Is Air Quality Improvement Being Undone by Data Centers? U.S. Study Warns of the Future in 2031

When you input a question into your smartphone, an answer comes back in seconds. AI that summarizes text, generates images, and even thinks through work procedures appears to be a nimble service operating solely on the screen.

However, the infrastructure that produces these responses occupies real land and continuously consumes electricity.

Depending on how that electricity is generated, long-standing efforts to combat air pollution could significantly regress. An article by The Verge, which highlighted an analysis by the Houston Advanced Research Center (HARC) in Texas, posed such questions to the infrastructure supporting growth in the AI era.

This is not merely a story of "AI consuming a lot of power." The issue is which power generation facilities, and in which locations, will supply the additional electricity needed. This choice affects the air quality in the region.

A scenario of increased emissions by 2031—though not a "predetermined future"

The Verge introduced a scenario where planned data centers are constructed and receive power from the assumed energy mix.

The article indicated that emissions of particulate matter from the power generation sector could more than triple, reaching levels not seen since 2006. Nitrogen oxides (NOx), which cause smog, could increase by 136%, and carbon dioxide (CO2) could rise by 90% compared to 2025.

However, caution is needed in handling these figures.

In HARC's official statement, one scenario of supply by grid power described NOx increasing by 136%, and PM2.5 and CO2 by about 90%.The increase rate for PM2.5 does not match the "more than triple" stated in The Verge's text. Whether this discrepancy is due to differences in comparison conditions or aggregation scope, or a matter of description, could not be confirmed, and this article does not treat the PM2.5 multiplier as a definitive value.

Moreover, emissions and the concentration of air that residents actually breathe are not the same. The estimate of a 136% increase in NOx emissions from the power generation sector cannot be directly read as "the air quality in all regions will worsen by 136%."

The figures indicate a warning that emissions of pollutants could significantly increase under certain conditions.

Looking only at the data center buildings doesn't reveal the environmental impact

When considering this issue, it's necessary to separate the buildings housing servers from the facilities that generate their electricity.

If a data center buys electricity from the grid, related emissions may occur at distant power plants. Conversely, if fuel-based power generation facilities are placed on-site or nearby, the relationship between power supply and emission sources changes.

Thus, a facility's clean appearance and lack of visible chimneys alone cannot determine the environmental impact of its operation.

For users, the "cloud" is connected to physical infrastructure. While AI service descriptions list model performance and response speed, information about the power supply is also a crucial element for understanding the service's reality.

Regarding the environmental goals set by businesses, it's necessary to confirm both CO2 reduction and the reduction of air pollutants related to the region. A single indicator cannot explain all environmental impacts.

Future predictions are influenced by construction plans and power source choices

HARC analyzed different combinations of power supply methods with three different data center demand forecasts.

The model assumes that facilities operate at 80% load throughout the year. Each scenario assumes that new data center demand is met by the same power supply method.

In reality, there may be delays or cancellations of plans, and different power sources or operational methods may be adopted for each facility. Therefore, a single line of estimation does not directly translate into future achievements.

This uncertainty does not render the warning meaningless. Rather, it indicates that future outcomes can be altered by decisions on equipment investment and operation.

It's necessary to question not only the scale of data center construction but also what will be used to supply electricity. This is the central point that can be drawn from HARC's analysis.

On social media, debates continue about "lack of transparency" and the need for comparisons with other industries

On social media, discussions about the environmental impact of data centers continue from different perspectives.

The topics introduced here are posts on the overseas bulletin board-style SNS "Reddit" related to the theme, not direct responses to the current The Verge article or HARC's new analysis. Additionally, trends in public opinion cannot be judged from the posts of a single community.

One discussion is dissatisfaction with the invisibility of environmental impact.

There were opinions that it's difficult to grasp the whole picture with only corporate disclosures, and that it's necessary to consider not only AI's learning and calculation during use but also semiconductor manufacturing and raw material extraction.

On the other hand, some posts argue that attention should not be focused solely on data centers and that resource consumption should be compared with agriculture and other uses. They suggest that design and efficiency improvements of facilities can reduce the burden.

In response, there were counterarguments that the large burden of other industries is not a reason to increase new burdens.

These posts are not scientific verifications in themselves. However, it is clear that the focus of the discussion is not limited to "Is AI good or bad?" but also involves disputes over what extent of burden to count, what to compare it with, and what information companies should disclose.

In Japan, it's also necessary to consider attraction and power supply together

The numbers from Texas cannot be directly applied to Japan due to differences in power source composition, facilities, location, and systems.

Nevertheless, the issues are familiar to Japan. In Japan's 7th Basic Energy Plan, data centers are listed as a factor when considering future power demand.

When considering the construction of new data centers in Japan, there are things to confirm alongside investment amounts and economic effects.

First, where to procure the additional electricity from. It's a question not only of the annual purchase volume but also of what kind of power sources will be responsible for supply during the hours the facility operates.

Second, if power generation facilities are to be installed, what will be the operating conditions and emission countermeasures? It's necessary to explain not only the installation of equipment but also the extent of its use under normal conditions.

Third, can explanations to the community be made verifiable? If information is available that allows comparison between explanations at the planning stage and the reality after operation, such as power usage and equipment operation records, residents can make more concrete judgments.

These are not predictions that the same pollution as in Texas will occur in Japan. They are perspectives for evaluating business plans that can be derived from overseas estimates.

The question is how to design the "system that operates AI"

AI has value in assisting work, making information easier to find, and enabling new expressions. Recognizing that value and examining the infrastructure burden can coexist.

This analysis should not be reduced to the conclusion that "using AI will inevitably worsen air quality." At the same time, one cannot postpone decisions on power supply by assuming that "technology will naturally solve the problem as it advances."

When deciding to expand data centers, consider the necessary electricity and emission countermeasures together. Businesses should present explainable data, and administrations and communities should verify the content. Such concrete judgments will influence future environmental impacts.

Even if the AI responses displayed on the screen are brief, the infrastructure behind them continues to operate over many years.

Where and from what the electricity supporting convenience is generated is a question that Texas's warning also poses to Japan.

Sources and References

1. The Verge:
   Reported on the potential for data center power demand to reverse air quality improvements, highlighting HARC's analysis.
   https://www.theverge.com/ai-artificial-intelligence/1015958/data-center-pollution-rising-report

2. HARC Official Announcement: Emission Estimates and the Importance of Power Source Selection
   Confirmed the outlook for increases in NOx, PM2.5, and CO2. Noted discrepancies in PM2.5 increase rates with the original article.
   https://harcresearch.org/news/data-center-power-demand-could-undo-decades-of-texas-air-quality-progress-harc-analysis-finds/harcresearch.org

3. HARC Research Page: Analysis Methods, Assumptions, and Limitations
   Confirmed uncertainties due to three demand forecasts, operating conditions, power supply scenarios, and construction delays.
   https://harcresearch.org/research/texas-data-center-air-quality/harcresearch.org

4. Reddit: Related Discussions on the Environmental Impact of Data Centers and AI
   Summarized dissatisfaction with information disclosure, calls for comparisons with other industries, and counterarguments. Not direct responses to the current report, and numbers in posts are not adopted as factual evidence.
   https://www.reddit.com/r/BetterOffline/comments/1tlnaxb/data_centerai_environmental_impacts/reddit.com

5. Ministry of Economy, Trade, and Industry: 7th Basic Energy Plan
   Referenced that data center demand is listed as a factor when considering Japan's future power demand.
   https://www.meti.go.jp/files/900015352.pdf