top | item 47185823

(no title)

babelfish | 2 days ago

Inference is already profitable (training is not)

discuss

order

nradov|2 days ago

So what. In a highly competitive industry they can't keep selling inference unless they continually train better models. It's like saying my airline is profitable if you don't count the cost of buying new airplanes.

stingraycharles|2 days ago

This is a completely new market and players are currently burning money in order to capture market share. The money will stop flowing in at some point, but until then, you can’t compare it to an industry like aviation which is extremely mature and heavily optimized.

zippothrowaway|2 days ago

[citation needed]

OpenAI have made this claim and maybe it is with API pay-per-use (there's also good evidence eveb that is not if you dive into how much a rack of B200s cost to operate), but I'd be very sceptical that the free, $20 or $200 a month plans are profitable.

Then the questions are if the market will bear the real cost and if so how competitive OpenAI are with Google when Google can do what Microsoft did to Netscape and subsidize inference for far longer than OpenAI can.

Kubuxu|1 day ago

Just try using Claude with API for an hour and you will see that the subscriptions are definitely not profitable (unless they percent off “partying but dormant” is very high).