With all this compute will open weight models, that currently boast of a 10-20x cost advantage over OpenAI, Anthropic, Gemini models be able to provide inference at competitive rates, even if they have 0 margins on the tokens? Wouldn't big labs over subsidize the large scale inference?
I think Chinese models would be the only ones that will survive the price war given state sponsored GPU clusters and energy access.
WDYT?