Value shifts to physical AI bottlenecks
Lynn Alden argued that value in the AI era will accumulate in physical bottlenecks like semiconductors and memory rather than hyperscalers due to lower switching costs and high CapEx demands.
The argument
Unlike the 2010s internet giants that enjoyed high-margin network effects with low CapEx, today's AI hyperscalers face an expensive CapEx arms race and lower customer switching costs, which Alden suggests will depress their return on invested capital. Consequently, she favors the physical bottlenecks—such as the highly consolidated memory and GPU markets—and the end-users who gain productivity.
The thesis, stress-tested
✓ What validates it
- ✓Consistently high margins for memory and GPU manufacturers
- ✓Declining return on invested capital (ROIC) for major hyperscalers
▸ Risks discussed
- ▸Overvaluation during vertical price runs
- ▸Cyclicality of data center CapEx spend
Hear it yourself
"You have thoughts on the models, anthropic OpenAI, which are privately traded and you you have a engineering background and at least when it comes to cryptocurrency are extremely in the weeds on on technical matters. I wanna ask you about semiconductors. I don't think we you and I in our many interviews we've done over the years, I've ever really talked about semiconductors as an investment opportunity. What what conclusions do you draw, and what do you think about that space? I'm tactically bullish here and there. I've not captured as much as I would have wanted to. There's some some stocks that's rode for five, ten years."
06:45
AFFILIATE LINK · ZORTIX MAY EARN A COMMISSION · NEVER A RECOMMENDATION TO TRADE