I think Ethan Ding's article about inference costs is thought-provoking.

· Bits and Bobs 8/11/25
  • I think Ethan Ding's article about inference costs is thought-provoking.
    • Absolute inference costs are going up, even as token cost declines.
    • The rate is getting cheaper but the volume is going up.
    • Plus, people keep showing a clear preference for the highest quality cutting edge model vs last year's best that is now cheap.
    • The same dynamic happens for iPhones.