Model inference compute over model lifetime is ~10x of model training compute now for major providers. Expected to climb as demand for AI inference rises.
replyFor sure and growth also costs money for buying DCs etc.
replyThey are constantly training and getting rid of older models, they are losing money
replyWhich part of "over model lifetime" did you not understand?
replyThat's not a sufficient condition for profitability if both inference and scaling costs continue to increase over time.
reply