Brent at a Six-Month High Three Days Before the Fed. France Dodged the Downgrade We Expected.

Saturday, October 24, 2026, written at 9:00 a.m. Eastern. This is a weekend wrap covering the week to Friday’s close. American markets are shut and futures do not reopen until Sunday evening, so every equity, yield, currency and commodity figure above is a settled mark rather than an indicative one. Two things land before the next session. France holds a third consecutive weekend of national protests today with roughly 70,000 police on the street, and the Federal Reserve goes into Tuesday’s meeting without the inflation print that would have told it what to do.

Top Call: The Inflation Impulse Arrived Before the Fed, and Without the Data to Judge It

Brent settled Friday at $116.72, up 3.1% on the day and 6.2% on the week, the highest settlement since April 29 and a third consecutive weekly gain. WTI settled at $103.27, up 3.8%. Both moved on Houthi strikes against Saudi energy infrastructure and a growing market assumption that American strikes on Iran resume once the November 3 midterms are past.

Price that against what else printed in the same session. The S&P Global flash composite fell to 57.6 from 58.4, which is still firm expansion, but the detail is what matters: new orders hit their highest level in four and a half years and selling prices hit their highest since November 2022. Final Michigan sentiment came in at 48.0, a five-month low on the headline, while one-year inflation expectations sat at 4.7% and the five-year measure rose to 3.5%. So on one day the market got a four-month high in crude, the fastest selling-price growth in four years, and consumers lifting their long-run inflation expectations.

That combination is the whole problem with next week. The Federal Open Market Committee meets Tuesday and Wednesday and is expected to hold at 3.75% to 4.00%, with 94 of 100 economists polled by Reuters forecasting no change and futures putting an October hike near 16%. December is priced somewhere between 76% and 90% depending on which day you take the quote. But September core PCE and third quarter gross domestic product do not arrive until Thursday October 29, the day after the decision. Chair Kevin Warsh has already made clear he does not intend to pre-commit, telling an IMF audience in Bangkok that he will not say now what he might do at later meetings.

Our read is that the committee has been handed an energy-led price impulse and no fresh read on the underlying trend, and that the asymmetry in Warsh’s position runs hawkish. A chair who rejects forward guidance, who took the committee to a unanimous hike in September, and who is looking at $117 crude and 4.7% one-year expectations has very little reason to spend Wednesday’s press conference talking the December hike down. We think the risk into Wednesday is a hold with language markets read as more hawkish than the hold itself, and that the duration buyers who stepped in during Thursday’s tech selloff are the ones positioned against that.

The Week Was a Breadth Story, Not a Drawdown

Put the weekly numbers side by side and the headline indices stop being the story. The Dow rose 0.41%, a third consecutive weekly gain. The S&P 500 fell 0.47%. The Nasdaq Composite fell 0.71%, a second consecutive weekly loss. The Russell 2000 rose 2.0%, and the equal weight S&P tracker gained 1.1% while the capitalisation weighted index fell.

Friday itself inverted the week’s pattern. The Dow lost 652.82 points or 1.23% to 52,215.16, the S&P 500 lost 0.52% to 7,623.68, and the Nasdaq Composite lost only 0.21% to 25,955.34. Technology steadied while the defensives that had been absorbing the rotation gave some back, with staples, utilities and healthcare all lower and energy the clear gainer.

We have described this tape as rotation rather than liquidation since the middle of the month, and the weekly breadth keeps supporting that. A market where small caps add 2% and the equal weight index rises while the Nasdaq falls is not one that is being forced out of equities. It is one that is being paid to own cash-generating assets instead of capital-spending plans.

The honest qualifier is that Friday complicates the clean version of this story. If the rotation thesis were purely about preference, the Dow should not have been the worst of the three on the session that crude settled at a six-month high. What happened on Friday looks less like rotation and more like a repricing of the input cost of everything that is not a software company, which is a different and less comfortable mechanism. One session does not overturn a week, but it does mean the next leg has to be watched for whether defensives keep absorbing the flow or start trading as the oil-sensitive cyclicals they partly are.

The AI Funding Ledger Got Longer, and One Number Deserves Context

Three entries went on the board this week and all three point the same way on funding costs while saying very little about demand.

CoreWeave’s five-year default swap pushed past 1,100 basis points, with reported prints as high as roughly 1,140 and a reading near 1,082 on October 19, against a threshold where S&P treats anything above 1,000 as distress. Its unsecured 2031 and 2032 notes changed hands between 76.5 and 81 cents on the dollar, with yields on the 2031s near 16%. Oracle’s swaps widened to roughly 300 basis points, which is a different order of magnitude. Chief executive Michael Intrator called the reaction irrational and pointed at the backlog.

