OpenAI Cut Its Own Round. The Funding Market Said No in Three Places at Once.

Saturday, October 31, 2026, written at 8:30 a.m. Eastern. This is a weekend wrap covering the week to Friday’s close and the last session of a month that changed the character of this market. American equities are shut and futures do not reopen until Sunday evening, so every level above is a settled mark rather than an indicative one. Clocks change tonight, which pulls the London to New York gap to five hours from Monday. Three things sit between here and Tuesday: Apple reports Monday afternoon, the midterms are Tuesday, and reporting says Washington has told Gulf allies to be ready for strikes on Iran as early as November 4.

Top Call: The Money Was Rationed in Three Markets on the Same Day

Friday produced four separate pieces of news, and they are the same piece of news. The Financial Times reported that OpenAI has cut its fundraising target to roughly $20 billion from $30 billion and its valuation to $1.25 trillion from $1.4 trillion, with reporting elsewhere suggesting only about $11 billion has actually been committed and that the first close could land below even the reduced target. Blackstone, the anchor investor in Anthropic’s $63 billion chip financing and until recently a forecaster of an investment grade rating by year end, now expects Anthropic to come public carrying a double B. S&P Global cut its outlook on Meta to negative, keeping the AA- rating but writing explicitly that it is no longer confident the company will moderate spending. And gold fell 3.5% to a one-year low while silver broke $50, which is what happens when people sell the thing they can sell to pay for the thing they cannot.

Put those in a row and you have the private market, the credit market and the public market all declining to fund the same build within a few hours of each other. We have spent two weeks arguing that this drawdown was about the cost of capital rather than about demand. Friday is the session where that stopped being an interpretation and became an observation, because the company that defined the cycle marked its own round down by a third without a single customer leaving.

The detail that matters most in the S&P action on Meta is the reasoning, not the rating. The agency projects roughly $140 billion of capital spending this year, above $225 billion next year and $252 billion in 2028, against negative free operating cash flow it puts somewhere between $22 billion and $39 billion for 2026 and $71 billion for 2027, and it notes that Meta, unlike Microsoft or Alphabet, has no cloud business through which to sell the compute back out. That is a rating agency pricing a spending plan rather than a balance sheet. On October 26 we wrote that if Meta’s own credit protection widened despite a $95 billion equity raise, it would mean the market was pricing the plan and not the company. That test has now resolved, and it resolved against the more comfortable reading.

What this does not mean is that the artificial intelligence build stops. Demand cleared everywhere it was tested this month, from SK hynix sold out to 2029 to Amazon Web Services accelerating to 39.3%. What it means is that the marginal dollar is being priced properly for the first time in this cycle, and every entity in the chain now has to be sorted by whether it can fund itself from operations. That sorting is not finished. It has roughly two weeks of visible runway in Anthropic’s pricing, OpenAI’s close and the Treasury refunding, and we would not treat any bounce before those resolve as the end of it.

The Month in Numbers, and Why the Headline Index Understates It

Friday was the worst session since June for both the S&P 500 and the Nasdaq, and it was enough to flip the S&P from a positive October to a negative one in a single afternoon. The Nasdaq Composite recorded its worst month since March 2025. The Nasdaq 100 finished 6.65% lower on the month and roughly 9% below its October 6 record close of 31,224.47, having touched correction territory intraday. Technology posted its largest monthly loss since March 2025 and communication services its deepest since April 2022.

Now look at the same month from the other side. Nine of the eleven S&P sectors finished October higher. The Dow gained on the month. Even on Friday, with the Nasdaq down 2.86%, energy rose 1.91%, staples 1.41%, utilities 0.67%, healthcare 0.45% and materials 0.40%, while the equal weight S&P fell only 0.79% against the capitalisation weighted index at 1.82%. The week told the same story, with the Russell 2000 up roughly 1.2% while the Nasdaq lost about 2%.

This is the most important thing to hold in your head over the weekend, because it decides how you position for Tuesday. A market in which nine sectors rise and the index falls is not a market losing confidence in equities. It is a market withdrawing capital from a specific funding structure and redeploying it into companies that already generate cash. The headline index is a poor instrument for measuring what is happening because its weights are the problem being priced. Roughly $2 trillion came out of the large technology complex in October while the median stock had a perfectly respectable month.

The symbolic marker arrived at the close. Apple fell 1.2% and still became the world’s most valuable company again at about $5.03 trillion, passing Nvidia at $4.93 trillion after the chipmaker dropped 4.2%. Apple is up about 27% this year and is the one member of the group that has conspicuously not committed to a hundred billion dollar annual build. The market crowned, on the last day of October, the company that chose not to spend. That is not subtle.

