In an effort to address the challenge of global climate change, Microsoft has published the findings of its two-year study that quantifies energy and water use and greenhouse gas emissions across four datacenter cooling techniques.

The study was a full lifecycle report that exhaustively detailed the many resources needed to create the semiconductors, virtual machines, servers, cooling and other required data center technologies to help IT architects design facilities that consume less natural resources. The study even included variables like the toll of extracting natural resources, transportation and eventual end-of-life disposal.

Published in Nature, the study examined four cooling technologies: Air cooling, cold plates, one-phase immersion and two-phase immersion.

Air cooling is the legacy standard for data center cooling, but liquid cooling has recently found more interest because it can dissipate heat more efficiently.

“Artificial intelligence has abruptly thrust liquid cooling into the klieg lights because it is the only way to cool systems with massive performance, density, and scale—which is what defines the paradigm, AI system, today. But it also lowers the power needed for a given workload, both as an efficient cooling mechanism itself and by improving the amount of compute per watt delivered by the systems it cools. This is an important way we will hopefully be able to wrangle exploding power demands and manage – or even someday reduce – the environment impact of these systems,” said Guy Currier, VP and CTO of Visible Impact, a Futurum Group company.

Cold plate technology pumps a coolant in a loop to a flat container that is positioned on top of the chips in a server rack. One-phase immersion pushes cooling fluid around servers in submerged tank. Two-phrase immersion places servers in a tank filled with different fluids whose boiling point is low, so the vapor condenses to cool the tank.

The researchers found that cold plates and both forms of immersion cooling are more effective than air cooling. They can lower greenhouse gas emissions by 15% to 21%, and require 31% to 52% less water over their lifecycle than the air cooling used by legacy data centers.

The problem is that immersion methods use liquid polyfluoroalkyl (PFAS) substances, which are called “forever chemicals.” The dangers of PFAS have earned them regulation by both the U.S. Environmental Protection Agency and the European Union.

Yet the study made a valuable discovery to avoid this issue: “It was interesting to see that cold plates could be as good as the two immersion cooling methods,” said Teresa Nick, director, natural systems and sustainability for cloud operations and innovation at Microsoft and the study’s co-author.

The study also calculated that using renewable energy sources rather than a traditionally powered electrical grid reduced greenhouse gases by 85-90%.

Many environmental impact studies are conducted after a facility is constructed. In contrast, said Husam Alissa, director of systems technology in Cloud Operations and Innovation at Microsoft and the study’s leader, “We’re advocating in this paper for the use of life cycle assessment tools to guide engineering decisions early on and also sharing the tool with the industry to make adoption easier.”

Large data center operators—most notably the cloud leaders, Microsoft, Amazon and Google—have been under pressure to reduce the enormous environmental impact of their massive data facilities. Some forecasts have indicated a doubling of power demands by data centers by 2030; a Goldman Sachs study predicted a potential 165% increase. This is exacerbated by the growth of AI, which requires complex compute infrastructures that are voracious consumers of energy. The power demand and the resulting heat it expends creates a clear challenge to environmental safety.

In response, large tech vendors and other power consumers have pledged to be carbon neutral. Google has announced that it is working to achieve net-zero emissions across all of its business by 2030, with plans for 24/7 use of carbon-free energy. Microsoft too has pledged to be carbon negative by 2030. And Amazon has committed to net-zero carbon emissions by 2040 for all of its businesses.

To aid the effort, Microsoft is offering the methodology from its data center study to other companies via a research repository. The Microsoft research was also presented at the Open Compute Project Global Summit, held recently in San Jose, California.