The yield broke to a 2007 high and the chips paid first. Software didn't.

No pre-open or mid-day edition was published today, and none has been since Sep 17. This close is graded against the last thesis this desk put on the board.

The standing thesis · exposure held, mechanism did notThe Sep 17 close called that session "a duration tape wearing a rally's clothes" and wrote down what would prove it wrong: if the 10-year resumed its climb, the cohort that led would be the cohort most exposed, and the thing to watch was whether software joined or kept lagging. Today tested both halves at once. The exposure call held: the 10-year pushed to a 2007 high, and semiconductors went from 6 for 6 to 0 for 6. The mechanism call did not hold. If a single duration trade were driving the board, software would have fallen with the chips. It led instead. This desk is recording that half of its read as wrong, in public, because that's what the receipts are for.

What led · security and softwareCybersecurity went 3 for 3, averaging +3.82%: CrowdStrike +4.97%, Okta +4.45%, Zscaler +2.03%. Software went 5 for 6, averaging +1.94%: Palantir +3.68%, ServiceNow +2.76%, Unity +2.30%, Salesforce +1.84%, Datadog +1.62%. Energy went 3 for 4 at +1.23% on average, led by ConocoPhillips +2.25%. IonQ's +4.42% came with a company-specific headline, covered below.

What lagged · the chips, every one of themSemiconductors went 0 for 6, averaging −1.30%: Micron −2.22%, AMD −1.47%, Nvidia −1.40%, TSMC −1.20%, Intel −1.02%, Qualcomm −0.52%. Coinbase fell 1.46%. Finance went just 1 for 4 (JPMorgan −0.73%, Bank of America −0.36%). For a second straight close, banks didn't confirm a rising-rate tape that is conventionally read as helping them.

The read · a sorting, not a sell-off26 of 49 names advanced and 30 carry a BUY bias, on a day the S&P 500 fell 0.8%. The index weight sat in the semiconductor complex, and that's what sold. The honest description of Wednesday isn't that growth got taxed. The yield taxed hardware and memory while software and security were bid, and a single discount-rate lever doesn't explain that sorting.

What would say this is wrongOne session of software strength against a 17-basis-point yield spike is thin evidence. If the 10-year keeps rising and software rolls over to join the chips, today was a lag, not a split. If software holds while yields hold near 5.1%, the board is pricing something other than the rate. That's an observation about today's structure, not a forecast about tomorrow's direction.

Rates, Hormuz, a summit on the calendar and a cluster of high-profile shorts all landed on the same session. Every item below is attributed to a named outlet, and every market link is stated as observation rather than cause.

Rates & the Fed · hot activity data, a hawkish governor, and a 19-year highStocks & News reported the 10-year at 5.10%, up 17 basis points to its highest level since July 2007, with the 2-year at 4.89%, and tied the move to "renewed inflation fears" after stronger private-sector activity data (Stocks & News · Sep 23). FXStreet reported September flash PMIs of 58.4 composite, 57.0 manufacturing and 58.7 services, together with Governor Barr's view that more hikes would likely be needed (FXStreet · Sep 23). Bloomberg headlined that five-year yields breached 5% for the first time since 2007. CNBC described the 10-year as having rocketed to a 19-year high, and Zacks framed the day as "Fed's Higher-for-Longer Stance: 3 Picks as AI, Oil & Markets Diverge." The thread: this is the fact the whole session traded against, and it came from the economy rather than the central bank. The group most sensitive to it, the semiconductor complex that led Sep 17 on a falling yield, gave that back on a rising one. The Zacks headline got the word right: the market diverged.

Energy & geopolitics · Hormuz at the UN, $100 Brent, and record dieselStocks & News reported that Iranian President Pezeshkian told the UN his nation "will not allow freedom of navigation through the Strait of Hormuz while sanctions and a U.S. blockade remain in place." It also reported WTI at $92.50, Brent at $103.40 and a record diesel price of $6.52. Benzinga headlined the President threatening to "annihilate" Iran in his own UNGA remarks. 24/7 Wall St. asked the supply question directly: "If Hormuz Stays Shut, How High Can Exxon Go?" The thread: energy was one of the few groups on the board that were meaningfully green. ConocoPhillips +2.25%, Exxon +1.63% and Chevron +1.53% put the sector at +1.23% on average. Oil also feeds the inflation input that drove today's yield move, which ties this item back to the first one. Our sentiment engine still reads Exxon neutral (3 positive, 1 negative), so the price moved further than the news flow did.

Trade & China · the summit is tomorrow, and a memory rival just showed upStocks & News noted that the focus turns tomorrow to the Trump–Xi Jinping summit in Washington, and 24/7 Wall St. ran "How the Trump-Xi Meeting Could Move Nvidia, Boeing and Your 401(k)." GuruFocus reported "Micron Falls as China Adds a 24-Gigabit Memory Challenger." Separately, Reuters reported Nvidia's chief executive saying on a podcast that AI firms should not get regulatory waivers. The thread: the group that went 0 for 6 is also the group with the most direct China exposure, both as a market and as a competitor. Micron, the worst name on the board at −2.22%, drew a China-specific headline on the same day. Software and cybersecurity, the day's leaders, have far less exposure to a summit outcome. That fits today's sorting, but we're not claiming it caused the split.

Corporate & people · a famous short, an insider sale, and a quantum deploymentProactive Investors reported that Michael Burry expanded short bets on Micron, Palantir and semiconductor stocks, and separately that IonQ shares rose on an NVIDIA quantum-computing deployment. 24/7 Wall St. reported an Nvidia director selling about $300 million of stock. MarketWatch described Micron as a "battleground" as the AI narrative shifts, and Benzinga reported that Martin Shkreli had also shorted Micron. Our sentiment basket reads Micron as bullish on the label but 2 positive against 3 negative on the count, and Nvidia as mixed to neutral across 15 headlines. The thread: the same reported short named both Palantir and the semis, and the tape split them. Palantir rose 3.68% while Micron fell 2.22%. IonQ's 4.42% gain came with a company-specific headline and wasn't part of the sector move. High-profile shorts in the headlines didn't move the software names today. In memory, the negative headlines lined up with the price.

What is actually being heldIf today's leaders are the book, it's a software-and-security position that rose against a 19-year-high yield. That's either real relative strength or a lag, and one session can't tell you which. If today's laggards are the book, it's the cohort that has now moved with the 10-year in both directions inside a single week.

Three threads land before the next reactionThe Trump–Xi summit in Washington touches the China-exposed chip complex directly. Hormuz rhetoric at the UN feeds oil, and oil feeds the inflation input. Sovereign yields reprice through the Asian and European sessions while US equities are closed. Any of the three can reach a position before it can respond. We're describing when news arrives relative to when a position can act on it, not calling a direction.

Not a weekend yetToday is Wednesday, so tonight's gap is a single session. But Thursday brings the summit and Costco's report after the bell, so Friday's close carries a denser file into the weekend than usual.

Estimates shown are consensus figures from our earnings calendar, not this desk's projections. The marquee print is Costco, Thursday after the close, a consumer read landing on a day already crowded with macro. Earlier today, General Mills (0.75 vs 0.717 est), Cintas (1.39 vs 1.35), Paychex (1.34 vs 1.32) and Cracker Barrel (0.99 vs 0.17) all reported above consensus before the open, per our calendar feed.

Overnight risk vetoThe input that drove today reprices while US equities are closed. Yields move through the Asian and European sessions before US hours. The leadership that split today, chips down and software up, can flip together or keep diverging, and there's no way to tell which from a single session. Define the level that says your thesis is wrong before the open, not after.