Saturday, September 12, 2026, written at 9:30 a.m. Eastern. This is a weekend wrap covering the week to Friday’s close, with US markets shut and the Federal Open Market Committee four days away. Everything below is a settled mark rather than a live one.
Top Call: The AI Trade Changed Hands on Friday and Nobody Rang a Bell
Oracle reported the quarter the bulls wanted. Cloud infrastructure revenue rose 121% to $7.4 billion, total revenue rose 30% to $19.3 billion, non-GAAP earnings of $1.92 beat $1.74, remaining performance obligations reached $664 billion, and the company delivered 850 megawatts and over 300,000 GPUs at 97.9% utilisation. The stock opened more than 7% higher and closed at $150.28, down 1.74% on the day.
In the same session Dell rose 11.98% to $567.30, Hewlett Packard Enterprise rose about 12%, Super Micro rose 7%, and Lumentum finished the week as the best performer in the S&P 500 at plus 9.02% with Ciena at plus 4.24%. Those names did not report anything on Friday. They rose because Oracle guided fiscal 2027 capital expenditure to $90 billion to $95 billion, and that line is a cost to Oracle and a revenue line to everyone standing downstream of it.
That is the call. The market has stopped treating hyperscale capital expenditure as evidence of the spender’s strength and started treating it as an order book for the supplier. We flagged the rule on Friday in the form the tape then gave us, that it pays the layer selling compute and discounts the layer consuming it. Friday extended the rule one rung up the ladder. Oracle was on the selling side of that trade for two years, and in one session the market moved it to the buying side. The implication is uncomfortable for anyone holding the AI complex as a single position: within the theme, the operating leverage now sits with servers, networking and optics, and the balance sheet risk sits with the cloud operators funding them.
The Ten Year Is the Other Half of the Same Sentence
The ten year closed the week at 4.974%, the highest since October 2023, with the thirty year at 5.36% after touching 5.373% and the two year near 4.58%. August core CPI at 0.3% against a 0.2% forecast moved September 16 pricing from about 71% on Thursday evening to 88% by Friday’s close, and some calibers put it near 90%.
The wedge from Thursday has not moved. Headline inflation runs 3.4% and core runs 2.4%, so essentially the entire gap is energy, with the energy index up 16.3% over the year and gasoline up 27.4%. Core is now lower than it was in July. What the committee will vote on Wednesday is not a domestic demand problem; it is a transit chokepoint priced in dollars. Our view has not changed and the market’s has: a hike at 88% is done, so the argument is no longer about the move but about the dots, and whether Kevin Warsh uses his first press conference to describe this as a one-off insurance step or the opening of a cycle.
The cleaner way to hold the rates picture is alongside the first section rather than separate from it. Oracle intends to spend up to $95 billion in a year against revenue of at least $90 billion, having produced negative free cash flow of $5 billion on $23 billion of operating cash flow. That gap is funded in credit markets, and it competes there with the Treasury at 4.97%. Every dollar of that guide that thrilled Dell holders on Friday is a dollar Oracle has to borrow into the most expensive long end in nineteen years. The equity market priced both facts correctly on Friday and priced them in opposite directions on the same piece of news, which is the most coherent thing it has done in a month.
Crude Had the Best Week of Anything and Gave Some Back
Brent settled at $104.61, down 2.8% on Friday and up 8.7% on the week after peaking near $108 on Thursday, its highest since May. WTI settled at $100.05, down 2.4% on the day and up 9.4% on the week, holding the hundred dollar handle into the weekend. Energy was the one sector that lagged on Friday and it remains the only durable trend of 2026, up roughly 42% year to date against a market that has gone essentially nowhere. Refiners told the same story from the margin side, with Valero, Phillips 66, Marathon Petroleum, PBF, HF Sinclair and Delek all at fifty two week highs on Friday.
The Friday giveback came on Iranian state media reporting planned talks with Gulf states in Oman. We would not pay for that. In the same week, OPEC reporting showed Saudi output at 6.238 million barrels a day, the lowest since 1990, exports at a thirteen year low, and reported market supply of 7.122 million against production of 6.238 million, meaning roughly 900,000 barrels a day of August deliveries came out of storage rather than out of the ground. A producer covering a ninth of its supply from tank farms can hold a normal face to the market for some number of months and not indefinitely. Three dollars of headline relief against that backdrop is positioning, not information.
One thing genuinely did change this week, and it was not diplomatic. The Houthi campaign moved from ships to Saudi soil, with attacks on Abha, Khamis Mushait, Jizan and Najran wounding at least 73 people and igniting fires at oil and utility sites. A conflict priced as a shipping problem now has a production leg, and that is a different distribution of outcomes even at an unchanged price.
