The rotation lasted one day

1. What the Fed actually did

The Committee left its target range at 3.50 % to 3.75 %, a fifth consecutive hold, on a 9-3 vote. Beth Hammack, Neel Kashkari and Lorie Logan all preferred a quarter-point increase. Logan had already argued in Houston on 16 July that modestly higher rates would better balance the outlook, adding that inflation had been too high for too long.

Three dissents in the same hawkish direction ties September 2016 for the most in almost a decade. It is worth being exact here: 2016 also had three, so this equals rather than exceeds it, and the September 2019 meeting had three dissents pointing in different directions. The signal is the unanimity of the objection, not the count.

Two things are easy to miss. The first is the chair. This was only Kevin Warsh's second statement, and his message afterwards was notably firm, including the line that there is no soft inflation target. The bond market did not take it as reassurance. Yields rose.

The second is the data. Core PCE has run above the 2 % goal continuously since April 2021 and printed 3.4 % year-on-year in May, the highest since late 2023. The statement itself does not use the five-year framing, but the series does.

2. The market was already positioned for a hike

A common reading of the sell-off is that investors were caught out. The evidence says otherwise. Futures pricing had put the probability of a hike by the September meeting at roughly 82 % as early as 27 July, and after the meeting a September move became the single most likely outcome.

So the 29 July fall was not surprise. It was confirmation being priced. That distinction matters, because it removes the most comforting explanation available, namely that the market overreacted to news it had not seen coming.

3. The rotation was real, and then it was not

Through most of July the broadening was genuine. Money left high-multiple semiconductors for defensives, industrials and value names, and the equal-weighted S&P 500 reached a record on 28 July. Morgan Stanley's Mike Wilson had been describing exactly this move since 6 July, framing it as a shift from chipmakers toward hyperscalers, consumer discretionary, transports and biotech.

On 29 July industrials fell 3.42 %. The sector that had been receiving the rotation took one of the heaviest hits of the day.

This is the part worth sitting with. A rotation is a bet that money is moving somewhere safer, and for three weeks that bet worked. It did not survive contact with a hawkish hold, because the destination of the rotation is not immune to the discount rate either. It is only less exposed.

4. Duration explains part of the tech drawdown, not all of it

The standard explanation is duration, and it is worth stating precisely because it is usually stated loosely. A share is worth its future cash flows discounted to today. Raise the discount rate and every future dollar shrinks, but a company earning most of its profit this year barely notices, while a company whose profits arrive in 2032 gets repriced hard.

That mechanism is real. It is not, however, what drove the memory selloff, and it would be sloppy to pretend otherwise. On 27 and 28 July, reporting confirmed that China had begun mass production of domestic immersion DUV lithography tools, with scanners heading to SMIC, Hua Hong and CXMT. Samsung and SK Hynix each fell more than 13 % on 28 July, and Micron fell almost 10 % in the same session.

That is a supply and competition shock, not a discount-rate effect. Duration and oversupply happened to point the same way this month, which makes the single-cause story tempting and wrong.

5. Where the capacity deals point

Two agreements are worth reading carefully, because they are being cited as evidence for conclusions they do not quite support.

Samsung and Broadcom announced an arrangement on 25 July valued above $200 billion through 2030. It is a five-year memorandum of understanding rather than a binding decade-long contract, and it spans HBM4 and HBM4E memory, sub-2nm foundry capacity and packaging together. It is not a packaging-only deal, so it does not by itself establish that packaging is the binding constraint. Intel and Lens Technology are separately collaborating on glass-substrate advanced packaging, with a pilot line running.

The fair conclusion is narrower than the one usually drawn: firms are contracting years ahead across several parts of the stack at once, which tells you they expect scarcity somewhere and are unwilling to guess where.

6. The financing structure we keep watching

Away from the headlines, a credit structure has been building. Lending facilities secured by GPUs as collateral have moved from novelty to investment grade, including a bankruptcy-remote vehicle used by CoreWeave and smaller non-recourse facilities elsewhere. JPMorgan has sized data-centre securitisation at $30 billion to $40 billion a year across 2026 and 2027.

The shape is familiar. When the asset securing a loan is the same asset whose price is falling, leverage stops multiplying returns and starts accelerating losses. Chinese DUV progress makes falling hardware prices a live scenario rather than a hypothetical. This is not a systemic event today. It is the kind of thing that becomes one quietly.

7. Four things on the desk

1. 01Energy, which reversed under everyone. Brent jumped almost 7 % on 29 July to settle near $89 after an Iranian ballistic-missile attack on US forces, US and Saudi strikes on Iran-backed militias in Iraq, and a Houthi missile strike on a Saudi tanker. Any argument resting on cheap energy supporting industrial and consumer margins was written for a market that no longer exists.
2. 02A September move. Three dissents plus a firm chair plus futures pricing a hike as the most likely outcome. The debate is no longer whether the risk leans hawkish, but what a confirmation does to multiples.
3. 03Memory oversupply. Chinese domestic DUV is entering production at modest volume, roughly five tools this year against about twenty planned for 2027, and each still needs line qualification. It compresses the technology gap that export controls were meant to preserve, and legacy and mid-tier chip pricing is where that shows up first.
4. 04The gap to consensus targets. The S&P closed at 7,316.15 on 29 July. Morgan Stanley raised its year-end target to 8,000 on 13 May, roughly 9 % above that level, in the same note that set a mid-2027 target of 8,300. Mike Wilson has separately flagged a possible dip toward 7,000 first. Both can happen in either order.

8. What we take from it

The honest summary of this week is that a comfortable narrative broke in a single session. Earnings outside technology really are good, and that has not changed. What changed is that two of the supports underneath the constructive case, cheap energy and a patient central bank, went away within a day of each other.

We are not drawing a directional conclusion from two sessions, and neither should anyone else. What the week does establish is narrower and more useful: the rotation that looked like a safe destination is not a hedge against the discount rate, and the energy tailwind that was flattering margins is contingent on a geopolitical situation that just moved against it.

Weeks like this reward one thing, which is knowing what you already hold, and in particular which duration, which supply chain and which commodity assumption sits inside positions that were never chosen for those exposures.