U.S. AI Data Centers Produce More Pollution Than Expected

U.S. AI Data Centers Produce More Pollution Than Expected, Study Finds

November 17, 2026 | U.S. Technology and Environment News

The rapid expansion of artificial intelligence is creating a new environmental challenge for the United States. While AI companies continue investing billions of dollars in computing infrastructure, a new Harvard-led study has raised concerns about the electricity consumption and carbon emissions associated with some of the country’s largest data centers.

Researchers examined 403 hyperscale data centers operating across the United States between May 2024 and April 2025. Their analysis estimated that these facilities consumed approximately 82 terawatt-hours of electricity during the study period under the central operating scenario.

That electricity use was associated with approximately 25.7 million metric tons of carbon dioxide emissions.

The findings, published in the journal Nature Sustainability, offer a more detailed estimate of the environmental footprint of large U.S. data centers. They also arrive as demand for AI computing continues to grow and technology companies announce plans for additional facilities.

The study suggests that the environmental cost of the AI expansion deserves closer attention, particularly because a significant share of the electricity used by these facilities was linked to fossil-fuel generation.

The findings do not mean artificial intelligence itself is necessarily harmful. Instead, they highlight an important question for the technology industry: Can the United States expand its AI infrastructure while reducing the pollution and resource consumption associated with powering it?

What Did the Harvard Study Discover?

The research team collected facility-level information on 403 large data centers in the United States. These facilities are known as hyperscale data centers because they have substantial computing capacity and are designed to handle large volumes of digital workloads.

Researchers used available information, satellite imagery, electricity data and modeling to estimate how much power the facilities consumed and which energy sources supplied that electricity.

Under the study’s central scenario, the 403 data centers consumed approximately 82 terawatt-hours of electricity between May 2024 and April 2025.

The researchers estimated that this electricity use was associated with approximately 25.7 million metric tons of carbon dioxide emissions.

The study also presented a range of estimates depending on assumptions about how heavily the facilities operated. Electricity consumption was estimated at approximately 68 to 99 terawatt-hours, with associated emissions ranging from roughly 21 million to 31 million metric tons of carbon dioxide under the study’s attribution method.

The central estimate represented approximately 1.8% of total U.S. electricity consumption during the period.

These numbers are significant because they show how much electricity a relatively limited group of large computing facilities can require.

However, the research should be understood as an estimate based on facility information and electricity-generation data, rather than direct measurements from every data center.

Why Do AI Data Centers Need So Much Electricity?

Artificial intelligence systems require powerful computers to process information, train models and respond to user requests.

Behind many online AI services are large data centers containing thousands of servers, networking equipment, storage systems and cooling infrastructure.

These facilities operate continuously. Their computing equipment produces heat while processing data, making cooling systems necessary to maintain safe operating temperatures.

As AI applications become more advanced, demand for computing capacity can increase.

Several activities contribute to the electricity requirements of modern data centers:

1. Training Large AI Models

Training an advanced AI model can require enormous computing resources.

Specialized processors perform repeated calculations using large datasets. Depending on the model and training process, this work can involve substantial electricity consumption.

As companies develop more capable AI systems, they may need additional computing infrastructure to support research and development.

2. Running AI Services for Millions of Users

Training is only one part of the process.

Once an AI model becomes available, it must respond to requests from users, businesses and other applications.

This process is often called inference. When millions of people use AI services, the computing infrastructure must handle a large and sometimes unpredictable number of requests.

That demand can require additional servers and electricity.

3. Cooling and Maintaining Equipment

Data centers generate heat because their computing equipment operates continuously.

Cooling systems help prevent overheating and maintain reliable performance.

Depending on the facility’s design, cooling can involve air conditioning, liquid cooling, pumps and other equipment that consume additional electricity.

4. Continuous Operation

Many digital services are expected to remain available around the clock.

Data centers therefore need reliable power systems, backup arrangements and infrastructure that can support operations even when demand changes.

The combination of computing, cooling and continuous operation makes electricity planning an important part of data center development.

More Than Half of the Electricity Was Linked to Fossil Fuels

One of the most important findings concerns the sources of electricity used by the facilities.

The Harvard study estimated that approximately 54% of the electricity generation attributed to the data centers came from fossil-fuel sources.

These include energy sources such as natural gas and coal, which release carbon dioxide when used to generate electricity.

The remaining share was associated with other electricity sources, including lower-carbon generation.

This matters because the environmental impact of a data center depends not only on how much electricity it consumes but also on how that electricity is produced.

A facility supplied primarily by low-carbon electricity can have a different emissions profile from a similar facility relying heavily on fossil-fuel generation.

The researchers used regional electricity-generation information to estimate emissions associated with data center demand.

