The problem of this "depreciation" rationale is that it presumes that all the cost is in training, ignoring that actually serving the models is also very expens

The problem of this "depreciation" rationale is that it presumes that all the cost is in training, ignoring that actually serving the models is also very expensive. I certainly don't believe they would be profitable, and vague gestures at some hypothetical depreciation sounds like accounting shenanigans. Also, the whole LLM industry is mostly trying to generate hype, at a possible future where it is vastly more capable than it currently is. It's unclear if they would still be generating as much revenue without this promise.