SanDisk Is the Cleanest Datapoint of the Week

If you wanted one print to explain the month, this is it. SanDisk beat on revenue, beat on earnings, guided the next quarter above consensus on both lines, and announced an eight-for-one stock split effective November 19. The shares fell 13.5% on Friday. There was no bad number in the release.

What sold it was the suspicion that memory pricing is at or near its peak, combined with a valuation that had already discounted several quarters of the same. The storage and memory complex has been the best performing corner of this market for most of the year on a genuinely excellent supply story, and the first moment the market decided to ask what the second derivative looks like, a flawless quarter was worth minus thirteen percent. We have written three times this month that demand was not the question. SanDisk is the proof in the most favourable possible test case.

The wider optics and hardware tape rhymed. Applied Optoelectronics fell about 15%, Coherent 11.3% to $210.32, Lumentum about 10% to $824.82 and Marvell about 10% to $197.50, with the reported $83 billion all-stock approach for Lumentum still unsigned and possibly landing next week. AMD fell 9.3%, Broadcom 5.5% and Micron 5.3%. The Philadelphia Semiconductor Index lost 7.2% in a single session. None of these companies reported anything bad on Friday. They are being repriced as a group because their revenue is someone else’s capital spending, and that someone else is now being asked how the spending is funded.

Intel and Amazon: What Actually Happened at the Close

Thursday night’s two prints drove Friday, and the closing marks correct part of what we wrote yesterday morning. Intel beat the quarter at $16.51 billion of revenue, up 21%, and $0.52 adjusted, then guided the fourth quarter to $15 billion to $16 billion against roughly $17 billion and $0.25 against $0.43, citing a personal computer slowdown running at least through the first half of next year and supply constraints on data centre components. It closed down 28.6%, among the worst single sessions in the company’s listed history.

Amazon is the more interesting one, and we were wrong about it intraday. The stock opened down roughly 7% on $77.3 billion of quarterly capital spending against $58.7 billion modelled and $35.1 billion a year earlier, with trailing twelve month free cash flow at minus $42.4 billion and the 2026 plan reaffirmed near $220 billion. It closed at $240.35, down only 1.41%. We wrote yesterday that we would not step in front of the capital spending disclosure. A buyer did, all day, and was right by the bell.

The honest read is that the capital spending punishment is real but it is not indiscriminate, and the distinction is the one S&P drew about Meta. Amazon has a cloud business that sells the compute, growing 39.3% with a $578 billion backlog, so the spending has a visible revenue counterparty. Meta does not, and its outlook got cut. Intel is spending into a market where it cannot source parts and cannot raise prices, and it lost a quarter of its value. The market is not refusing to fund artificial intelligence. It is refusing to fund artificial intelligence without a contracted buyer, and that is a far more rational tape than the word selloff suggests.

Gold at a One-Year Low Is a Liquidity Signal

Spot gold fell about 3.5% to roughly $3,881, the lowest since October 2025, and finished its worst October since 2008 with a decline near 6%. Silver dropped 6.5% to about $49, under $50 for the first time in eleven months. Platinum hit a thirteen month low and palladium went below $1,000. Gold miners were hit harder than the metal, with Kinross down 7.7% and the sector broadly down about 7%.

Take this seriously, because it is the hardest thing on Friday’s tape to explain with a story. Crude rose. The dollar was flat. Yields fell. Those are three conditions under which gold normally rises, and it had its worst day in a long time anyway. Commerzbank called the move exaggerated, and Kitco and Mining.com both reported the same mechanism: people sold metal to meet margin calls generated by losses in technology. Saxo flagged the risk that further fund liquidation pushes silver into the mid forties.

A session in which the hedge is sold to fund the position is a session about leverage, not about views. It tells you that the drawdown in technology has reached accounts that are financed rather than long only, and that is usually a later stage of the process than the headlines imply. We would read Friday’s metals complex as a better indicator of how far this unwind has travelled than anything in the equity sector tables, and we would watch whether gold stabilises on Monday. If it keeps falling while crude holds above $100, the margin story is still running.

Energy and Geopolitics: Two Trading Days to the Midterms

Brent’s December contract settled at $117.64, up 2.4% and its highest in six months, before expiring; the January contract settled at $113.43, so the apparent gap in quoted prices is a roll and not a collapse. WTI settled at $104.12, up 2.1%. For October, Brent gained 13.6% and WTI 15.2%, a fourth consecutive monthly advance. Goldman continues to put around $27 a barrel of geopolitical premium on top of a fair value near $85.

