it. CME FedWatch odds of a 25 bp hike at the September 15-16 FOMC stood near 60% on Monday, reached 61.4% Wednesday, roughly 70% Thursday after a hot August PPI (+0.4% MoM, +5.4% YoY, energy +4.2%, diesel +24.1%), and 86% Friday on an August core CPI print of +0.3% MoM against a 0.2% consensus. Polymarket’s 2026 hike contract moved from 72% to 89% across the week, an 11-point jump on Friday alone. The 2-year yield added 25.3 bps to 4.630% and the 10-year 18.8 bps to 4.972%, flattening 2s10s from 40.7 bps to 34.2. A 30-year auction cleared at 5.308% against 5.216% prior, and Treasury ran its first long-end buyback above the $4 billion floor. Economists switched calls on Friday: EY-Parthenon’s Greg Daco said “We are changing our Fed call from a hold to a 25bps hike,” and BMO’s Ian Lyngen said the report “clears the path.” Fed officials were in blackout from September 5 through 17 and could not respond. Yet Friday’s equity session was broadly higher — S&P +0.86%, Dow +0.98% — with the VIX down 11.21%.

Why it matters:A market that rallies into a near-certain hike is not disagreeing with the hike. It is making a claim about what the inflation is: an oil impulse, and a reversing one. That claim is testable on Monday and it is the single largest positioning risk into next week — if the Salalah meeting fails and crude resumes, equities have priced away an inflation problem they have not solved, into a Fed that will already have tightened. Two pieces of the week’s tape corroborate the oil reading rather than a benign one. Precious and industrial metals crashed together on Thursday — platinum −7.09%, silver −6.72%, copper −5.33% — during a war escalation that would ordinarily bid gold, which identifies the real-rate path rather than fear as the dominant driver (see Commodities table in Section B). And the front end led the whole move, flattening the curve into an energy shock instead of steepening it. The uncomfortable arithmetic is what a 25 bp hike does from here: 2s10s at 34.2 bps leaves very little room, and the Committee arrives holding evidence that household inflation expectations have come unanchored.

What to watch:The decision at 14:00 ET on Wednesday September 16 with a Summary of Economic Projections. The dot plot, not the hike, is the information — it says whether 86% odds bought a one-and-done or the first of a sequence. Watch the 2-year through 4.75%.

↑ back to summary

3. Oracle Guides $90–95 Billion of Capex and Falls; Dell Rises 11.98% to a Record for Receiving It

The core facts:The trade ran in two halves on consecutive sessions. On Thursday the entire AI-capex chain was sold into Oracle’s fiscal Q1 report — Lam Research −5.65%, Intel −5.63%, Oracle −5.38%, Dell −5.35%, Micron −4.66%, with AMD −3.36% and Nvidia −2.37%, dragging the Nasdaq 100 down 1.08% on scrutiny of Oracle’s negative free cash flow and backlog concentration, with options pricing an 11% post-earnings move. Oracle then beat on both lines and, on the post-close call, CFO Hilary Maxson guided full-year capital expenditure to $90–95 billion and said the spending would flow to vendors supplying AI racks, cooling systems and networking equipment, naming Dell and HPE specifically. Friday inverted the Thursday trade: Dell +11.98% to a record $567.29, its largest single session of the year and the S&P’s top gainer; HPE +10.70%; HP Inc +10.13% to a 52-week high; Super Micro higher. Networking followed — Arista +5.61%, Amphenol +4.57%, Cisco +4.37%, Marvell +4.03%. Oracle itself fell 1.74%. BMO cut its Oracle target to $195 from $220 on sequential cloud gross-margin decline and RBC to $165 from $190 on heavy data-centre capex; RBC initiated Dell at Outperform, $640. Industry-wide hyperscaler capex is estimated at $775–800 billion for 2026, roughly 64% above 2025.

Why it matters:The same announcement moved one company down and three others up 10% or more, which is the market drawing a line through the AI trade that did not exist a week ago: it will pay a high multiple for the revenue and charge a penalty for the capital intensity that produces it. Oracle is now the chosen expression of that penalty, and it is being charged for spending its suppliers are being rewarded for receiving. Two things make this a durable positioning fact rather than a two-day curiosity. First, the hardware complex is levered to a single variable — the persistence of a spending plan whose own author is being punished for announcing it. Second, the split is not only buyer-versus-vendor but layer-versus-layer within the vendors: on the same Friday, SanDisk fell 3.50% on a DeepSeek claim that its new model needs a quarter of the memory and an eighth of the storage, and Seagate fell roughly 4%. Demand for racks and demand for the bits inside them decoupled in the same session. Racks and networking took the capital; memory took the efficiency risk (see weekly movers tables in Section B, where nine of the ten largest gainers are in this chain).

What to watch:Oracle’s cloud infrastructure gross margin next quarter. BMO has said explicitly that triple-digit infrastructure growth alongside stabilising margins would strengthen its thesis; a second sequential decline turns a vendor windfall into a capex-plan risk, and the vendors are now priced for the plan.