Asia Broke Harder Than Anywhere Else
Japan was the week’s real casualty. The Nikkei fell 1.93% on Friday to 64,011.34 and lost about 1.55% on the week, with the selling concentrated in Kioxia, Advantest, SoftBank and Tokyo Electron, which is to say in exactly the names that led the year. The yen strengthened 1.26% over the week to around 154.17, so this was not a currency event. Japanese producer prices rose 7.6% in August and the Bank of Japan meets on Friday the 18th, where a move to 1.25% would be the highest policy rate in roughly thirty one years. When Asian AI names sell off on a rate move rather than a demand datapoint, the market is telling you that AI exposure in Asia now trades as a duration asset.
Hong Kong lost about 3.3% on the week to 24,805 and Shanghai gave back roughly 1% to 3,888. Europe held up in price and not in tone, with the Stoxx 600 up 0.3% on Friday to close a week down 2.2%, its worst since April, the DAX at 25,568.56, the CAC at 8,179.77 and the FTSE at 10,650.44. Banks and aerospace led the Friday session, which is the same real economy bid that has defined the European tape all year.
Developments Worth Carrying Into Monday
- Breadth: nine of eleven S&P sectors rose on the week, led by technology, industrials and communication services, while healthcare and utilities lagged. That is a better internal picture than the index suggests, and the index still fell, because the losses were concentrated in the largest weights.
- Small caps: the Russell 2000 fell 2.39% on the week against the S&P at 0.80%. Small caps are the cleanest domestic funding proxy available, and a three to one underperformance into an 88% hike is consistent rather than anomalous.
- Volatility: the VIX fell 11.2% on Friday to 15.84, into an FOMC with a dot plot, a Bank of Japan meeting, a quarterly expiry and a war in the Gulf all inside seven sessions. That is the single most complacent number on the page and it was also the most complacent number on Friday’s page.
- Healthcare: the sector produced the week’s worst tape. Cooper Companies fell 22.98% on a cut outlook, Amgen fell 12.48% after the Novartis and Ionis pelacarsen trial missed on its primary cardiovascular endpoint, and Casey’s fell 16.46% on a same store sales print of 3.2% against 3.8% expected. The Amgen move is the more interesting one, because a failure in someone else’s lipoprotein(a) programme took a quarter of a trillion dollar company down double digits, which is what happens when an entire sector’s valuation rests on pipeline optionality.
- M&A: Copart agreed to buy ACV Auctions for $1.9 billion at $10.50 a share, sending ACV up more than 40%, and Constellation agreed to buy the RISEC plant from Shell for $715 million, adding merchant generation in New England. Deal flow is alive in power and in unglamorous marketplaces, and absent everywhere the index actually trades.
- Space and optics: Rocket Lab flew its sixteenth Electron of 2026 and its ninety fifth overall, carrying an Earth observation satellite for an undisclosed customer into a 500 kilometre orbit. Combined with LightPath’s 197% backlog growth and Frequency Electronics’ record funded backlog earlier in the week, the small cap photonics and launch complex is running at capacity while nobody models it.
- Trade: Canadian tariffs of 15% to 50% on more than 700 American goods are live, Washington has no talks scheduled, and a US federal procurement ban on Canadian products takes effect on September 29. This is a slow grinding cost input rather than a headline risk, and it lands in the same core goods basket that diesel is already pushing on.
Earnings That Mattered
Oracle was the print of the week and the reaction is now more interesting than the print.
| Metric | Consensus | Actual | Verdict |
|---|---|---|---|
| Revenue | $19.1bn | $19.3bn, +30% | Beat |
| Non-GAAP EPS | $1.74 | $1.92 | Beat |
| Cloud infrastructure | n/a | $7.4bn, +121% | Large beat |
| FY27 capex guide | n/a | $90bn to $95bn | The problem |
| Friday close | n/a | $150.28, -1.74% | Faded a +7% open |
Our take: a company cannot beat on demand this comprehensively and close red unless the market has repriced the business model rather than the quarter. What changed is the denominator. At $90 billion of capital expenditure against $90 billion of revenue, with negative free cash flow and more than $100 billion of long term debt, Oracle is an infrastructure developer with a software attachment, and developers are valued on cost of capital rather than on growth. We think the credit rating is now a more important variable for this equity than the backlog, and that the RPO line, which every desk still leads with, has stopped being the thing that moves the stock.
