To The Moon

$Apple(AAPL)$ beat on revenue for its fiscal third quarter and fell 5.74 per cent after hours. Once the same call had wrapped up, $SanDisk Corp.(SNDK)$ rose 4.61 per cent in extended hours and $SK hynix(SKHY)$ 3.76 per cent.

Between the two sits a single remark from Tim Cook: memory supply is tight, it will hit next quarter, and the company is stockpiling hard to get ahead of significant supply constraints.

On Apple's books that sentence is a cost. On a memory maker's books it is revenue. One fact, two income statements, opposite signs.

Before we put the whole 5.74 per cent on that one line: the same results missed on iPad and on services revenue, and sales in Greater China are slowing. Any one of those three is enough to take a stock down on its own. Which slice belongs to which, I can't separate, and neither can anybody else. But one thing does separate cleanly. A weak iPad quarter doesn't turn into somebody else's revenue. Nor does a slowdown in Greater China. Only the memory line does. It is the one number in these results that travels off Apple's books and onto someone else's.

So the point isn't how far Apple fell. It's that rising memory prices had until now been a sell-side assertion propped up by second-hand evidence — Samsung's roughly thirteen-fold jump in profit, and so on. On Thursday night a buyer signed for it. And not just any buyer: the one with more bargaining power than anybody else alive.

I ended the last piece with a question — which side Apple and Amazon would land on. Thursday answered it.

$Amazon.com(AMZN)$ rose 9.43 per cent after hours. What lifted it wasn't the capital expenditure: the same results took the 2026 figure up to roughly US$220 billion, a number that three days earlier would have knocked the stock down. What lifted it was AWS growing at its fastest rate in five years. Those two numbers have to be read together — the scale of the money going out is on the table, and so is the speed at which it turns into revenue. When the two reconcile, US$220 billion gets booked as investment. $Apple(AAPL)$ ran the other way: it beat on revenue, and then had next quarter's guidance pinned down by a variable it doesn't control. The one spending harder went up. The one with the tidier books went down.

The market never objected to the spending. It objects to spending that doesn't reconcile.

$Microsoft(MSFT)$ rose 15.51 per cent in the regular session, and one set of results choked off a six-day losing streak on the Nasdaq Composite. The same day, $Meta Platforms, Inc.(META)$ closed down 7.95 per cent — same money, same destination, and the ticket written to one name only. Put those two side by side and the sorting machine from the last piece hasn't been overturned. It has been reinforced.

The rebound in memory was violent. $Micron Technology(MU)$ +18.36%, $SanDisk Corp.(SNDK)$ +25.99%, $SK hynix(SKHY)$ +17.52%, $Roundhill Memory ETF(DRAM)$ +16.70%. On the leveraged tape: $CSOP SK Hynix Daily (2x) Leveraged Product(07709)$ +55.13%, $Tradr 2X Long SNDK Daily ETF(SNXX)$ +51.88%, and $iShares Semiconductor ETF(SOXX)$ down 26.29% — the same fund that had gained 16.35% the day before. Two days, one round trip, and a slice taken off both ends of the leverage.

That run needs its timing straightened out. By the time Cook spoke, the US market had closed. The daytime move was pushed by Samsung's profit and by a bounce off oversold levels — short positions being carried out. Cook's sentence lands only on the 4.61 per cent and the 3.76 per cent in extended hours. The first half was leverage draining out; only the second half was a buyer signing. The two happened back to back. They aren't the same event.

Let me make the strongest objection for you: the shortage has now been confirmed out loud by the largest customer, so this is genuine fundamentals, not a short squeeze. I believe it. But fundamentals give you a slope; they can't give you 26 per cent in a session. Twenty-six per cent in a day, fifty-five per cent in a day — that's leverage's handwriting. I'll commit to direction. Not to size.

One box doesn't tick. $NVIDIA(NVDA)$ rose only 2.65 per cent. Across a V-shaped rebound in AI, the sharpest gains went to $Advanced Micro Devices(AMD)$ 13.00 per cent, the memory complex, and $NEBIUS(NBIS)$ 27.13 per cent — the leader was the weakest of the lot. That isn't conviction returning. What returned was beta.

There's a second thing in that same box. $NEBIUS(NBIS)$ fell 12.65 per cent on Wednesday and rose 27.13 per cent on Thursday. Its US$775 million of debt funding was read as leverage risk on the Wednesday and as capital efficiency on the Thursday. Not one word of the fact changed; only the column it was filed under.

Line those three sums of money up and they're the same manoeuvre. Cook's memory purchases: a cost on one side, revenue on the other. Amazon's US$220 billion: bookable as expense or as investment. Nebius's US$775 million: risk one day, efficiency the next. The hand doing the bookkeeping shakes far more than the facts do.

But the memory line differs from the other two, and that difference is what's genuinely new about these two days. Whether Amazon's US$220 billion is filed as expense or investment moves only Amazon's own valuation. Whether Cook's money is a cost or revenue is two companies splitting one pot. The default setting of this AI cycle so far has been that everybody wins together — compute up, software up, power up, and nobody has to settle the bill. A shortage is the first thing to put the bill on the table: one side's gross margin has to go down for the other side's revenue to go up. Zero-sum has entered the room.

The macro contributed background noise and little else: core inflation (core PCE) ran at 3.3 per cent year on year in June, a shade below May, and the Nasdaq 100 posted its best session since May 2025. But this move was lifted by earnings, not by liquidity. Someone said of Kevin Warsh's Fed that you watch the actions and ignore the talk. On that I agree.

Cook's sentence is already written into Apple's guidance for next quarter. Whose revenue it is also written into, the market is still counting name by name.