CHARBONNEAU – AI is hungry for electricity, thirsty for water

WE HEAR A LOT about the promise and peril of artificial intelligence but little about mundane things such as how to feed the lumbering savant.

The promise is that generative AI will use supercomputers to create and interpret text and audio; that AI generated video will permeate countless products and services. Dreary jobs will be eliminated, freeing the mind for creative endeavours.

The peril is that misinformation will be amplified, biases entrenched, and privacy eroded. Automation will lead to job loss; deepfakes will result in political instability; and there’s the dreaded singularity in which a powerful super-intelligence will surpass humans’.

For that to happen, the world will need more supercomputers, more data centres, more energy to power them, and more water to cool them.

Data centres are already energy hogs but generative AI will fuel the need. By 2026, AI could contribute to a doubling of global electricity use by data centres. That’s about twice what Canada uses in a year.

An investment by Canada’s feds of $CA2.4 billion pales compared to the US$100-billion supercomputer planned by Microsoft along with its partner OpenAI. Also, Meta plans to increase spending this year to US$40-billion to support AI.

Supercomputers create a lot of heat. That heat can consume a lot of water for cooling.

Google’s data centres, alone, consumed enough water to supply all of metro Vancouver last year. Only about 6 per cent of Google’s freshwater was returned to the source; the rest lost though consumption and evaporation.

Kamloops’ data centre on Bunker Road has solved the water problem through use of glycol-cooled air conditioning units and under-floor airflow distribution.

This allows for efficient use of water at the Kamloops centre. A measure of water use is to compare water use in litres to power consumption in kilowatt-hours. While the average data center uses 1.8L per 1kWh, Kamloops uses only 1.07L per 1kWh.

Because Kamloops data centre is serviced by BC Hydro, its electricity consumption is 100 per cent renewable energy. Rooftop solar panels reduce their energy consumption.

However, only 30 per cent of the world’s electricity comes from green sources, most of that from hydroelectricity.

A Canadian company called Denvr Dataworks, based in Calgary, has come up with an innovative solution to cooling.  Instead of water, the company uses immersion cooling, in which computer chips are submerged in a chemical mix that transfers heat away from the chips. From there, the heat is sent to a cooler where it is exchanged with outside air.

Generative AI could make data centres even thirstier. With climate change causing drought in some areas, water consumption is a prime consideration in locating data centres.

Also, climate change will make data centres hotter, which will require more water cooling and air condoning.

By 2027, AI computing could be responsible for up to 6,600 billion litres of water withdrawal –enough to supply the United States for a year according to Professor Shaolei Ren, at the University of California who studies water use in data centres.

While we should marvel at the potential of AI and worry about its dangers, we can never overlook the environmental impact of the servers that feed the pseudo-intellectual beast.

David Charbonneau is a retired TRU electronics instructor who hosts a blog at http://www.eyeviewkamloops.wordpress.com.