AI Is Hot, Quite Literally, But Microsoft's New Solution May Not Drain Local Water Supplies
According to Microsoft, microfluidics performed three times better than cold plates at heat removal.
Cooling has been one of the most controversial aspects of AI technology. Some of the largest critiques leveled at the way AI is designed involves the environmental cost that it may exact and whether the human race is in any position to pay it.
According to a March 2025 report, ChatGPT consumes the equivalent of 500 ml of water to generate 10-50 medium length responses. An older report from 2021, published in Nature posited that potable water makes up 57% of the source for AI coolant technology.
Microfluidics — Microsoft's New Cooling Technology
Amongst these concerns, Microsoft has come up with a new way of cooling its data centres known as "microfluidics". This involves minute channels being etched into the back of a silicon chip through which coolant fluid flows, cooling the chip directly.
The grooves are thin and have similar dimensions to human hair, leaving very little margin for error. They are inspired by the veins in a leaf or a butterfly wing.
The software giant said that it used AI to find out the spots on the chip that heated up the most and built channels there to have more liquid coolant circulate in those areas.
Performance
The most widely preferred method of cooling was through the use of "cold plates" which are separated from the chips by many layers and involve a similar principle of coolant being circulated inside it to pick up heat from the chips its sitting on.
"With all those layers of insulation removed and coolant directly touching the hot silicon, the coolant doesn’t need to be anywhere near as cold in order to do its job. That would save energy that won’t be needed to chill the coolant, while doing a better job than current cold plates," the release said.
According to Microsoft, microfluidics performed three times better than coldplates at heat removal. The company reported that the technology reduced the maximum heat of the silicon inside graphical processing units by 65%.
The company stated that it will also considerably improve performance and reduce power consumption. The software goansts said they can make their servers run faster and harder without worrying about the heat melting down their systems due to this technology.
AI's Impact On The Environment
In an April 2025 report, the Guardian highlighted Amazon's plans to build data centres in the Aragon region of Spain that uses more water than the inhabitants of that area uses as of now.
This lead to widespread objections from farmers and interest groups in the area. Amazon, Microsoft and Google all pledged to engage in "water-offsetting" to deal with the issue, pleading to be "water-neutral" by 2030.
Guardian said that this may not be enough as experts told the publication that water is a very localised natural resource, different from carbon.
They said that increasing access to water for a certain area does not account for all the water that communities much farther away have lost access to.
Water and soil pollution are also potential issues to be addressed as according to a market analysis report by Klean Industries, industrial cooling operations for data centres contaminate ground water and soil with "forever chemicals" that can have serious side effects on hormones.