↑ back to summary

4. A Drug Class Fails Its First Outcomes Trial and Takes the Whole Sector Down — Amgen −15% on the Week, Healthcare the Worst Sector

The core facts:Novartis disclosed after Friday September 4’s close that pelacarsen — the first cardiovascular outcomes trial of an Lp(a)-lowering drug — missed its primary endpoint, while confirming that Lp(a) levels were successfully lowered. US markets had no session to react until Tuesday, and when they did, the read-through hit harder than the sponsor. Amgen fell 10.08% to $393.17 on roughly three times average volume — reported as its steepest single session since October 2000 — as investors marked down olpasiran, its own Lp(a) candidate whose Phase 3 data is not due until 2027 or later. Novartis then posted a second Phase III failure in five days on Tuesday, the HARBOR trial of del-desiran in myotonic dystrophy, and fell 13.93% to $137.70 on a $294.8 billion market capitalisation. BMO cut Amgen to Market Perform with its $450 target intact; HSBC cut to Hold on Wednesday evening. Amgen’s own positive Phase 3 DeLLphi-305 lung-cancer readout, released that same Tuesday morning, was ignored. By Friday, Amgen had lost 15.03% on the week and Healthcare was the worst of the eleven sectors at −4.70%, with Morgan Stanley additionally cutting Novo Nordisk to Underweight on semaglutide concentration.

Why it matters:This is the only story of the week that owes nothing to oil or to the Fed, and it produced the largest single loss in the mega-cap universe. The reason it travelled so far is the shape of the failure: the mechanism worked and the outcomes did not follow, which does not admit the usual wrong-dose, wrong-patients rescue and impeaches the hypothesis underneath the entire Lp(a) field rather than one molecule. What the weekly view adds is the measure of how far the contagion actually reached, and it reached further than the class. Five of the ten worst weekly movers in the market — Amgen, Merck, UnitedHealth, Johnson & Johnson and Lilly — are Healthcare names, and Merck and J&J fell through fresh target increases from Guggenheim and HSBC. A sector selling names whose analysts are raising numbers is not processing one trial result; it is repricing late-stage pipeline productivity generally, a question that had gone quiet while Healthcare ran +19.08% over twelve months. Note what this is not: a valuation story. Healthcare is still positive on every horizon beyond one month.

What to watch:Whether Amgen restates olpasiran’s Phase 3 design, endpoint or timeline, and whether Lilly comments on lepodisiran, which sits in the same class. The congress at which Novartis releases the full pelacarsen dataset is the offsetting catalyst — subgroup detail decides whether the category is dead or merely narrowed.

↑ back to summary

5. Saudi Arabia Loses Its Hormuz Bypass and Reports Output Down 1.9 Million Barrels — the Physical Story the Tape Ignored

The core facts:Two disclosures landed 24 hours apart and both concern supply rather than sentiment. On Thursday, in a direct communication published with OPEC’s September Monthly Oil Market Report, Saudi Arabia told the group its August crude output had fallen 1.9 million barrels a day to 6.238 mb/d; the same report cut OPEC’s 2026 world demand growth forecast to 380,000 b/d, a fifth consecutive downward revision. On Friday the Saudi Ministry of Energy announced the East-West Pipeline “was shut down as a precautionary measure” after multiple attacks on pump stations in the Riyadh and Madinah regions. The line runs roughly 1,200 km from the Eastern Province fields to Yanbu on the Red Sea, has a full pumping capacity of about 7 million barrels a day, and is the Kingdom’s only export route that does not transit the Strait of Hormuz. The Foreign Ministry blamed drones launched from Iraqi territory and said it would hold off retaliation at the request of Iraq’s prime minister; attribution is contested, with earlier coverage pointing to the Houthis, and no restart timeline has been given. The IEA’s September report, also Friday, cut 2026 global demand a further 940 kb/d to a 2.5 mb/d decline, put world supply 5.7 mb/d lower year on year with more than 10 mb/d of Gulf output shut in, and recorded 507 mb of cumulative stock draws since February. Crude fell on both days.

Why it matters:Story #1 is what the market traded; this is what actually happened to the oil system, and the gap between them is the thing to carry into next week. Since Iran’s effective closure of Hormuz, the East-West line has been the mechanism by which Saudi barrels reach a customer at all — the workaround, not a supplement. With it down, Saudi export optionality collapses back onto the chokepoint the pipeline was built to avoid, while the Houthis have taken Mokha at the other end. Both outlets are under pressure simultaneously, and the market marked oil lower on the session it learned this. The buffer arithmetic is what makes it consequential: with a tenth of world supply shut in and seven consecutive months of inventory draws, there is very little left to absorb a further disruption, and the IEA has now deferred the Middle East supply recovery to 2027 — which removes the argument that this is a spike to look through. For US portfolios the transmission is refined product rather than crude, and it is already visible: retail diesel set successive all-time records through the week, reaching $6.0556 a gallon on Friday.

What to watch:A Saudi announcement of a pipeline restart timeline, and whether Riyadh’s deliberate non-retaliation survives a second strike. Watch the Brent-Dubai spread and US diesel cracks rather than headline crude, and the October IEA report for whether the 2026 demand cut deepens a second time.

