The AI Trade Just Split in Two, and Earnings Lost
SK Hynix grew operating profit more than sixfold and the stock fell 13 percent. A Chinese memory upstart with no such record closed up 450 percent. That gap tells you exactly what the market is buying now, and it is not profit.
On July 29, SK Hynix reported a more than sixfold jump in quarterly operating profit. The stock fell about 13 percent.
Two days earlier, CXMT, a Chinese memory maker with none of that track record, raised at least 8.6 billion dollars on Shanghai's STAR Market and closed up more than 450 percent on its first day of trading.
Same industry. Same week. Opposite verdicts. The market has stopped paying for AI earnings and started paying for two other things: proof that the demand lasts, and control of the supply.
## What got punished
Reuters gave two reasons for the SK Hynix selloff: delays in some advanced chip shipments, and mounting investor doubt about how sustainable hyperscaler AI infrastructure spending really is. Sit with what that implies. A sixfold profit increase was read as a miss, because it arrived with a question mark attached to its duration.
The same week, the purest expression of the AI-long thesis took damage. CNBC reported that the 24 billion dollar hedge fund founded by former OpenAI researcher Leopold Aschenbrenner is unwinding many of its trades after heavy losses on AI stocks and a bad short against software names.
I would not read that as the thesis being disproven. It looks more like crowded positioning getting flushed. But it removes a marginal buyer from exactly the names that need one.
## What got paid
Microsoft shares jumped 15.5 percent, their biggest one-day gain since 2008, on Azure growth and stronger-than-expected profits. That is contracted, recurring, visible demand. Duration, priced.
CXMT got a 450 percent debut on something else entirely: position. Chinese buyers now have a strategic reason to source memory domestically more or less regardless of price, which turns a commodity supplier into a protected one.
And the European Union committed 10 billion euros to seven AI gigafactories, each housing at least 100,000 advanced chips, more than doubling the bloc's compute. Nobody ran a return calculation on that. It is a sovereignty purchase, and it is now a category of AI buyer that does not care what the multiple is.
## The rule that is forming
Three buckets, and the market is now sorting them ruthlessly.
One: proven durable demand, booked and contracted. Microsoft.
Two: control of constrained supply, especially where a government is behind it. CXMT, the EU gigafactories, and most of what you see moving through our deal tracker on the sovereign side.
Three: everything growing fast simply because AI is growing fast. That is where SK Hynix sits this week, and it is where the repricing is happening.
The tell is that this discrimination showed up in memory first. Memory is the most supply-constrained and most commodity-like layer of the AI stack, which makes it the place duration risk gets priced before anywhere else. It is a leading indicator, not an isolated bad quarter.
## What to do with this
Read supplier results backwards. Before the profit line, find the shipment-timing line and the customer-concentration line. That is what actually moved SK Hynix, and the market got there faster than the headlines did.
Then watch the next round of hyperscaler capex guidance. If guidance holds firm while memory names stay punished, this was positioning and the selloff is a buying window. If guidance softens even slightly, SK Hynix was the early warning and the whole third bucket reprices behind it.