That approach helps provide a clearer picture of the relationship between computing infrastructure and the wider power system.

However, it is important not to interpret the study as proof that every individual data center uses the same energy mix. Electricity sources vary by location and regional grid conditions.

Why the Findings Matter for the United States

The United States has become a major center for cloud computing, artificial intelligence development and digital infrastructure.

Technology companies are expanding data center capacity to support AI products, cloud services, business applications and other online platforms.

This expansion brings potential economic benefits, including construction activity, investment, technical jobs and additional digital services.

But it also creates pressure on electricity systems.

When large data centers are built in a particular region, local utilities may need to plan for additional electricity demand. In some areas, that can create questions about power generation, transmission infrastructure, energy costs and long-term reliability.

The environmental impact is another part of the discussion.

If additional electricity demand is met through fossil-fuel power generation, emissions could increase. If new facilities are supported by cleaner energy and more efficient systems, their carbon footprint may be reduced.

The Harvard research gives policymakers and energy planners another source of information as they evaluate the consequences of the country’s growing data center industry.

Plans for More Data Centers Raise Further Questions

The study’s findings come at a time when the United States is seeing major investment in new digital infrastructure.

Technology companies and developers have announced plans for additional data centers to support the expected growth of AI and cloud computing.

The Associated Press reported that plans for approximately 2,000 additional data centers have added to concerns about future electricity demand.

The number of planned facilities does not mean all of them will be constructed, operate at full capacity or produce the same emissions.

Their eventual environmental impact will depend on several factors, including location, computing capacity, operating efficiency, cooling technology and electricity sources.

Nevertheless, the potential scale of future development makes energy planning increasingly important.

Local communities are also paying attention to proposed projects, particularly where residents have questions about electricity costs, water use, land development and environmental effects.

For developers, obtaining access to reliable electricity and suitable infrastructure is becoming an important part of data center planning.

AI’s Environmental Impact Is Not Limited to Carbon Emissions

Electricity consumption and carbon emissions are major concerns, but they are not the only environmental issues associated with data centers.

Water use is another important consideration.

Some facilities use water-based cooling systems to manage heat generated by servers. The amount of water required can vary considerably depending on the cooling design, climate and operating conditions.

In areas already facing water stress, large industrial facilities can create additional pressure on local resources.

Data centers also require physical infrastructure, electronic equipment and materials.

Servers and other components eventually need replacement, creating questions about electronic waste, recycling and responsible hardware management.

The International Telecommunication Union’s new Antalya Artificial Intelligence Pledge, launched during COP31 on November 16, addresses several of these issues.

The voluntary initiative encourages participating organizations to improve energy and water efficiency, use cleaner electricity, manage hardware responsibly and increase transparency about the environmental impact of AI systems.

The pledge also encourages the use of AI to support climate-related solutions.

It is not a binding U.S. law or a guarantee that participating companies will immediately eliminate their emissions. Instead, it establishes areas for cooperation and encourages measurable action.

COP31 Launches a New Initiative for Sustainable AI

The environmental debate surrounding AI is no longer limited to researchers and technology companies.

At the COP31 climate conference in Antalya, Türkiye, the International Telecommunication Union and the COP31 Presidency launched the Antalya Artificial Intelligence Pledge.

The initiative focuses on four broad areas.

First, it encourages more sustainable AI infrastructure, including greater use of low- and zero-emission electricity and improved energy and water efficiency.

Second, it calls for responsible design and management of AI systems throughout their life cycles.

Third, it supports better assessment and reporting of the environmental impact of AI.

Fourth, it encourages investment and technical support for using AI to address climate challenges, particularly in developing countries and vulnerable communities.

The initiative reflects a growing international discussion about how to balance technological development with environmental responsibility.

For American technology companies, the pledge provides another framework for considering how their infrastructure decisions affect energy use and emissions.

Can AI Help Reduce the Pollution It Creates?

Artificial intelligence has an environmental challenge, but researchers and climate advocates also see opportunities for the technology to help reduce emissions in other sectors.

AI tools can potentially support more efficient electricity grids, improve building energy management, assist weather forecasting and help industries optimize their use of resources.

For example, AI systems may help utilities forecast electricity demand more accurately or identify opportunities to reduce energy waste.

In manufacturing, software tools can help improve production planning and reduce unnecessary material consumption.

However, the possible environmental benefits of AI should not be treated as proof that its own energy consumption is automatically offset.

The actual results depend on how AI systems are developed, powered and used.

A technology that improves efficiency in one area may still increase electricity demand elsewhere if its computing infrastructure expands rapidly.

The central challenge is to measure both sides of the equation rather than relying only on promises of future environmental benefits.