Friday’s news flow was the worst of the month. Reports put at least four tankers hit in the Strait of Hormuz, including the Qatari owned liquefied natural gas carrier Al Mayeda, with the Financial Times reporting limpet mines. Iran’s Revolutionary Guard claimed three tankers exploded and a fourth struck a mine; Central Command rejected the account and maintains key lanes have been cleared. Qatari negotiators spent roughly twelve hours in Tehran and left without an announced agreement, after a letter from Vice President Vance proposing direct talks with the nuclear programme on the table. Iran’s parliament is reported to have received at least ten separate proposals to leave the Non-Proliferation Treaty. Non-emergency American staff have reportedly been told to leave Saudi Arabia and tanker aircraft have shifted away from Al Udeid.

Against that, President Trump said late in the session that peace talks are taking place and added that there is time, which pared the move after settlement. This is the eighth or ninth time this month that a headline has taken a dollar out of crude without anything physical changing, and we keep making the same point: the premium is attached to transit and optionality, not to lost barrels. The asymmetry into next week is the sharpest it has been. Reporting points to strikes becoming possible from November 4, the day after the midterms, which is the first moment at which the stated political constraint lapses. There are two trading sessions between now and that window. We would not be short energy into it, and we would be careful with anything whose input cost is a barrel.

Rates: A Rally Nobody Should Enjoy, Into a Heavy Week

The two year finished at 4.56%, down five basis points, the ten year at 4.94%, down four, and the thirty year at 5.31%, down one. December hike pricing drifted down toward 70% to 75% from roughly 84% immediately after Wednesday’s hold, which came with Cook and Hammack dissenting in favour of an increase.

The shape of that is a front end rallying on growth worry rather than on inflation relief, and it is keeping bad company. Crude is at a six month high, euro area inflation printed 3.8% in September, and the Employment Cost Index on Friday showed real compensation down 0.4% over the year. A market pricing fewer hikes because activity is softening, while the energy input that drives headline inflation runs, is the configuration that leaves equities without rescue from either direction. Duration bought this week is paid only if the slowdown is genuine, and the evidence for that is one soft Chicago PMI.

Next week settles it. The Treasury’s quarterly refunding lands Wednesday, with the market focused on whether guidance on nominal coupon and floating rate auction sizes is maintained, and October payrolls arrive Friday with consensus near 55,000 after 29,000 in September, unemployment expected unchanged at 4.2% and average hourly earnings at 0.3% monthly. Barclays is at 50,000 with a 4.1% unemployment rate and argues that combination would keep a December move live. Our view is that the committee that just dissented hawkishly twice needs a genuinely weak number to be deterred, and 55,000 is not it.

Europe gave the week’s other rating news. Fitch revised France to negative from stable at A+ in an unscheduled review, citing weaker growth, higher borrowing costs and a wider deficit since August, and projecting debt rising from about 115.7% of output in 2025 to roughly 124% by 2028. The spread over Germany is back around 164 basis points. France is now the economy that generates a fiscal premium while failing to grow, and with Christine Lagarde leaving the European Central Bank on February 7 the successor question is the live one for European duration.

Asia and Europe

Asia had the worse week and China had the worse month. The Nikkei fell 2.12% to 63,290.11 on Friday even as Japanese industrial production, employment and inflation all beat, with the Bank of Japan holding at 1.25% on an eight to one vote and accelerating its exchange traded fund sales. The Kospi dropped 3.48% to 5,730.74 and tripped a sell side sidecar, with SK hynix down 7.08%. The CSI 300 lost 1.39% to 3,985.50, taking October to roughly 8% lower and marking a twelfth consecutive weekly decline, the steepest monthly loss in more than a decade. The Shanghai Composite closed near 3,599, down 1.44%, with Shenzhen off 2.73%.

Hong Kong was again the exception, up 1.46% to 23,294.42. We have flagged this several times in October and it keeps holding: Hong Kong trades on mainland policy and liquidity rather than on American technology earnings, which makes it a genuine diversifier in precisely the sessions when correlation goes to one everywhere else. That the CSI 300 and the Hang Seng moved in opposite directions on the same day is itself worth noting, since it says the mainland weakness is domestic rather than imported.

Europe closed the month well. The Stoxx 600 slipped 0.36% on Friday to 646.75 but recorded its best month since June, the DAX fell 0.85% to 25,818.23 and the FTSE 100 edged up 0.07% to 10,729.78. A region with almost none of the capital spending exposure and a cheaper starting multiple has quietly been the better place to sit through October, and nothing in the funding story argues against that continuing.

