Dario Amodei on Dwarkesh

AI
https://www.youtube.com/watch?v=n1E9IZfvGMA

The conversation spends over an hour in a very pedantic economics segment revisiting compute, revenue, and valuation assumptions without converging on a single model. Dario would have been a more interesting guest if he challenged Dwarkesh's assumptions about model capabilities and got more specific than just "put it in the context".

But that economics segment gives insight into Dario's perspective of model performance as he continues framing AI takeoff as a "country of geniuses in a data center," with his near-term economics timing of ~100Bbyend2026,around100B by end-2026, around 1T by end-2027 (with one- to two-year uncertainty) — roughly aligning with METR benchmark projections that put consumer apps near 2027, full B2B SaaS around 2028, and frontier LLM training across multiple data centers by 2029 with task length doubling very 128 day.

Published on February 14, 2026