Adobe beat and fell about 3.7%, the same lesson from the other end.
| Metric | Consensus | Actual | Verdict |
|---|---|---|---|
| Revenue | $6.70bn | $6.76bn | Beat |
| Non-GAAP EPS | $6.09 | $6.13 | Beat |
| AI-first ARR | n/a | above $650m, +150% | Real |
| Q4 revenue guide | $6.96bn | $6.80bn to $6.85bn | Miss |
Our take: AI recurring revenue growing above 150% is not the issue. Adobe’s AI revenue is additive while its seat revenue is being eroded by the same technology, so the sum guides down. Nothing re-rates here until the additive line visibly outruns the erosion, and that is a 2027 question rather than a fourth quarter one.
The most useful earnings signal of the week was not an earnings report at all. Dell and Hewlett Packard Enterprise both reported in early September, Dell with $60.9 billion of AI server orders, $16.4 billion of recognised AI server revenue and a $95 billion backlog, and both spent Friday adding 12% on someone else’s guidance. When suppliers reprice on a customer’s capital budget rather than on their own bookings, the market is valuing the order book as an annuity. That works for as long as the customers can fund it, and the funding is now priced off a 4.97% ten year.
Key Events Next Week
| When | Event | What matters |
|---|---|---|
| Mon | BRICS summit, New Delhi, day two | Xi, Putin and Pezeshkian in one room |
| Tue 15 | China August activity data | Retail sales, fixed asset investment, property |
| Wed 16 | US retail sales, August | The consumer after a 27% gasoline year |
| Wed 16 | FOMC decision and dot plot, 2pm | 88% priced, the dots carry the information |
| Wed 16 | Warsh press conference, 2:30pm | His first as chair: insurance step or cycle |
| Wed 16 | Lennar earnings | Housing at a 4.97% ten year |
| Thu 17 | FedEx earnings | The cleanest read on diesel pass through |
| Fri 18 | Bank of Japan | 1.25% would be a 31-year high |
| Fri 18 | Quarterly expiry | Gamma into a post-FOMC tape |
The two that matter are stacked. Wednesday’s dot plot tells us whether September is a single insurance move against a barrel or the start of something, and Friday’s Bank of Japan decision arrives with the yen already 1.26% stronger on the week and Japanese producer prices at 7.6%. A hawkish Fed and a hiking Bank of Japan in the same week removes the last cheap funding currency in the system, and the Nikkei’s AI names have spent this week telling us they already know it.
Geopolitical Risks
The Gulf is the live one and its shape changed this week. American forces destroyed five Iranian crude carriers on September 8, Iranian forces struck vessels near Khasab and fired on US warships, the Houthis hit four Saudi cities, and by Friday Iranian state media were briefing talks with Gulf states in Oman. That sequence is now familiar enough to model: escalation, then a diplomatic signal short of an agreement, then three to four dollars of premium out of the barrel. By our count this is at least the fifth such signal this year and none has produced a signed corridor framework. The base rate on them is poor and the market keeps paying full price.
The more consequential development is at the BRICS summit in New Delhi, which runs today and tomorrow with Xi Jinping, Vladimir Putin and Masoud Pezeshkian all attending, Xi in India for the first time in seven years. The agenda includes cross border payments, a logistics and supply chain framework, and energy cooperation. We would not overread the de-dollarisation headlines, and analysts close to the bloc are explicit that nobody is trying to substitute the dollar. What is worth tracking is narrower and more real: a payments and settlement rail among producers and consumers of physical commodities, convened in the same month that an American Fed is hiking, an American administration is running a tariff war against Canada, and the Strait of Hormuz is being policed by the US Navy. Institutions built in conditions like these tend to be used later even when they are unimpressive at launch.
Elsewhere, the Canada tariff exchange escalates on September 29, the Bank of Japan decides on Friday, and the Xi and Trump summit later this month still sits over unresolved rare earth and optical transceiver disputes, the second of which now runs directly through the names that led Friday’s tape.
Accountability
We got Friday wrong in a specific and useful way. On Friday morning we wrote that the market paying 7% for Oracle’s delivery was the right reaction and that the constraint had loosened. Oracle closed down 1.74%. The demand read was correct and the conclusion about the stock was not, because we treated the capital expenditure guide as a reservation attached to the equity when the market treated it as a transfer of value to other equities. Being right on the mechanism and wrong on who captures it is the more expensive error of the two, and it is worth naming plainly: the correct expression of last week’s Oracle view was long the suppliers, not long the print.
We were also right on the rates leg and understated it. We said the discount rate half of the AI trade was underpriced. The ten year closed at 4.974% and September pricing went from 71% to 88% in an afternoon, which is faster than we expected.
Everything above settles on Friday’s close. The dot plot on Wednesday and the Bank of Japan on Friday are the two things that can make this note read differently by the following weekend.