↑ back to summary

6. Two Chip Makers Raise Multi-Year Revenue Targets in One Session and Finish the Week as the Market’s Two Best Mega-Caps

The core facts:Tuesday produced two independent upgrades to multi-year AI silicon demand. Marvell chairman and CEO Matt Murphy told CNBC that evening the company now expects roughly $12 billion of revenue this year and $18 billion in fiscal 2027, with more than $15 billion of next year’s total from data centres against roughly $2 billion in 2023 — taking the combined two-year outlook to $30 billion from the $23.5 billion guided in December. He described Marvell as working “across GPU and XPU platforms” rather than being tied to one. The same day AMD’s chief financial officer raised the company’s 2030 total addressable market to $2–3 trillion from $2 trillion at Citi’s technology conference, and CLSA lifted its price target to $710 from $575 citing MI-455 GPU demand and pricing. Separately on Tuesday, Qualcomm disclosed a multi-generational inference-silicon collaboration with Amazon and issued AWS warrants over 25,000,000 shares at $161.26 expiring in 2036, vesting in tranches against purchase orders up to a maximum $60 billion of payments — of which only about 3.75 million shares vested on issuance. Across the full week AMD closed +13.15% and Marvell +13.06%, the two largest gainers above $200 billion; Intel was third at +12.29%.

Why it matters:Tuesday is the session that gives this its weight. Crude broke higher, ten of eleven sectors fell and the Russell dropped — and Technology finished at −0.05% with the Nasdaq 100 at −0.29%, the two best showings on the board after Energy. A macro shock that leaves one complex flat is telling you that complex is priced off a different variable, and two multi-year guidance raises in a single evening are that variable. The composition matters more than the headlines: data-centre revenue going from roughly $2 billion to more than $15 billion in four years is a change in what Marvell is, not a cyclical upturn in what it was. The Qualcomm-Amazon structure is the other half of the same story and the more repeatable one — a hyperscaler committing to a second merchant inference source and paying in equity rather than cash, aligning the supplier’s upside with its own purchasing. That instrument is becoming standard in this cycle, and the discipline point is that $60 billion is a ceiling stretching to 2036, not an order book: the stock opened +7% and closed +3.17%, which is shareholders saying exactly that.

What to watch:The first Qualcomm warrant tranche vesting beyond the initial 3.75 million shares — the only observable confirmation that binding orders are being placed against the ceiling. And whether semiconductor-equipment names close the gap to the AI-silicon names: Applied Materials and KLA made the weekly top ten, but both sit deeply negative over one and three months, so the market is still paying for silicon demand rather than a broad capex cycle.

↑ back to summary

7. Section 338 Goes From “No Established Practice” to a Customs Entry Procedure in Four Sessions — and the Canadian Bans Replace the Tariffs Rather Than Stack on Them

The core facts:The week moved a dormant 1930 statute through three stages. Tuesday: Canada’s counter-tariffs on US$27.6 billion of US goods took effect at 12:01, duties of 15–50% across steel, aluminium, dairy, appliances, agricultural equipment and more, framed as a dollar-for-dollar response to US Section 338 duties — and the same day the USITC opened Investigation No. MISC-053 conceding it “does not currently have an established practice for identifying applicable discriminatory actions,” asking the public how it should gather information and naming “the risk… of direct or indirect reprisal from a foreign government.” Wednesday: five proclamations signed Tuesday evening prohibited importation of Canadian malt beer, wines, cider, whiskies, vodka, spirits, non-alcoholic beer, whey, molasses and larger motorcycles effective September 29 — and the bans replace the 50% tariffs rather than stacking, with goods imported but not consumed before September 29 remaining dutiable at 50%. The President also directed removal of Canadian-origin products from federal procurement schedules covering $50 billion, with no effective date. Friday: CBP issued bulletin CSMS #69851916 at 16:49, the first operational entry guidance, confirming duties apply from 12:01 ET September 15, adding 122 HTSUS classifications and deleting nine, with drawback available. Proclamation 11064 states the Section 338 duties apply “in addition to” Section 232 duties.

Why it matters:The replace-rather-than-stack construction inverts the read for anyone modelling this as a tariff. A 50% duty is a price; a prohibition is a quantity of zero. Substituting the second for the first eliminates the revenue and the trade together, so this is not an escalation of one instrument but a switch to a different one — and it is the instrument whose administering agency conceded in the same week that it has no practice for operating it. That sequencing is the structural risk: the legal architecture underneath the current regime is being built retrospectively, which is exactly the condition in which litigation risk accumulates, and the February IEEPA reversal is the template. The nearer-term exposure is mechanical. Two dates now differ in kind — September 15 is a scope change at existing rates, softened by drawback; September 29 is when the bans bite, creating a fourteen-day pull-forward window with a hard edge and a considerable customs load inside it. The gap the CBP bulletin conspicuously does not address is Section 232 stacking and USMCA treatment on vehicles, which is a live and unquantified cost for every North American assembler two weeks out.

What to watch:A follow-up CSMS bulletin addressing Section 232 stacking and USMCA treatment before September 29 — its absence, not its content, would signal the bans are being allowed to bite at full rate. Also the MISC-053 comment deadline and who files: a docket dominated by steel, aluminium and agricultural petitioners would signal the next wave of country actions.

