Apparently, the idea is to construct massive neural networks to perform global-sized problems. But each "neuron" is tantamount to a separate mainframe computer, in order to simulate all the inputs and feedbacks and networking. And they need many thousands of them.
This is like removing a mountain with an army of spade-workers with pails. So, the facilities are gigantic in order to clobber the problem with brute force---before someone else gets there first. Thus, the crash-program atmosphere.
Here are some links to descriptions of the big center being built in Abilene, Texas.
On the other hand, at least the Crusoe facility's closed-water and energy-sourcing is a step in the right direction. If we're going to have all these monsters all over the landscape, they should at least be built in such a way as to not totally destroy the water tables and completely drain the grid.
The more I think about this, the more I am convinced that consumption of environmental water is not necessary. Most computer centers attain their system cooling by extreme air conditioning. Works just fine. People have built hangar-sized buildings for the storage of structural composite pre-preg which is maintained at icy temperatures by refrigeration (I've been in one). Food processing and canning plants maintain huge volumes of frozen goods using ammonia refrigerant.
I can understand that some electronic configurations might have high heat production per square inch, for which a water flow would be a desirable coolant (high specific heat, high density), but the water can be a closed loop, with heat exchangers passing it off to the ambient atmosphere.
There must be more to this story.
(I chuckle at their hat tip to "renewable" energy. I recall a recent winter when all the renewables were not showing up for duty, and poor Texas was being pinched by the cold.)
Apparently, the idea is to construct massive neural networks to perform global-sized problems. But each "neuron" is tantamount to a separate mainframe computer, in order to simulate all the inputs and feedbacks and networking. And they need many thousands of them.
This is like removing a mountain with an army of spade-workers with pails. So, the facilities are gigantic in order to clobber the problem with brute force---before someone else gets there first. Thus, the crash-program atmosphere.
Here are some links to descriptions of the big center being built in Abilene, Texas.
https://constructionreviewonline.com/crusoe-11-6b-to-expand-ai-data-center-campus-in-abilene-texas/
https://www.crusoe.ai/resources/blog/an-inside-look-at-the-abilene-ai-data-center
Yikes.
On the other hand, at least the Crusoe facility's closed-water and energy-sourcing is a step in the right direction. If we're going to have all these monsters all over the landscape, they should at least be built in such a way as to not totally destroy the water tables and completely drain the grid.
The more I think about this, the more I am convinced that consumption of environmental water is not necessary. Most computer centers attain their system cooling by extreme air conditioning. Works just fine. People have built hangar-sized buildings for the storage of structural composite pre-preg which is maintained at icy temperatures by refrigeration (I've been in one). Food processing and canning plants maintain huge volumes of frozen goods using ammonia refrigerant.
I can understand that some electronic configurations might have high heat production per square inch, for which a water flow would be a desirable coolant (high specific heat, high density), but the water can be a closed loop, with heat exchangers passing it off to the ambient atmosphere.
There must be more to this story.
(I chuckle at their hat tip to "renewable" energy. I recall a recent winter when all the renewables were not showing up for duty, and poor Texas was being pinched by the cold.)