Tech companies have been pouring substantial investments into building data centers, which are critical to the rapidly accelerating demand for artificial intelligence (AI). Recently, Meta’s CEO Mark Zuckerberg announced plans for a $600 billion investment in US infrastructure, including data centers, by 2028, while OpenAI is set to invest $1.4 trillion.
A newly published study in Nature Communications delves into the environmental impact of these data centers, assessing where they should ideally be located to mitigate harmful effects. The research compiles various data sources, projecting environmental ramifications through the end of the decade. It highlights that tech giants’ commitments to achieving net zero emissions as they expand may not hold up against the substantial energy and water needs required by these facilities.
Fengqi You, a professor at Cornell University and co-author of the study, explains that the timing of this research is critical to understand AI’s impact on climate and water use. The study suggests that optimum locations for data centers in the coming years will be states like Texas, Montana, Nebraska, and South Dakota, which balance energy requirements and water resources efficiently. States that rely more on renewable energy sources will significantly lower the carbon emissions from these centers.
Historically, regions such as Virginia and Northern California have been favored for data center construction due to their geographical advantages and existing infrastructure. Virginia is home to over 650 data centers, benefiting from its proximity to Washington, D.C., and tax incentives. However, the state’s growing energy demands from data centers may jeopardize its goal of achieving 100 percent clean energy by 2045. California faces similar challenges, grappling with chronic water issues that could hinder continued data center growth.
The study also cautions that as data center operators often choose locations based on energy and water needs, they may be overlooking regions that could serve the industry better. While certain states continue to attract significant investments and development, there is concern that this concentration could exacerbate resource strains.
Data center construction in Texas has surged, making it the second-most popular state for data centers, while states like Montana, Nebraska, and South Dakota are just beginning to emerge in this sector. Despite the potential for expansion, they still have a long way to go compared to established hubs like Virginia.
The study underscores uncertainty in predicting future demands due to variables such as improvements in technology and possible fluctuations in the AI market. The political climate surrounding fossil fuels also plays a significant role in the energy landscape needed for sustaining these data centers.
The future may witness a scenario where the U.S. data center expansion could produce an additional 44 million tons of carbon dioxide annually if the industry’s growth far exceeds efficiency improvements and the shift to renewable energy falters. This projection raises concerns about the long-term environmental impacts of the industry’s quick expansion.
Conversely, breakthroughs in cooling technologies and proper siting of data centers could lead to significant reductions in their carbon footprint and water usage. Experts, including MIT’s Computing and Climate Impact Fellow Noman Bashir, emphasize that innovations in energy sources and efficiency could influence the data center sector’s overall emissions.
Bashir acknowledges that the likelihood of tech giants meeting their net-zero promises diminishes in light of this rapid expansion. Transparency regarding emissions linked to data center operations is essential for ensuring sustainability as infrastructure continues to grow.
Ultimately, the study suggests prioritizing sustainable practices from the outset will be critical for the future of AI computing infrastructure in the U.S.
