From the article: 1.859 billion kWh. Divide by 1 year = 212 megawatt. Definitely not small. Not that gigantic either, compared to current day (US) AI/hyperscalers.
Storage densities these days are kinda amazing, it's not that much of a datacenter. Assuming you chunk it with triple redundancy, that's 220k TB raw. 10k 22 TB disks, you put them in one of those 4U 50 disk storage pods. 200 pods, 10 of those in a rack with some space left for a switch and power, so that's only 20 racks.
Adding more 9s is costly, and AI training is very suitable to be throttled and/or interrupted. I'm not talking about days or weeks of downtime, but these things are definitely being considered. Source: I'm working at a Google datacenter.
I was thinking the same, who on earth would want that, so did my technical colleagues, but since the AI summaries rolled out over here, non-technical folks I've asked about it actually seem to like it.
Not really, the whole point of this type of cloud offering is that it doesn't phone home to Google / the US. Sure, it will be left to the partner to support all of it, but it can't be shut down from one day to the other.
Not really. Even the most evil Google one can imagine would realise "your data" is the most valuable thing they possess, selling it would be bad for business. They're selling ads to the highest bidder who's looking for someone with a profile based on your data, but not your data itself.
Amazon will pay perhaps $50 for a 10TB disk, 10x it to cover for redundancy and the servers and datacenters to put the disks in, and you're looking at "only" 16 mil savings for an exabyte.
Belgium (my country) imports about 400% of its domestic use in LNG, to then sell it to neighbouring countries via pipelines. Doesn't make sense if LNG was too expensive.