Discussion about this post

User's avatar
Seixas's avatar

The diffusion gap framing is a helpful way to separate model capability from actual revenue. The PNC figure of only about 2% of US households paying for AI sits awkwardly next to capex commitments sized for something much larger. I'd say the capex case rests less on power users spending more and more of the time, and more on whether enterprise adoption closes the gap, since enterprise contracts tend to scale differently from consumer subscriptions. Which of the two do you find more convincing as the main driver, and what would you watch to tell them apart?

Valentina's avatar

The browser and Maps examples make me think the next step goes further, and the average user may never "use" AI much at all. They'll hand a decision to an agent, and the agent does the using. If diffusion happens that way, the revenue shows up as agents spending on people's behalf rather than the 2% of households with a subscription becoming 20%, so who captures it: the model owners, or whoever owns the agent the user trusts?

5 more comments...

Ready for more?