Also On The Tape

• OpenAI is reported to have delayed its own listing to 2027, which follows logically from a down round. A company that has just marked itself from $1.4 trillion to $1.25 trillion in the private market has no reason to ask a public one
• Anthropic filed its S-1 on October 26 and intends to list on Nasdaq as ANT. KBRA already had it at BB+ in June. If Blackstone’s double B call is right, the $63 billion Broadcom accelerator financing gets repriced before it funds, and that is the single most consequential number in the complex over the next month
• The Treasury is reported to have borrowed a record amount in a single week, a gross figure that overstates the net position but which lands two days before a refunding announcement that will set coupon guidance into the new year
• Exxon and Chevron both beat on record refining results, the product of an effectively closed Hormuz and strikes on Saudi and Russian capacity. Both chose balance sheet repair over larger buybacks, which is the correct handling of a windfall that depends on a war continuing
• Friday’s gainers were unglamorous and sizeable: FedEx Freight rose 14.9%, Regeneron 13.3%, Cavco 13.1% and Global Payments 13.0%. Reddit fell 29.2% and Illumina 18.6%. Dispersion inside the index is now wider than the index move, which is what a rotation looks like from the inside
• A Trump and Xi meeting is reported for November 18 and 19, and a United States, Russia and Ukraine meeting is planned in the United Arab Emirates in November, with Washington approving a possible $2.1 billion sale of 500 PAC-3 interceptors

The Record

Yesterday’s note was written at 9:55 a.m. with Amazon down roughly 7% and we said we would not step in front of the capital spending line. Amazon closed down 1.41%. The direction of the argument held across the complex, but on the specific name the opening print was the low and the intraday recovery contradicted the framing. We also wrote that the S&P 500 was still positive for October. Friday’s 1.82% decline took it negative, so that sentence was true when written and false by the close, which is the risk of writing at the open on the last day of a month.

What we got right, and we want it on the record because it was pre-committed on October 26, is the Meta test. We wrote then that if Meta’s credit protection widened despite the $95 billion equity raise, it would mean the market was pricing the spending plan rather than the balance sheet. S&P’s negative outlook says exactly that in the agency’s own language, naming the absence of a cloud business as the reason. The equity-for-debt substitution we identified does not fix the problem, it relocates it.

We also want to correct our framing on memory from earlier in the week. We treated sold out order books through 2029 as unambiguously bullish for the producers. SanDisk beat, raised, split and fell 13.5%, and Intel showed the same shortage destroying a buyer’s guidance. The correct formulation is that scarcity rent is good for whoever collects it only until the market starts discounting the far side of the cycle, and that discounting began this week.

The new test we will carry forward: if this is a funding event rather than a demand event, the first clean confirmation will come from Anthropic’s pricing range in its amended filing. An issue that prices near the top of the indicated range with a double B rating means capital is available and expensive. An issue that is cut or delayed means it is not available at all, and the second outcome would make the past three weeks the opening act rather than the correction.

Key Events Next Week

• Monday: Apple reports, the first quarter under chief executive John Ternus, with consensus near $1.98 on about $113.3 billion. Palantir follows after the close. United States ISM manufacturing, consensus 54.7. China’s National People’s Congress Standing Committee convenes through Friday with no fiscal items on the draft agenda
• Tuesday: Midterm elections, with 35 Senate seats, all 435 House seats and 36 governorships. Forecasters put Democrats between 82% and 94% for the House and between 62% and 75% for the Senate. AMD reports after the close, alongside Amgen, Pfizer, Shopify and Arista. The Reserve Bank of Australia is roughly 79% priced for a hike to 4.85%. Japan is closed
• Wednesday: Treasury quarterly refunding. ISM services, consensus 54.7. ADP employment and euro area flash inflation. The Riksbank is expected to hike 25 basis points to 2.00%. McDonald’s, Uber, Qualcomm and Arm report
• Thursday: The Bank of England is expected to begin a tightening cycle with a 25 basis point increase, with the vote split the thing to read. Norges Bank expected on hold at 4.50% and Banxico at 6.50%. Canadian federal budget. Challenger job cuts. Toyota, ConocoPhillips and Sony report
• Friday: October payrolls, consensus 55,000 with unemployment at 4.2%. Preliminary Michigan sentiment. This is also the first full session inside the window in which renewed American action against Iran becomes politically possible