That would certainly be a more apt concern, but the article is specifically talking about the network bandwidth:

The article is, (in my opinion) unfairly and intentionally, confusing the non-consensual download energy use with the optional inference energy use.