↑ back to summary

8. The ECB Hikes Into the Energy Shock Six Days Before the Fed Decides — and Lagarde Warns It Could Intensify

The core facts:The European Central Bank raised its deposit rate 25 bps to 2.50% on Thursday, its second increase in three months, citing the energy shock from the Iran war after euro-area inflation reached 3.3% in August — a three-year high and a sixth consecutive month above target. Christine Lagarde told the press conference the energy shock could intensify further and that second-round effects on other prices and wages could be larger than previously expected. ECB staff held the 2026 inflation projection at 3.0% but raised 2027 to 2.5% and 2028 to 2.1%. Bloomberg reported Lagarde called the hike a “no brainer” as markets bet on more. The global long end sold off in step, with the US 10-year adding 13.3 bps the same session.

Why it matters:A major central bank tightening into a supply shock rather than looking through it is the strongest external evidence available for what the Fed does on Wednesday, and it arrived in the same 24 hours that took US hike odds to roughly 70%. The signal is not the 25 basis points; it is the framing. The central bank closest to the energy disruption — Europe absorbed the sharpest gas move of the week, Dutch TTF +12.37% against Henry Hub’s decline — has decided this is an inflation problem rather than a growth one, and has raised its 2027 and 2028 projections to say so. The mechanical consequence shows up in a place worth noting: the dollar index finished the week unchanged at 99.13 despite 25 bps being added to the US front end, because both sides tightened and the differential did not move (see Vol & Treasuries table in Section B). That removes the usual dollar offset to higher US yields, which matters for anyone who was relying on currency to hedge a rates view.

What to watch:Whether the Fed’s September 16 statement follows the ECB in describing the energy shock as a persistent inflation risk rather than a transitory one. That single characterisation, more than the rate itself, sets the 2027 path.

↑ back to summary

9. Enterprise Software De-Rates on Agentic AI While the Hardware It Runs On Rips — and Analyst Upgrades Stopped Working

The core facts:OpenAI shipped GPT-6 Astra on September 3 with a staged rollout that completed across the closed Labor Day weekend — Microsoft’s Foundry catalogue marked it generally available at 17:02 Pacific on September 7, hours before Tuesday’s open. Its headline capability is “computer use,” navigating software as a person would. Tuesday, the first session able to price it, marked the complex down: Salesforce −3.90%, ServiceNow −4.99%, Shopify −7.57%, Accenture −4.12%, Adobe −3.47%, Intuit around −5%. Several fell on days their price targets were raised — Shopify’s lifted to $180 by Piper Sandler, ServiceNow’s to $170 by BTIG. The pattern repeated on Friday in cybersecurity: Wedbush reset coverage, cutting Fortinet to Neutral while raising its target to $155 from $125 on a roughly 100% year-to-date gain, and resumed Outperform on CrowdStrike at $250, Palo Alto at $400, Zscaler at $215 and Datadog at $275 — whereupon CrowdStrike fell 1.82%, Palo Alto 2.32%, Datadog 0.81% and Zscaler 0.03%, on a day Technology rose 1.12%. Salesforce closed the week −6.32%, the fourth-largest mega-cap decline.

Why it matters:The recurring tell across the week is that price and analyst opinion moved in opposite directions in two different subsectors, on two separate occasions, five sessions apart. When a stock falls 7.57% on the day its target is raised, and four more fall on the day they are resumed at Outperform, the sell side and the tape are pricing different horizons — and the tape is pricing displacement risk no broker model yet contains. That is a positioning signal rather than a valuation one, and it has a clean counterpart in the same week: Meta rose 6.12% and made the weekly top five on the strength of Muse, a paid agent, plus a JPMorgan upgrade recasting its AI spending as a revenue line. Agents are being priced as an opportunity for whoever sells them and a threat to whoever sells the workflow software they operate. The rotation completed on Friday, when capital moved decisively into AI hardware (story #3) and the software and security complex underperformed a rising tech tape — an opex story losing to a capex story, heading into a Fed that is about to tighten.

What to watch:Whether the security group keeps lagging on up-days for the Nasdaq — two or three more sessions of that pattern confirms a rotation out of software rather than a single-day funding move into hardware. And pricing or usage disclosures for the Meta Model API, the first evidence of whether the non-advertising revenue now being underwritten exists at scale.

↑ back to summary

10. Alphabet Buys 22 Years of a Finnish Reactor’s Output — the AI Constraint Moves From Transistors to Electrons

The core facts:Announced in the European morning on Wednesday, Alphabet committed €13 billion — about $15.1 billion — to Finnish AI infrastructure over two years: new data centres at Kajaani, Muhos and Vaala, expansion at Hamina, plus grid upgrades, clean energy projects and battery infrastructure. Ruth Porat, Alphabet and Google president and chief investment officer, described it as the company’s largest single investment in Europe. In the same announcement Google signed a 22-year power purchase agreement with Fortum for up to 50% of the output of the Loviisa nuclear plant — from 2028 at reduced capacity and the full 50% for 2030 through 2049. Loviisa runs two VVER-440 units of 507MW each, employs around 580 people and supplies roughly 10% of Finland’s electricity; the contract underwrites approximately €1 billion of Fortum life-extension capital expenditure and, on Fortum’s account, is what allows the plant to run past 2030 at all. It is Google’s first nuclear agreement outside the United States. Fortum shares rose; Alphabet fell on the session, GOOGL −2.28% and GOOG −2.09%.

