Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

Ai does us a crap-ton of water. Most data centers use closed loop liquid cooling with heat exchangers to water cooling. (At least all the big ones like Google and Amazon do)

I’m curious what evidence you think you’ve seen to the contrary. from my side, I used to build data centers and my friends are still in the industry. As of a month ago I’ve had discussions with Google engineers who build data centers regarding their carful navigation of water rights, testing of waste water etc.



> Most data centers use closed loop liquid cooling with heat exchangers to water cooling.

If these data centers are so water efficient, please explain the Dalles data center use > 25% of their water supply?

https://web.archive.org/web/20230130142801/https://centralor...

https://web.archive.org/web/20251014013855/https://www.orego...


The Dalles data centers use a large fraction of the water supply of The Dalles because the data centers are extremely large and the town of The Dalles is of negligible size. It is also true that the paper mill of Valliant, Oklahoma uses 50 million gallons of water per day and that the town of Valliant, Oklahoma, population 819, uses less than 1% of that amount, so the paper mill can be said to be using > 99% of the local water supply but this is also a meaningless comparison.


Did they say it was efficient? The "closed loop" is only one part of the system that cycles water between the heat exchanger and the building/servers.

The second part of the system is an open loop that uses water to cool the closed loop at the heat exchanger.


They implied that DCs somehow save water because of being closed loop. The closed loop is a red herring, since the outer loop dumps potable water.


So we'll move the datacenters from the tiny town to just outside of a giant city which will probably move that percentage down to only a few percent if even that. Problem solved!

You're looking at the wrong metrics to compare here if we're trying to just gauge how efficient a datacenter is or is not. This metric could be useful if the datacenters are attached to the municipal water system and thus begin to be a massive load compared to what was originally planned/built, but in terms of understanding the total water use compared to other industrial users its kind of a meaningless statistic.


Parent says consume, you write use.

I’ve been unclear on this. What datacenter out there is using an open loop cooling system that does not return the water after cooling for other uses?

It seems extremely inefficient to have to filter river water over and over then to dump it into the ground so deep it doesn’t go back to getting into an aquifer.


The water that is "used" by data centers is evaporated. That's where it goes. The sky.


So you are saying it’s an open loop, and can we not calculate when these million of gallons of water are going to come back down?


As is always the case when discussing systems, the answer changes depending on where you draw the system boundary. In some cases you would expect water to fall as rain in the same watershed where it was drawn. This is the case for example of water "used by" California rice fields that are irrigated by flood. In other cases, you can expect the water to disappear into a distant system. This would be the case for water drawn from fossil aquifers.


That water becomes rice.

Does the water that cool datacebters become AI? Do we ship water bearing AI around the world?




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: