Thirty Minutes Apart.

On Thursday the President took a war off the calendar, and half an hour later the FT took $20 billion off OpenAI's revenue line. The index split in two. The average stock rose. The AI complex fell. That is the tape we said in May we were waiting for.

In The Other 493, on 17 May, we wrote that the next leg of this market would be one where the index moves and the AI names are not the reason — and that the catalyst, when it came, would run through Iran, oil and the long end of the curve rather than through anything a chipmaker said. Thursday was the cleanest version of that tape we have seen this year. It was also messier than the headline makes it look, and the mess is worth understanding before anyone trades on it.

The sequence. Through the morning, the market was trading a report that the White House was weighing strikes on Iran before the 3 November midterms. Brent pushed above $104, the 10-year sat near multi-decade highs, and tanker attacks in the Strait of Hormuz were running at their highest weekly rate since the war began. Then the President posted that the United States "will not be attacking Iran at any time prior to the Midterm Elections," adding that there were "productive discussions with the Islamic Republic of Iran." Oil came off its highs and Treasury yields reversed to finish at the lows of the day.

About thirty minutes later, the Financial Times reported that OpenAI's annualised revenue was running at roughly $50 billion — about $20 billion below the ~$70 billion run-rate the Street had been modelling off last month's reporting. The semiconductor index fell as much as 4%. By the close, the S&P 500 was down half a percent and the equal-weight S&P was up more than half a percent.

01 / The splitTwo markets, one session.

The composition is the whole story. Technology was down 1.8% and was the only sector down more than half a percent. Energy rose 2.9% and staples 2.1%. Of the ten largest S&P constituents, only Apple finished higher — and nearly all of the twenty worst performers in the index were AI-linked. Meanwhile roughly forty S&P names rose more than 3%, against eight the day before.

Look at who was on the losing side. Not just the chip designers, but the whole physical supply chain of the build-out: optical networking (Coherent, Lumentum), glass and fibre (Corning), the neoclouds (CoreWeave, Nebius), on-site power (Bloom Energy, −6.4%), and the hyperscale landlord most exposed to OpenAI's commitments (Oracle). The market did not sell "tech." It sold a single customer's ability to pay its bills, and every company whose order book leads back to that customer.

For a year the market has treated AI as a sector. On Thursday it priced it as what it is: a chain of receivables that ends at a handful of model labs.

02 / AttributionWhat the post did, and what it did not.

The easy story is that a de-escalation headline rotated money out of AI and into everything else. It is not quite right, and we would rather say so here than have a reader discover it on their own.

• The AI selloff was the OpenAI report, not the Iran post. Chips were soft from the open, but the leg down came after the FT. Nothing about a pre-midterm strike pledge changes a GPU order book.
• The post's fingerprint is on the rates market. The 10-year closed at 5.23%, down about 5 bp, having traded near its highs earlier. A $22 billion 30-year auction then cleared well at 5.618%. Lower long yields are the transmission into the other 493 — and the equal-weight index is, in effect, a bet on that transmission.
• Energy did not rally on peace. Brent still settled up 4.1% at $104.28 and WTI up 3.6% at $91.49. The post took oil off its highs, but Hormuz tanker attacks and a Gulf of Mexico storm kept it bid. Refiners and producers rose with the commodity, not with the de-escalation.
• Some of the winners were idiosyncratic. PepsiCo beat on the quarter. Chipotle rose on an FT report that Starbucks has explored a takeover. Strip those out and the rotation is still visible in breadth — NYSE up-volume of 66.6% is not two stocks — but it is smaller than the sector table implies.

03 / Why it mattersThe correlation broke.

For most of 2026, the S&P has behaved like a single trade: when AI went up, the index went up; when AI wobbled, everything wobbled with it, because the same passive dollar owns both. Thursday is a datapoint that the two halves are now priced on different inputs. The 493 are trading on oil, war risk and the long end. The AI complex is trading on whether the end customers can fund the capex already committed in their name.

That decoupling is what makes the next four weeks interesting. The President has, in effect, written a dated put on geopolitical risk that expires on 3 November. Over that window, the rate-sensitive half of the market has a cleaner runway than it has had since the war began. The AI half, by contrast, now has a revenue question that a single news cycle will not answer — OpenAI's Q4 run-rate, the hyperscaler prints later this month, and whatever Oracle says about its backlog.

04 / The bear caseWhy one session is not a regime.

• The pledge has an expiry date. "Not before the midterms" is a statement about 3 November, not about Iran. If anything it concentrates event risk into the weeks after the election, and the market will start pricing that well before it arrives.
• The Fed is the bigger rate story. Governor Waller and President Musalem both said on Thursday that further hikes are likely needed. A 10-year at 5.2% with the Fed leaning toward tightening is a headwind for small caps and dividend payers that a 5 bp rally does not offset. The Russell 2000 finished roughly flat — not the tell of a full-throated rotation.
• $50 billion is still a very large number. A run-rate $20 billion below a whisper is a miss against expectations, not evidence of collapse. If the market decides it was a sell-the-number event, the dip in Nvidia and Broadcom gets bought, and the spread snaps back as quickly as it opened.
• We have been here before. In June we wrote about the chip unwind and the case for staples. Equal-weight has had several false dawns this year. One good breadth day does not make a trend; three weeks of it does.

05 / ScenariosInto the midterms.

Our probability-weighted view over the four weeks to 3 November, framed as the spread between the equal-weight S&P and the Nasdaq 100:

That weights to a modest edge for the broad market over the AI complex into the election — roughly 1 to 1.5 points of relative performance, with a fat left tail that hits both. It is a positioning tilt, not a conviction call. The asymmetry is in the composition: the 493 need only for nothing to go wrong for four weeks. The AI complex needs a revenue number to be revised back up.

06 / PositioningWhat we are doing.

• Holding the staples and defensive core we built through the summer. Thursday is what it was built for — the book had a portion that did not care what OpenAI's run-rate was.
• Not chasing energy on a 3% day driven by the commodity rather than by fundamentals. Oil above $100 is a regime we hedge, not one we buy into.
• Not re-shorting semiconductors into a 3.4% decline. We took our gains in June. A crowded short after a sharp down day has the same problem as a crowded long after a sharp up day.
• Watching three things into 3 November: the equal-weight/cap-weight spread for follow-through, the 10-year against its recent highs, and any second-source confirmation — or denial — of the OpenAI figure. Until it is confirmed on the record, it is a press report, not a fact.

The lesson of the session is compact. When a market is held together by one theme, a single headline moves everything. When it starts to respond to two headlines in two directions at once, the theme has stopped being the market and become a position inside it. We have spent five months arguing that the other 493 were waiting for permission. On Thursday, for one afternoon, they got it.