Why it matters:The power contract is the consequential half and the half that gets copied. A 22-year offtake for half a reactor is not a procurement decision, it is a financing decision: it converts a plant with no economic case beyond 2030 into one with a funded life to 2050, which means a hyperscaler’s balance sheet is now the reason a piece of European baseload generation continues to exist. That is a different relationship between compute and the grid than the renewables PPAs of the past decade, where the buyer took output from an asset that was being built anyway. Placed against the week’s other AI stories the point sharpens into a sequence: story #6 says silicon demand is real, story #3 says the racks are being paid for, and this says the binding constraint is increasingly electrons. The week supplied a corroborating discomfort from the other direction: Utilities were the worst sector over six months at −9.07%, and Commerce finalised solar duties of up to 234% on India, Indonesia and Laos on Friday — raising the cost of exactly the generation the buildout is counting on. Note also what a €13 billion two-year commitment with an embedded 22-year power liability is not: a line item that reverses if demand disappoints.

What to watch:Whether Microsoft, Amazon or Meta announce a comparable non-US nuclear offtake in the next two quarters — that is what turns a Finnish deal into a template. And the ITC injury vote on the solar duties on October 14, the single binary event that would void them.

↑ back to summary

1. Both August Price Prints Run Hot on Energy, but the Cores Disagree (BLS, Thu Sep 10 & Fri Sep 11)

What they’re saying:August PPI rose 0.4% MoM, in line with consensus, and 5.4% YoY, with energy prices up 4.2% and diesel alone up 24.1%; core PPI cooled to 0.2% MoM against a 0.3% estimate. Friday’s CPI inverted that split: headline matched at +0.4% MoM and held at 3.4% YoY, while core came in at +0.3% MoM against a 0.2% consensus — a beat on the measure the Fed watches. June’s monthly pace had been 0.1%.

The context:Gasoline drove over a third of the monthly CPI increase, which makes the headline a direct function of the Hormuz shock rather than of domestic demand — and diesel’s 24.1% jump in the PPI is the same shock arriving one stage earlier in freight, agriculture and construction costs. The two cores pointing opposite ways is precisely the energy-versus-underlying dilemma a supply shock creates for a central bank. The market resolved it as hawkish: hike odds moved to 86% from roughly 70%, the 2-year added 8.0 bps on Friday to finish the week 25.3 bps higher, and Polymarket’s 2026 hike contract jumped 11 points in a day (see Polymarket table above and the Vol & Treasuries table in Section B). Two economists switched to a hike call on the print.

What to watch:September CPI in mid-October will be the first print able to show whether the energy pass-through persisted once crude stopped rising. Before that, Wednesday’s FOMC statement language on energy is the read.

2. Consumer Sentiment Collapses to 47.8, Second-Lowest on Record — and Inflation Expectations Come Unanchored (University of Michigan, Fri Sep 11)

What they’re saying:The preliminary September sentiment index fell to 47.8 from 51.7, missing a 51.0 consensus by a wide margin and marking the second-lowest reading in a series beginning in 1952, behind only May 2026. One-year inflation expectations jumped to 4.6% from 4.0%, the highest since June. The collapse was attributed to record-high September gasoline prices and to renewed trade tensions following the Canadian retaliatory measures.

The context:Sentiment itself has been a poor predictor of consumption for three years, and a reading driven by pump prices tends to mean-revert when pump prices do. The expectations series is a different object entirely. The Fed’s case for tolerating an energy shock is conditional on expectations staying anchored — that is the explicit basis on which a supply-driven price rise gets looked through — and a 60 basis point jump is evidence the anchor is dragging, arriving with a hike already 86% priced. It corroborates rather than contradicts the same morning’s hot core CPI: households are both experiencing and anticipating faster price growth. Note the uncomfortable combination for next week — a consumer this pessimistic, facing $4.295 gasoline and $6.06 diesel, going into a tightening, is a demand-destruction path, and it is the same one the IEA marked down global oil consumption for on the same day.

What to watch:The 5-10 year expectations series in the final September reading later this month. A move there, rather than in the one-year, is what would force the Fed’s hand beyond a single hike.

3. The Street Splits Openly on Wednesday’s Decision — From Two More Hikes to No Hike At All (Multiple Houses, Tue Sep 8 – Fri Sep 11)

What they’re saying:Deutsche Bank said on Tuesday the Fed must reverse 2025’s three 25 bp “insurance cuts” with two hikes this year to a 4.1% funds rate, pause through 2027 and hold off cuts until 2028, citing core PCE near a 34-year high. Mohamed El-Erian argued the same day that Treasury and hyperscaler bond issuance rather than Fed policy is driving yields — “the amount of issuance… far exceeds what you can count on in terms of reliable buyers” — and that the Fed should hold. Pantheon Macroeconomics expected Friday’s CPI to keep the Fed on hold through year-end. J.P. Morgan called for a hike; Goldman Sachs called one “very unlikely.” Officials were in blackout from September 5 through 17.