The arithmetic behind the quote is the part worth carrying. The company reported $35.6bn of debt against $103.7bn of contracted backlog, and first-half operating cash flow of about $3.7bn against roughly $14.1bn of capital spending. A backlog is a claim on future cash. Debt service is a claim on present cash. The swap is not pricing whether the contracts exist, it is pricing the gap between those two timelines, and that gap is roughly $10bn per half year.

OpenAI cut the valuation it is asking from new investors to $1.25tn from $1.4tn on a round of up to $30bn, with commitments wanted by October 30 and an initial public offering now described as early 2027 business. For scale, March’s raise took $122bn at an $852bn valuation, so the company is still marking up by nearly half while accepting an 11% haircut to the ask.

Anthropic is the entry that needs a caveat rather than a headline. Reporting on its draft prospectus put the first-half net loss at $80.7bn against $13bn a year earlier, a number that has been circulating all week as evidence that the economics do not work. About $72bn of it was a non-cash increase in the estimated value of bonds that can convert into Anthropic shares. Strip that and the operating picture is revenue of $16.3bn in the first half against $1.1bn a year earlier, with compute and infrastructure spending near $9bn against $2.3bn. The company has committed to at least $518bn of cloud and compute spending over a decade, and a public filing was expected as early as Friday ahead of a possible November listing, with investors reportedly weighing something closer to $1.5tn than the $2tn once discussed.

We want to be precise here because the headline is doing work the number does not support. A $72bn mark-to-market on convertible paper is a statement about how much the equity has appreciated, not about how much cash the business burned. The real disclosure is the $518bn commitment and the fifteen-fold revenue growth that has to service it. Treating the accounting loss as the scandal lets the actual question, which is the duration mismatch between decade-length compute commitments and annual revenue, go unexamined.

Earnings: The Demand Side Keeps Clearing

Western Digital remains the most instructive print of the week precisely because it did everything right and lost anyway. Revenue of $4.38bn grew about 55% year on year, non-GAAP gross margin reached 58.1%, cloud hit a record 88% of revenue, and the fiscal second quarter guide of $5.1bn implies roughly 69% growth with gross margin stepping to 60% to 61%. Management then told the market that capital spending will run modestly above its normal 4% to 6% of revenue band for two years and that the money goes into HAMR heads, media technology and factory automation rather than new capacity. That is a supplier saying in public that it intends to keep nearline storage tight and harvest the price. The shares fell 6.4% on the print and 3.8% more on Friday. Being right about the physical bottleneck has stopped paying, and the reason is visible in the section above: the marginal buyer is now asking who finances the building, not how many exabytes ship.

Lam Research and Texas Instruments are the pair that frames the semiconductor cycle honestly. Lam cleared published consensus on both lines and still traded down because it missed the buy-side number, which is what happens to an equipment name when expectations have run ahead of the print. Texas Instruments landed inside its own guidance range while telling the market industrial revenue is still below its 2022 peak and customer inventories have not been rebuilt. Read together: the artificial intelligence capital cycle is intact and the broad analog cycle is not, and anyone treating semiconductors as a single exposure is holding two different businesses.

SLB is the quarter that reads straight through to the energy section. Revenue grew 5% sequentially and 6% year on year with adjusted EBITDA margin up 71 basis points to 21.9%, and Middle East revenue of $1.9bn rose 13% from the prior quarter on recovering activity. Free cash flow was $828mn. Year on year earnings still fell, so this is a company whose geography mix is doing the work rather than a sector-wide recovery, but Middle East spending accelerating while Hormuz is a war risk headline is a signal about where the barrels of 2028 are being contracted.

American Express stays the cleanest read on the consumer. Billed business up 8% to $454.0bn, provisions for credit losses down to $1.0bn from $1.3bn a year earlier, and the full-year adjusted guide raised to $17.60 to $17.90. The reconciliation with the 48.0 Michigan print is that premium households keep spending and keep paying while sentiment reflects prices and rates. The stress in this cycle sits on corporate balance sheets financing compute, not on the household ones.

Energy: A Freight Problem Wearing the Clothes of a Supply Shock

The move in crude is real and the mechanism is narrower than the price suggests. Physical production has not been interrupted at scale. What has broken is transit. Goldman puts the global average crude freight rate above $20 a barrel against roughly $3 a year ago, which means something like $17 of the delivered price is now the cost of moving the barrel rather than the cost of lifting it.

The week’s catalysts were a Revolutionary Guard claim that seven vessels were hit or mined in and around the Strait of Hormuz, Saudi Aramco telling staff to leave its Dhahran headquarters, and the Tuesday attack on King Fahd airport that killed two people alongside an intercepted drone at the Shaybah oilfield. The Associated Press could not verify the vessel claim and President Trump said Iran is inventing incidents to lift the oil price. That qualifier matters more than it sounds, because it tells you the market is paying a premium for the possibility of an interruption rather than for one that has occurred.

