So the cost of caching should be a rounding error as well. If The Internet Archive can afford to cache vast swathes of the web, then surely the big AI companies can do so.