The context:A dispersion this wide six days out is itself the data point, and the blackout is why it went unresolved — no official could arbitrate between the camps, so the argument ran on the sell side and in the pricing. El-Erian’s objection is the one worth holding onto because it is testable and it was tested this week: if issuance rather than policy is setting the long end, a hike tightens conditions without addressing the imbalance. Wednesday’s $39 billion 10-year reopening argued against him, drawing a 2.71 bid-to-cover, the strongest since April 2016, with indirect bidders taking 79.2% at a yield the 10-year had not touched since August 2007. Thursday argued for him, with the 30-year clearing at 5.308% against 5.216% prior and Treasury running its first long-end buyback above the $4 billion floor. Both can be true at different maturities, which is the actual finding.

What to watch:The blackout lifts September 17, the day after the decision. The first post-meeting speeches will show whether the internal split matched the external one, and the dot plot on Wednesday is the closest thing to a count of it.

4. Existing Home Sales Hit a 14-Month Low as the Mortgage Rate Reaches a 15-Month High (NAR / Freddie Mac / MBA, Wed Sep 9 & Thu Sep 10)

What they’re saying:Existing home sales fell 2.0% in August to a seasonally adjusted annual rate of 3.98 million — in line with consensus but a third consecutive monthly decline and the slowest pace in over a year — while the median price rose 1.6% YoY to an all-time August high of $429,100. The MBA’s 30-year fixed rate rose to 6.85% for the week ended September 5 from 6.79%, with applications down 2.7% and the purchase index easing to 157.5; Freddie Mac’s measure reached 6.76%, a 15-month high, against 6.35% a year ago.

The context:Housing is where the week’s rate move lands first and hardest, and the 10-year finished at 4.972% — within 3 bps of 5.00% — so the affordability squeeze recorded in these prints understates where the market now sits. The sell side began writing it into 2027 models during the week, which is the more durable signal: Wells Fargo re-rated US building materials on an “increasingly cautious” 2027 view citing higher interest rates and constrained public budgets, cutting Vulcan to Underweight, while Bernstein and Evercore trimmed Home Depot and Lowe’s on a delayed recovery. Martin Marietta was upgraded in the same note and still fell 1.76%, which says the group is being sold on the rate view rather than on relative calls (see Real Estate at −2.13% on the week in the sector rotation table in Section B).

What to watch:Housing Starts and Building Permits on Thursday September 17, the first housing data after the FOMC, and Lennar’s results on Wednesday hours after the decision — its order book and cancellation rate are the cleanest same-day read on rate transmission.

5. Recession Odds Near 50% at One House While GDPNow Tracks 4.4% and Claims Sit at Cycle Lows (Moody’s / Goldman / Atlanta Fed / DOL / NY Fed, Tue Sep 8 – Thu Sep 10)

What they’re saying:Moody’s Analytics put 12-month US recession odds near 50% on Wednesday and Goldman Sachs raised its own estimate to roughly 30%. Against that, the Atlanta Fed’s GDPNow tracked Q3 real growth at 4.4%, trimmed from 4.7% a week earlier and 4.8% at the start of the month but still more than double trend; initial claims were 206,000 for the week ending September 5 against a 205,000 consensus, with continuing claims at 1,774K. The NY Fed’s August Survey of Consumer Expectations held one-year inflation expectations at 3.6% but showed mean unemployment expectations jumping 1.6 points to 44.4%, the highest since April 2020, even as the perceived probability of losing one’s own job fell to 13.8%.

The context:These signals do not merely differ in degree, they point in opposite directions, and the market sided decisively with the hard data: Polymarket’s recession contract did not move at all, holding at 7% across the week while the hike contract added 17 points. That is a market pricing a policy problem and no growth problem whatsoever — a position the labour data supports and the survey data does not. The NY Fed’s split is the most interesting item in the set, because consumers see a more fragile aggregate labour market while feeling secure in their own jobs, which is the signature of anxiety absorbed from news rather than from experience. Note too what the week added on the wealth side: the Fed’s Z.1 accounts put household net worth at $195.9 trillion, up $12.8 trillion in the quarter, with $10.7 trillion of that from equities — so the buffer supporting consumption is a mark-to-market on the same AI-concentrated index that fell four sessions running.

What to watch:Initial claims on Thursday September 17 for the first labour reading after the decision, and the next GDPNow update for whether the Q3 trim continues. Whether any institutional recession-odds revision follows the hike is the cleaner test of whether 50% was about the Fed or about the oil price.

1. Oracle (ORCL): −5.38% Thu, −1.74% Fri | A 10% EPS Beat, Revenue Up 30%, and the Shares Still Finished Lower

The Numbers:Released AMC at 16:16 ET Thursday September 10, for the quarter ended August 31. Adjusted EPS $1.92 against $1.74 consensus, a 10.41% surprise; GAAP EPS $1.56. Revenue $19.35 billion against $19.13 billion consensus, +1.10% on the estimate and up almost 30% year over year. Full-year capital expenditure guided to $90–95 billion by CFO Hilary Maxson. The share path across the week: −5.38% Thursday into the print, roughly +7% in extended trading that evening, then −1.74% Friday to close at $150.28 on a $432.9 billion market capitalisation.

