←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: 0.212s | source
Show context
pityJuke ◴[28 Aug 25 13:26 UTC] No.45051892[source]▶
From https://www.theverge.com/command-line-newsletter/759897/sam-..., Sam Altman said:

> “If we didn’t pay for training, we’d be a very profitable company.”

replies(7): >>45052382 #>>45052398 #>>45052467 #>>45052890 #>>45053259 #>>45053709 #>>45056275 #
1. tootie ◴[28 Aug 25 14:10 UTC] No.45052382[source]▶
Yeah I've seen the same sentiment from a few others as well. Inference likely is profitable. Training is incredibly expensive and will sometimes not yield positive results.