It's frustrating that there's no way for people to (selectively) mirror the Internet Archive. $25-30M per year is a lot for a non-profit, but it's nothing for government agencies, or private corporations building Gen AI models.

I suspect having a few different teams competing (for funding) to provide mirrors would rapidly reduce the hardware cost too.

The density + power dissipation numbers quoted are extremely poor compared to enterprise storage. Hardware costs for the enterprise systems are also well below AWS (even assuming a short 5 year depreciation cycle on the enterprise boxes). Neither this article nor the vendors publish enough pricing information to do a thorough total cost of ownership analysis, but I can imagine someone the size of IA would not be paying normal margins to their vendors.

I mean no disrespect by this but when I saw the headline on HN I immediately thought it was about Scott Adams the text adventure guy. And then I started watching the video and was a bit confused at first before it all clicked.