My main thoughts on an AI bubble remain where they have been for years:
- The same thing—dot-com rerun—can’t happen twice, because the first time it happened it wasn’t influenced by its own history. Something else will happen. This goes for most phenomena that humans participate in.
- We will be compute-constrained forever.
- The best product-market-fit for this technology is software development.
- Models are commodities.
- The hyperscalers are making real money selling compute.