For all the attention they are currently receiving, data centers are not new. Over the last few decades, processing capacity, applications management, storage of digital information, and the like have moved to the cloud, but physical infrastructure is still required. Instead of living on-site, an organization’s countless routers, switches, firewalls, storage systems, and servers are housed now in massive facilities offsite, where they are combined with those of numerous other organizations.
What is new is the sharp increase in public awareness of data centers, thanks to the rise of artificial intelligence and its accompanying need for more computing power and storage. With awareness has come disapproval. A Gallup poll from earlier this year found that 7 in 10 Americans oppose the construction of data centers for AI use where they live. Forty-six percent of respondents reported worrying “a great deal” about data centers’ environmental impact, and 24 percent “a fair amount.”
But what exactly are those environmental effects? According to Wei Peng, an assistant professor of public and international affairs and the Andlinger Center for Energy and the Environment, the centers’ biggest impact is the amount of electricity it takes to power them and not just, as is commonly thought, the water they consume.
With data centers emerging as a campaign issue, Peng, whose research uses computational models to quantify difficult tradeoffs of energy policy across social, political, and environmental aspects, offered some timely context.
In general terms, what are the environmental effects of an average data center?
While a data center has direct water impacts from their cooling needs, a substantial amount of its environmental impacts come from the electricity used to power it. Rather than simply asking, “What are the environmental impacts of a data center?” it is also important to ask, “Where does its electricity come from, and what air pollution, water, and carbon impacts are associated with that electricity supply?”
In fact, we already know a lot about this thanks to decades of power-sector research. If a data center follows a BYONCE approach (i.e., Bring Your Own New Clean Electricity, such as wind and solar), its environmental footprint can be quite small. But if it adds onsite natural gas generation or draws from a grid where fossil generation is meeting the additional demand, the impacts can be much larger.
In addition, some impacts are easier to tackle than others. Air pollution can be reduced with end-of-pipe controls, and dry cooling can substantially reduce water use. Carbon emissions are much harder. For fossil generation, the main option is carbon capture and storage, but there is little economic incentive to deploy it without strong climate policy.
What is it about data centers that affect the environment so strongly?
What is really making people worry is the potential scale of growth. Take the U.S. as an example. Electricity demand was largely flat from the mid-2000s to around 2020, growing by only about 0.1% per year. Over the same period, the power sector also became much cleaner, first through the shift from coal to natural gas after the shale gas boom, and later through the rapid growth of wind and solar as their costs came down. Before the AI boom, a lot of researchers and policymakers had started to view electricity as one of the “easier” sectors to clean up, and were shifting attention toward harder sectors like transportation and industry.
The AI boom is changing that picture. By 2030, data center electricity use in the U.S. could double, or even quadruple in some high-growth scenarios. That increase alone would add demand equivalent to roughly 5-15% of today’s total U.S. electricity use. This pace of growth, while familiar in some countries such as China and India, would be a big change for the U.S. And as the grid starts to rapidly grow again, where the new electricity comes from matters a lot for the overall environmental impacts.
Is there any way to mitigate these effects?
Yes, and there are two levers at the high level. The first is to reduce the data center’s own electricity and cooling-water demand. This can include more efficient computing and cooling systems and operational strategies such as shifting flexible workloads across time or locations to reduce peak electricity use and cooling needs.
The second is to get electricity from cleaner sources. This is where the BYONCE approach comes in. The basic idea is that a data center should bring enough new clean generation to supply its additional electricity demand, so that its growth does not require new fossil generation or increase fossil generation elsewhere on the grid.
Has AI shown any promise in counteracting the effects of climate change?
This is one of the central questions of our time, but at this point, everyone is still speculating. In fact, the direct environmental impacts of data centers, as we discussed above, are likely to be a small part of the overall picture. The much bigger question is how AI will affect the economy, and therefore influencing overall energy demand, emissions, and climate change.
In general, a larger economy tends to require more energy, which can make emissions mitigation more difficult. At the same time, a thriving economy can make people less vulnerable and better able to manage and adapt to climate impacts. How these interacting forces may play out in the future is still an open question. In fact, this was one of the homework exercises I asked students to work through in the course I am teaching this fall, Integrated Assessment Modeling for Climate Policy Making. Nobody can predict the future. But we can use model scenarios to explore what futures are plausible, what tradeoffs may emerge, and what factors matter most.
Left: Wei Peng; Right: Aerial of data center.
Headshot courtesy of Wei Peng, data center photo by Shutterstock.
