Cracks emerge in the AI CapEx trade
The AI infrastructure trade faces structural headwinds as hyperscalers reach CapEx limits, while open-source models and rising supply erode margins.
The argument
The speakers argued that hyperscalers are facing a prisoner's dilemma: they must continue heavy CapEx to support their private AI investments, yet this spending is leading to negative free cash flow and widening credit spreads. Simultaneously, the emergence of highly capable, low-cost Chinese open-source models (like Qwen 3) challenges the monetization assumptions of frontier models.
The thesis, stress-tested
✓ What validates it
- ✓Hyperscalers cutting CapEx guidance in upcoming earnings calls
- ✓Further widening of mega-cap tech credit spreads
- ✓Widespread enterprise adoption of open-weight models over proprietary APIs
▸ Risks discussed
- ▸National security imperatives may force continued government-backed AI spending regardless of economics
- ▸A short squeeze in mega-cap tech could temporarily rescue share prices
Hear it yourself
"And what makes me nervous about this AI drawdown is, like, so much money has piled in from the retail investors for months and months and months. We saw basically a gamma squeeze. And because of all the inflows so you have the inflows. Mike Green just talked about this on on a recent podcast, but these products have grown immensely, and then you could get this extra gamma from the inflows, which works in reverse on the outflows. And so this is exactly what we're seeing. The problem is is when something hits, like, a I think it's, like, a 120 implied vol, which is a lot of these stocks."
06:40
AFFILIATE LINK · ZORTIX MAY EARN A COMMISSION · NEVER A RECOMMENDATION TO TRADE