No it's not. It never makes any distinction between training and inference. It just lumps it all together as "running" the models.

But he's not wrong. Training + inference on free customers is the black hole here.

Except it's not. And the footnote that might be expected to clarify turns out to be a joke footnote.