To The Moon

$Microsoft(MSFT)$ up 7.70% after hours. $Meta Platforms, Inc.(META)$, down 7.10%. Same evening, same underlying fact — both companies are spending unholy amounts of money on AI — and fifteen points of daylight between them.

Read those two numbers alone and you'd conclude the market turned on AI capex. It didn't. It did something more annoying: it started sorting.

Start with who flipped the switch. The July FOMC didn't move rates and didn't change a word of the guidance. Warsh spoke, and within thirty minutes the ten worst names on the tape were almost all AI. The S&P 500 closed down 1.52% at 7,316.15, the Nasdaq Composite down 1.74%, $Invesco QQQ(QQQ)$ down 2.04% — the worst day since April 2025. What actually came loose wasn't the policy rate. It was the assumption that the Fed will catch the market. While that assumption holds, every distant cash flow queues at the same discount rate. Once it slips, the queue re-forms.

The first thing sorted was capex. $Microsoft(MSFT)$'s spending has Azure's revenue curve running behind it — money goes out, receipts come in — so it counts as investment. $Meta Platforms, Inc.(META)$'s return is still parked in guidance, so it counts as expense. Identical cash out the door, and the line item gets chosen by whoever reads the books, not by whoever writes the checks. There are your fifteen points.

I know where the strongest objection sits: $Meta Platforms, Inc.(META)$'s compute has been earning its keep inside the recommendation stack for years, that return is real, and it simply doesn't reduce to one quarterly slide. I believe it. The market this week isn't asking whether the return exists. It's asking whether it reconciles now. In March those were the same question. Last night they weren't.

The second sort was memory, into fundamentals versus leverage. $Micron Technology(MU)$ −9.94%. $SanDisk Corp.(SNDK)$ −7.32%. $Roundhill Memory ETF(DRAM)$−6.11%. And one box that doesn't tick: $SK hynix(SKHY)$ the name that cratered two days ago on the CXMT news, fell 2.60%, while $Micron Technology(MU)$ and $SanDisk Corp.(SNDK)$, with no direct bad news at all, fell hardest. The leveraged tape gives it away — $Direxion Daily Semiconductors Bear 3x Shares(SOXS)$ up 16.35% in a session, $Direxion Daily Semiconductors Bull 3x Shares(SOXL)$ down 16.02%, $CSOP SK Hynix Daily (2x) Leveraged Product(07709)$ in Hong Kong down 14.13%. When the resilient name is the one that already blew up and the broken ones had no news, that isn't repricing. That's leveraged money being carried out. As for the two bearish headlines everyone kept repeating: senators warning Apple off Chinese memory chips is, on any horizon longer than a week, good for $Micron Technology(MU)$, $SanDisk Corp.(SNDK)$ and $SK hynix(SKHY)$; and a CEO's share sale, on its own, has close to zero predictive power — it's just easy to say in a panic.

The third sort was the quietest and the one worth remembering. $NEBIUS(NBIS)$ −12.65%, $CoreWeave, Inc.(CRWV)$ about −9%, triggered not by earnings but by the overnight cost of insuring their debt. Equity and credit went into different bins. These businesses are asset-heavy, levered, and expand off prepaid contracts; every step up in funding cost takes a slice out of the expansion. Equity can change its mind overnight. Insurance doesn't.

So yesterday was neither "AI is over" nor an overreaction. It was a sorting — and a sorting is harder to trade than a selloff. You can wait out a selloff. With a sorting you have to know which bin your ticket landed in.

I read these two days in memory as a leverage flush rather than the end of the supercycle, and saying that out loud means stating the cost with it: nobody can tell you how deep the flush goes or how long it drags, and the NAV leveraged products surrender in tape like this doesn't come back. I'll commit to direction. Not to size.

$Apple(AAPL)$ and $Amazon.com(AMZN)$ hand in their papers tonight. They aren't just two more prints. They're the next input to the sorting machine — watch which side they land on.