The Problem/Win:The win is the infrastructure line, which remains the strongest part of the business and answered the demand half of the pre-print debate outright. The problem is beneath the beat, in two places the headline does not reach: a sequential decline in cloud gross margins, and SaaS growth that BMO called somewhat disappointing. Options had priced an 11% post-earnings move; the realised move across Thursday evening and Friday was a round trip to slightly lower. The analyst dispersion is the cleanest measure of genuine disagreement — BMO cut to $195 from $220 while keeping Outperform, RBC to $165 from $190 on heavy data-centre capex and delayed buildouts, UBS moved up to $250 from $245, and BofA, Cantor, Oppenheimer, DA Davidson and Stephens all reiterated unchanged. Six houses, three directions, one set of numbers.

The Ripple:Larger than the print itself, and it ran the other way — Dell +11.98% to a record, HPE +10.70%, HP Inc +10.13%, with Arista, Amphenol, Cisco and Marvell following. Oracle was the only major name in its own capital chain to fall on Friday. Note also the layer that did not participate: SanDisk fell 3.50% and Seagate roughly 4% the same session on a competing efficiency claim, so the spending lifted racks and networking while memory was marked down.

What It Means:Owning Oracle is now a position on gross-margin stabilisation rather than on backlog growth — the growth is not in dispute and is not being paid for. The stock has become the market’s instrument for expressing AI capital-intensity risk, which means its multiple will track capex discipline rather than revenue.

What to watch:Cloud infrastructure gross margin next quarter. BMO has stated explicitly that triple-digit infrastructure growth alongside stabilising margins would strengthen its thesis; a second sequential decline would not, and the vendors now priced off this capex plan would hear it too.

2. Adobe (ADBE): +1.37% Friday | Record Quarter, Raised Outlook, and the Targets Went Both Ways

The Numbers:Released AMC Thursday September 10. Record third-quarter revenue $6.76 billion against $6.69 billion consensus, +0.99%, up 13% year over year; adjusted EPS $6.13 against $6.08, +0.90%; GAAP EPS $4.62. Total ARR exiting the quarter $27.50 billion, record operating cash flow $2.52 billion, roughly 9.5 million shares repurchased. FY2026 targets raised to revenue of $26.576–26.626 billion and non-GAAP EPS of $24.45–24.50, with Q4 guided to $6.80–6.85 billion and $6.30–6.35. Shares closed Friday at $252.23 on a $100.26 billion market capitalisation — back above the $100 billion coverage floor after reading $98.91 billion the previous session.

The Problem/Win:The win is durability: a record quarter, a beat on both lines and a raised full-year range from a company the market has spent eighteen months treating as an AI casualty, delivered in the same week its peers were being marked down on exactly that thesis. The problem is the size of it. Beating by roughly one percent on each line is not a quarter that re-rates a stock, and +1.37% is the market saying so. The analyst reaction divided precisely along that line — BMO raised to $270 from $230, Baird to $250 from $230 and UBS to $255 from $225, while Citi cut to $250 from $301 and JPMorgan to $315 from $340. Estimates converged toward the middle rather than moving as a group, which is itself the verdict. The print also lands a week after Anil Chakravarthy was named to succeed Shantanu Narayen as CEO on December 1, so the raise is the outgoing management’s last full statement on the AI question.

The Ripple:Almost none, and instructively so. Adobe reported the same evening as Oracle and produced no sector move at all, while Oracle’s guidance moved three hardware mega-caps 10% or more. Application software was simply not where capital went: Technology rose 1.12% on Friday on hardware, and the software and cybersecurity complex underperformed it outright.

What It Means:Adobe has now demonstrated across a full quarter that the AI-cannibalisation thesis is not visible in its numbers, and the market has responded by paying it slightly more rather than re-rating it. Execution is no longer the open question; the multiple is. ARR growth, not the EPS beat, is the line that decides whether the discount narrows.

What to watch:Firefly ARR near $300 million and Creative freemium monthly active users above 90 million against the 10.2% FY2026 ending-ARR growth target. Separately, whether the market capitalisation holds above $100 billion — it has crossed that line in both directions across four consecutive sessions and the level now matters for index and mandate screens.

Q2 2026 reporting is complete and Q3 2026 has barely begun — FactSet counts two S&P 500 companies with Q3 actuals and two more scheduled in the coming week. No company above $100 billion in market capitalisation reports on any of the next five business days, the second consecutive session with an empty forward list, and the calendar is correspondingly thin.

Monday, September 14 — largest reporter Kestra Medical Technologies (KMTS), $1.34 billion, AMC. Nothing on the day approaches the coverage threshold. The market’s attention is elsewhere: the Iran-Gulf ministerial in Salalah, the promised Treasury bank designation, and the FOMC convening the following morning.

Tuesday, September 15 — largest reporter Trip.com Group (TCOM), $25.35 billion, AMC, which is both below the threshold and an ADR. Forgent Power Solutions (FPS), $9.69 billion, reports BMO. FOMC day one, and the Section 338 Canada scope changes take effect at 12:01 ET.