What makes an unverifiable claim worth three dollars is the absence of a cushion. Effective OPEC+ spare capacity has been running near a quarter of a million barrels a day and observed inventories are hundreds of millions of barrels below where they began the year. At that level of slack, optionality is expensive and the market will pay for it on a rumour. The structural view we have held does not change: a 2027 crude surplus sits underneath a present shortage of distillate and freight, and the risk premium is attached to transit rather than to production. The addition this week is that the premium now has a date behind it, because the planning reporting points past November 3 and Washington’s own ambassador to the United Nations has said the Iran conflict could run another year.

Europe: France Dodged the Downgrade and We Did Not

We wrote on Friday that a one-notch cut to A1 from Aa3 with a negative outlook was widely expected after the close and that the only part worth watching was the outlook language. Moody’s instead concluded a periodic review and affirmed Aa3, leaving the negative outlook in place. We conveyed a downgrade as the base case and it did not happen, and we are not going to dress that up as a technicality.

The rationale is worth reading closely because it is conditional rather than reassuring. Moody’s said the Aa3 rating assumes a deficit-reducing 2027 budget is adopted and that financing costs and social tensions abate somewhat in coming months. That is not an endorsement of the fiscal position, it is a stay of execution priced on two things that have to happen. The finance ministry credited the credibility of the 2027 finance bill, which still has to pass a parliament where opposition parties can reject it and topple the government in a confidence vote. The next scheduled agency verdict is December 11.

Where our framing does hold is that the bond market was never waiting for the label. The ten-year has been trading near 5.06% after touching 5.14%, its highest since 2002, with the spread over Germany at its widest since 2011. The affirmation removes a forced-selling catalyst from index and mandate holders. It does not change a 5.4% deficit, a flash composite PMI at 48.4 with both services and manufacturing contracting against a eurozone composite at 51.9, or the fact that about 70,000 police are deployed today after a weekend that produced roughly 2,400 arrests and almost 1,000 injured officers. The savings target behind the unrest is EUR 54bn, including a public-sector salary freeze worth around EUR 2bn.

The Stoxx 600 closed Friday at 635.77, down 0.31%, for its steepest weekly loss since early September, with the FTSE 100 at 10,690.36, the CAC 40 at 7,782.83 and the DAX at 25,152.30. Kering did most of the single-stock damage.

Asia: The Memory Crown Changes Hands Before the Prints

The most consequential Asian item this week was an analyst note rather than a company release. Bernstein’s read of export data suggests Samsung overtook SK Hynix in high bandwidth memory revenue in the third quarter, with SK Hynix up about 11% sequentially to roughly $7.6bn against a Samsung figure estimated nearer $12bn, helped by HBM4 shipments lifting Samsung’s share above 30%. Both companies report in the coming week, with SK Hynix expected between Tuesday and Thursday and Samsung’s full report tentatively Thursday.

If that holds in the actuals it is a genuine change in the structure of the trade. For two years the clean way to own accelerator demand without owning accelerator valuations was SK Hynix, on the argument that it held roughly half the HBM market and had the Nvidia qualification lead. A Samsung that has taken the revenue lead on HBM4 turns a near-monopoly rent into a duopoly margin, and duopoly margins compress. The caveat is real: Bernstein is working from monthly export proxies rather than company data, and neither Nvidia nor SK Hynix has confirmed the shift. That is exactly why next week’s prints matter more than the usual record-quarter headlines.

Elsewhere the Nikkei closed at 62,693.16, down 0.34%, with Japanese core inflation at 2.1% and the Bank of Japan deciding on Friday. Hong Kong added 0.69% to 23,093.37. The CSI 300 bounced 1.71% to 4,105.33 and still finished lower for an eleventh consecutive week, which is a streak with a pause in it rather than a turn. Korea closed up 0.30% at 5,550.30 after SK Hynix abandoned the Solidigm listing, and holders appear to have preferred a parent that keeps the asset whole to one that sells a stake into a bad tape.

