AI value may accrue to consumers, not creators
The guest argued that the immense capital expenditure on AI models may ultimately benefit consumers rather than model creators due to rapid hardware depreciation and intense commoditization.
The argument
The guest compared the current AI build-out to the fiber-optic and railway booms, noting that AI hardware has a much shorter useful life (5-7 years) and depreciates faster. He suggested that intense competition among multiple models will drive prices down to commodity levels, preventing creators from earning supernormal returns.
The thesis, stress-tested
✓ What validates it
- ✓Rapid price cuts for API access to frontier LLMs
- ✓Shorter replacement cycles for AI data center hardware forcing massive write-downs
▸ Risks discussed
- ▸A single dominant AI could achieve a winner-take-all monopoly
- ▸Enterprise adoption of premium models could remain highly price-inelastic
Hear it yourself
"Do you think that to some extent, maybe the valuation could be sort of justified there because of those? And it's it's sort of interesting. Like if you think we we've talked about this too with you before Toby. We we we talked about buffet's you know apple trade and I think when he bought the stock back in whatever it was twenty fifteen twenty six I think 2016 you know the p e of apple was around 10 to 12 times I mean apple today is like you know over 30 times trailing earnings so it's just amazing that if you think"