Wednesday, September 16 — Lennar (LEN), $19.13 billion, AMC, consensus EPS $1.29 on revenue of $8.32 billion. Below the Section F threshold, but worth flagging for a different reason: a homebuilder reports hours after an FOMC decision priced at 86% for a hike, into a 15-month-high mortgage rate. Its order book and cancellation rate will be the cleanest same-day read available on rate transmission to housing.

Thursday, September 17 — one row on the entire calendar: Innate Pharma (IPHA) ADR, $208.88 million, BMO.

Friday, September 18 — the calendar returns no scheduled reporters at all. The date was requested twice to confirm the empty result.

Q3 2026 earnings season begins in earnest in mid-October, with the large banks first. Until then the macro calendar, not the earnings calendar, is what moves the tape.

1. Does the Salalah meeting actually convene on Monday? The oil complex gave back roughly a third of a week’s supply-shock premium on a spokesman confirming that a GCC-Iran ministerial was being planned. Nothing was agreed and no transit resumed. A postponement or collapse puts 9% of crude premium back on the tape hours before the FOMC convenes — and this is the one event next week that could make Wednesday’s decision look wrong within a day of it being made.

2. If the Fed hikes with 2s10s at 34 bps, what is left of the curve? The front end added 25.3 bps this week against the long end’s 18.8, and another 25 at the two-year leaves almost no room before inversion. Watch whether the dot plot frames this as terminal or as the first of a sequence: the second reading flattens the curve through zero and re-opens a recession conversation that Polymarket, unchanged at 7% all week, is not currently having.

3. Do the AI hardware vendors hold their re-rating without a second capex confirmation? Dell, HPE and HP Inc added 10–12% in one session on one CFO’s guidance, while the company doing the spending fell. Nothing next week can confirm or refute the plan — there are no mega-cap reporters at all — so the group trades on sentiment for a full week. Watch whether memory keeps decoupling from racks after SanDisk’s 3.50% fall on the same day.

4. Which way does Healthcare break after a −4.70% week? Five of the ten worst weekly movers were Healthcare names and two of them fell through fresh target increases. Either the Lp(a) failure is being correctly generalised to late-stage pipeline productivity, in which case a sector still up 19% over twelve months has further to fall, or the selling is indiscriminate and the upgrades were right. No sector catalyst is scheduled, which makes next week a clean read on positioning rather than news.

5. Does the fourteen-day pull-forward window before the September 29 Canadian bans show up in the data? Business Inventories on Wednesday and Import Prices the same morning are the first series able to capture it, and CBP has still not addressed Section 232 stacking or USMCA treatment on vehicles. The absence of a follow-up bulletin, rather than its content, is what would tell you the bans are being allowed to bite at full rate.

WHY THIS CHARTThree of this week’s four candidates were excellent and looked backwards — a labour-market rule clearing its exit line, the global tightening count, the record gap between housing sellers and buyers. This one answers the question the week actually posed and left unresolved: if the inflation is an oil impulse, why did the 10-year keep climbing when oil collapsed 40% earlier this year? Its decomposition — only 15 of the 10-year’s 98 basis point climb is inflation compensation and 83 is a higher real cost of money — is the single most useful frame for reading Wednesday’s FOMC, and it explains in one picture why gold fell during a war escalation and why the equity market’s bet that a reversing barrel fixes the rate path may be the wrong bet.

ORIGINAL CHART ANALYSIS — FROM FRIDAY’S MIBBrent fell nearly 40% from its 31 March peak to 1 July, below its pre-war price, and the 10-year Treasury yield rose 18 basis points (0.18 percentage point) anyway. Two clocks run inside that yield, and only one keeps time with oil. The breakeven, the extra yield charged for expected inflation, swings like a pendulum, moving with crude day to day and giving back its gains: the five-year measure rose 14 basis points into the March peak, fell 28 in the slide and rose 20 on the rebound. The real yield, what inflation-protected Treasuries pay, works like a ratchet, higher in every phase, including 25 basis points while oil collapsed. So of the 10-year’s 98-basis-point climb to 4.95%, its highest close since October 2023, only 15 is inflation compensation; 83 is a higher real cost of money. The evidence points to the Federal Reserve: the two-year yield, the maturity most tied to policy, is up 118 basis points since the war began, more than the 10-year, and futures give better-than-even odds of a quarter-point hike on 16 September. Households live on both clocks. Gasoline swung with the pendulum, from a $4.50 May peak to $3.78 at oil’s low; the 30-year mortgage rode the ratchet, from 5.98% before the war to 6.43% at that low and 6.76% now. A ceasefire could cut the pump price again; the spring’s slide says it would not cut the mortgage.

What it means: for households, the war’s lasting cost is the loan, not the pump. If you’re waiting for a ceasefire to buy a home or refinance, expect gasoline to fall and your mortgage rate not to — this spring showed that. Watch Wednesday’s Fed decision, not the oil headlines. This is wrong if cheaper oil brings the 30-year mortgage rate back under 6%, its pre-war level.