You wouldn't be the one paying for it, like PyPi you would upload your package to them.

When you bootstrap your package ecosystem using git forges for hosting there's no index at all so I'm not really sure what the argument is.

The target audience for the article are people building these systems, so the people who would have to pay for the centralized infrastructure.

With git there's a sync protocol built-in that allows anybody who's interested to pull a copy of the index (this shouldn't be the default distribution model for the package clients, but anybody who truely wants it can pull it). PyPi is keeping their index private and you'd have to scrape all data through a heavily rate-limited API.