Could Data Centers Reuse Their Waste Heat?

Another possible way to reduce the environmental impact of data centers is to capture and reuse the heat generated by computing equipment.

Servers produce heat during operation, and many facilities release that heat into the surrounding atmosphere through cooling systems.

Instead of allowing all of it to go unused, some projects are exploring ways to redirect the heat toward nearby buildings, businesses or industrial processes.

The International Energy Agency has estimated that data center waste heat could provide enough energy to heat around 10% of European homes by 2030 if it were fully integrated into district heating systems.

That estimate relates to Europe and should not be interpreted as a confirmed projection for U.S. homes.

Nevertheless, it demonstrates the potential value of treating data center heat as a resource rather than simply a byproduct.

In the United States, projects involving heat recovery could become more attractive in locations where nearby communities or industrial facilities can use a consistent supply of thermal energy.

The approach is not suitable for every location, and infrastructure is needed to transport and use the heat efficiently.

What Happens Next for America’s AI Industry?

The latest Harvard research is likely to add to discussions about how quickly the United States should expand its data center infrastructure and what environmental standards should apply.

Technology companies will face increasing pressure to explain how their facilities use electricity, where their power comes from and what steps they are taking to improve efficiency.

Utilities and regional authorities will also need to consider how new computing facilities affect electricity demand and infrastructure planning.

For policymakers, the challenge will be finding a balance between supporting innovation, maintaining reliable electricity supplies and limiting environmental damage.

The study does not suggest that the United States should stop building data centers. Instead, it provides evidence that the environmental cost of digital infrastructure needs to be included in decisions about future expansion.

The next stage of AI development may depend not only on faster processors and more powerful models but also on the ability to operate computing infrastructure more sustainably.

Frequently Asked Questions

1. How much electricity did the 403 U.S. data centers consume?

The Harvard study estimated that 403 large U.S. hyperscale data centers consumed approximately 82 terawatt-hours of electricity between May 2024 and April 2025 under its central scenario.

2. How much carbon dioxide did these data centers produce?

The study estimated approximately 25.7 million metric tons of associated carbon dioxide emissions under its central scenario.

3. What percentage of U.S. electricity did the data centers use?

Their estimated electricity demand represented approximately 1.8% of total U.S. electricity consumption during the study period.

4. Why do AI data centers consume so much electricity?

They require power for servers, specialized AI processors, data storage, networking equipment and cooling systems. Many facilities operate continuously to support digital services.

5. Does the Harvard study prove that every data center uses fossil fuels?

No. The study estimated that approximately 54% of the electricity generation attributed to the facilities came from fossil-fuel sources. The energy mix can differ between locations and facilities.

6. What is the Antalya Artificial Intelligence Pledge?

It is a voluntary initiative launched at COP31 by the COP31 Presidency and the International Telecommunication Union. It encourages cleaner AI infrastructure, better energy and water efficiency, environmental reporting and responsible AI development.

7. Can AI help fight climate change?

AI may help improve energy efficiency, electricity-grid management, forecasting and industrial processes. However, its benefits depend on real-world implementation and do not automatically cancel out the emissions associated with data centers.

Final Thoughts

The expansion of artificial intelligence is changing the American economy, technology industry and everyday life. But the infrastructure behind this transformation requires enormous computing resources and electricity.

The new Harvard study offers an important estimate of the environmental footprint of 403 major U.S. data centers. Its findings show that AI infrastructure is becoming a significant part of the country’s electricity and emissions discussion.

The challenge ahead is not simply to build more powerful data centers. It is to make sure that the next generation of AI infrastructure is supported by cleaner electricity, efficient cooling, responsible water use and transparent environmental reporting.

If the technology industry can combine innovation with sustainable infrastructure, AI may become part of the solution to some environmental challenges rather than adding unnecessarily to them.

Reporting date: November 17, 2026
Research period: May 2024–April 2025
Primary research: Nature Sustainability, Harvard-led study on U.S. hyperscale data center energy consumption and carbon emissions.


Sources and Further Reading

  1. Associated Press — Study estimates magnitude of U.S. data center pollution as climate talks unveil clean AI pledge.
  2. Nature Sustainability — Assessing the carbon emissions and energy consumption of U.S. hyperscale data centres.
  3. Harvard T.H. Chan School of Public Health — Hyperscale data centers’ energy consumption and carbon emissions newly estimated by researchers.
  4. International Telecommunication Union — COP31’s Antalya AI Pledge to tackle artificial intelligence’s climate footprint.

Update Policy: Update the article if new U.S. data center energy figures or major clean-energy commitments are announced.

Trend Now

Leave a Comment

error: Content is protected !!
U.S. House Approves $1 Trillion Defense Bill