Data Center Water Consumption: A Lack of Transparency (2021)
nature.com
nature.com
The section on direct evaporative cooling was interesting and was what I was expecting but then you have evaporative losses seemingly bundled in with the upstream water usage of their power sources. I have probably 20 or 30 gallons of water in my hydronic heating system but because it's a closed loop system it doesn't actually consume that much water year over year unless it's leaking. So it's certainly possible to move a lot of BTUs in or out of a space using comparatively little water.
I was expecting a lot more meat on the bone in the form of some analysis of speculative numbers in the conclusion.
Because the Columbia River has a requirement that its use must be net zero, Google (and others, but the article is about Google) can't use river water for evaporative cooling. https://www.google.com/about/datacenters/locations/the-dalle...
Google bought the water rights from an aluminum smelter... which allows it to withdraw 3.9 million gallons per day from the groundwater. ( https://www.registerguard.com/story/news/2021/10/25/big-tech... https://columbiacommunityconnection.com/the-dalles/google/da... )
As Use of A.I. Soars, So Does the Energy and Water It Requires
This seems part of a coordination between various legacy media interests (print organizations, academics and research journals, and others) aimed at running the AI is Evil narrative. There are many, many, things to criticize about the way the world works, water consumption by datacenters is tone deaf whinging about the thread count of your throne's seatcushion.
Datacenters and computers are some of the most highly efficient and effective uses of power humanity has ever come up with. That single, highly efficient use of resources is the basis of modern society. For all the ills and evils, it's allowed flourishing and coordination in the effective and efficient use of resources at a global scale, making life a lot better for a lot more people than at any previous time in all of human history. Life is staggeringly better than 100 years ago, with absolute poverty and quality of life reaching what would have seemed absolutely miraculous levels by any other historical standards.
They've got to find better stories to tell, or we might stop taking them all seriously. AI can't replace these second-handers soon enough.
The externalities of data centers should be effectively measured, captured, and accounted for, same as anything else. I don't think, as things humans do, they're particularly drastic consumers; that doesn't mean they shouldn't be explored and understood, and there are probably worthwhile conversations to have about how to economically incentivize positive uses of large-scale compute and, similarly, how to disincentivize negative ones we as societies don't want to see grow.