←back to thread

Are OpenAI and Anthropic losing money on inference?

(martinalderson.com)
507 points martinald | 1 comments | 28 Aug 25 10:15 UTC | HN request time: 0.269s | source
1. SamInTheShell ◴[28 Aug 25 15:35 UTC] No.45053519[source]▶
Those same H100's are probably also going to be responsible for the R&D of new model versions. Running a model is definitely cheaper than training them.