Also On The Tape

• SpaceX is in talks for access to at least 80,000 American cell sites, roughly half the country’s purpose-built towers, with Crown Castle named among the counterparties. Tower landlords rose and the three incumbent carriers fell. No lease has been confirmed, and the carriers’ selloff prices a competitor that does not yet exist
• Strategic deal flow stayed heavy despite the funding noise. Salesforce confirmed its purchase of Braze at $47.50 a share in cash for about $5.66bn of enterprise value, a reported 58% premium, with Tradeweb taking the credit venue Trumid for roughly $1.97bn and Maximus buying Agile Defense for $660mn. Smith & Nephew’s talks with Johnson & Johnson reportedly collapsed over who would run the combination
• CrowdStrike was among the strongest large caps on Friday, up between 3.8% and 8.2% depending on the tape you read, on an Nvidia partnership and federal warnings on PHP vulnerabilities
• Gold settled near $4,011.71, down 0.30% on the day and 3.8% on the week, while silver near $54 printed a year-to-date low. The metals falling as crude rallies is the week’s most underdiscussed divergence and argues the bid in oil is a transit premium rather than an inflation hedge
• Bitcoin is about $80,700 after spot funds saw $635.6mn of outflows on Friday and more than $1bn across three sessions. It has been range-bound between roughly $81,000 and $84,500 since mid-month and sits about 35% below its October 2025 peak
• Chinese hyperscalers including ByteDance, Alibaba and Tencent have reportedly placed more than $35bn of Nvidia orders, largely for RTX PRO 5500 workstation cards, with Nvidia warning it cannot guarantee the quantities requested
• Regional banks closed out a solid reporting week, with PNC growing loans 14.2% and Regions beating at $0.73 adjusted against $0.66, though PNC’s non-performing loans rose to about $2.2bn from roughly $2.0bn at June 30
• Healthcare produced three approvals worth noting: AbbVie’s subcutaneous Skyrizi induction regimen in Crohn’s disease on Friday, GSK’s Hibsago in chronic hepatitis B, and Revolution Medicines’ daraxonrasib in pancreatic adenocarcinoma six and a half months ahead of its deadline
• United States and China have extended their trade truce only to January 10, with a tariff reduction covering about $30bn of goods. Both leaders are expected at APEC in Shenzhen in November and the G20 in Miami in December

The Record

One call goes against us, two tests stay open, and the week’s biggest number gets a correction.

France resolves against us. We presented a Moody’s downgrade as the expected outcome on Friday and framed the whole section around what the outlook language would signal. The agency affirmed Aa3. The substance of our argument, that the bond market had priced the fiscal problem weeks ago and that the rating was a label rather than an event, is intact and arguably reinforced, because the affirmation moved nothing structural either. But the directional expectation was wrong, and the practical consequence is that a forced-selling catalyst we thought was arriving did not arrive. Anyone positioned for spread widening into the verdict was positioned on our framing.

The Anthropic correction. We would have led with the $80.7bn loss if we had written it earlier in the week, and that would have been a mistake, because roughly $72bn of it is a non-cash revaluation of convertible instruments. We are flagging it here rather than quietly dropping it, because the figure is going to keep appearing in bubble commentary and it does not mean what the commentary says it means.

Test one stays open, due Friday. We said on Friday that if CoreWeave’s five-year default swap is still above 1,000 basis points at the close on October 30, the funding stress is structural and the thin dealer-quoted market objection is dead, and that a return below 800 without new financing means we overweighted one illiquid instrument. The spread sat above 1,100 this week, so the test is live and unresolved.

Test two stays open, due Wednesday and Friday. Our claim has been that the equity drawdown in the artificial intelligence build is itself a tightening of financial conditions that substitutes for policy. The test as written: a hold on October 28 with December hike odds ending that week below 60% and the ten-year still under 5.10% confirms the substitution read. December odds above 80% with the ten-year above 5.15% after a $117 Brent print kills it, and makes this a sector event rather than a macro one. Friday’s close puts the ten-year at exactly 5.00% and December pricing in the high seventies, so this one is going to resolve on Warsh’s language rather than on the decision.

New test, Thursday. If Samsung’s and SK Hynix’s third quarter disclosures confirm that Samsung has taken the high bandwidth memory revenue lead, we will treat the memory trade as a duopoly margin story rather than a scarcity rent and say so. If SK Hynix’s own numbers show it still ahead on HBM revenue, the Bernstein export-proxy method was wrong and we should not have given it a section.

Key Events Next Week

• Monday Verizon before the open, with Electronic Arts and Southern Copper after
• Tuesday and Wednesday FOMC meets, decision at 2:00 p.m. Eastern Wednesday and Warsh’s press conference after it, a hold at 3.75% to 4.00% priced at roughly 84%
• Wednesday Microsoft and Alphabet report after the close, a few hours after the decision, with consensus near $4.71 and $2.94. Bank of Canada and the Reserve Bank of New Zealand also decide
• Wednesday to Thursday SK Hynix third quarter results, with Samsung’s full report tentatively Thursday
• Thursday September PCE and third quarter gross domestic product at 8:30 a.m., the day after the Fed decides. Amazon reports after the close on consensus near $2.01, and the European Central Bank announces
• Friday Bank of Japan decision
• Monday November 2 Apple reports fiscal fourth quarter results after the close
• Watching CoreWeave and hyperscaler default swaps into the October 30 close, whether OpenAI gets its $30bn signed by Friday, whether Anthropic’s public filing lands ahead of a November listing, Hormuz transit and freight rates, and the French budget’s path through parliament before the December 11 verdict