While AI's overall environmental footprint remains small compared to transportation and manufacturing, the technology's concentrated local impacts on water supplies, air pollution, and community health are emerging as the real concern. By 2030, data centers could consume as much energy as Japan and as much water as needed to supply everyone in sub-Saharan Africa, with pollution potentially causing 1,300 additional deaths annually in the United States alone.

Why Is AI's Environmental Impact Concentrated in Local Areas?

The paradox of AI's environmental challenge is that it looks manageable on a global scale but devastating at the local level. Currently, all data centers account for just 0.5% of global CO2 emissions, which seems minor compared to sectors like aviation or manufacturing. However, the speed and concentration of AI data center construction is creating what researchers call a "perfect storm" of localized environmental stress.

The core issue stems from infrastructure constraints. Existing electricity grids cannot keep up with demand from new data centers. In Virginia, the world's largest data center market, there is a seven-year wait for new grid connections in areas serviced by Dominion Energy. Rather than waiting, developers are bypassing the grid entirely and installing on-site power generation, typically using gas turbines and other carbon-intensive sources.

New data from Global Energy Monitor reveals the scale of this shift. The pipeline for gas projects dedicated to data centers in the United States jumped from 97 gigawatts at the end of 2025 to 189 gigawatts by the end of June 2026, nearly doubling in just six months. This means new data centers are disproportionately likely to use more carbon-intensive energy sources than the average power grid mix.

What Are the Specific Health and Environmental Risks?

Beyond carbon emissions, data centers powered by on-site gas turbines create immediate air quality problems for surrounding communities. In May 2026, the NAACP filed for an injunction against xAI's power plant in Southaven, Mississippi, claiming it could emit thousands of tons of nitrogen oxides per year, along with formaldehyde and other harmful chemicals.

A preprint study by researchers at UC Riverside, led by Shaolei Ren, estimated the cumulative health toll. The analysis projects that by 2030, pollution from data centers could cause an additional 1,300 deaths in the United States annually and impose approximately $20.9 billion in public health costs each year. These are not theoretical numbers; they represent real mortality and healthcare burden concentrated in communities near data center sites.

The environmental footprint extends beyond air quality. Data centers consume enormous amounts of water for cooling, and as GPU chips burn out and are replaced, they generate electronic waste comparable to the annual waste output of entire nations. By 2030, data center electronic waste could match the total waste generated by Denmark, Norway, or Austria, while also straining global supplies of critical minerals needed for semiconductor manufacturing.

How Does the Jevons Paradox Undermine Efficiency Gains?

One of the AI industry's primary arguments is that environmental impact will decrease over time as each new generation of chips becomes more energy efficient. However, this assumption overlooks a well-documented economic principle known as the Jevons paradox, which observes that making a technology more efficient actually increases demand to such an extent that it can outweigh any efficiency savings.

In the context of AI, this plays out as follows. Better GPUs allow companies to train larger models more efficiently, which makes those models cheaper to operate. Lower costs drive broader adoption and more frequent use, ultimately increasing total energy consumption despite per-unit efficiency improvements. Sasha Luccioni, a former AI researcher at Hugging Face and co-founder of Sustainable AI Group, explained the dynamic: the micro trend of improving GPU efficiency is being completely overwhelmed by the macro trend of AI becoming embedded in everything, all the time.

Recent research underscores this concern. A consortium of researchers, including representatives from Cohere and Meta, developed the AI Energy Score, which rates models based on their energy efficiency during inference, the process of running a trained model to generate outputs. They found that reasoning models, which work through prompts step by step in a process designed to mimic human thought, use 30 times more energy than models without reasoning capabilities.

Where Does Most AI's Energy Consumption Actually Come From?

A critical insight often overlooked in discussions of AI efficiency is that the bulk of environmental impact comes not from training new models, but from inference costs incurred when people query them. A 2022 Google paper revealed that approximately 60% of its machine learning footprint was inference, not training. As AI adoption accelerates, this balance is expected to shift even further toward inference, meaning that efficiency improvements in training will have minimal impact on total energy consumption.

Steps Communities and Policymakers Are Taking to Address AI Data Center Impacts

• State-Level Moratoriums: American lawmakers in several states have proposed moratoriums on data center construction, with New York signing one into law, citing concerns over water and energy consumption and local environmental impacts.
• Local Advocacy and Legal Action: Communities are filing injunctions and organizing vocal protests against planned data center construction, forcing developers to address environmental and health concerns before projects proceed.
• Emissions Monitoring and Health Impact Studies: Researchers are conducting detailed analyses of pollution from data center power plants and quantifying public health costs, providing evidence for regulatory action and community negotiations.

The environmental challenge posed by AI expansion is not a single, global problem but rather thousands of smaller, localized ones. Alex de Vries-Gao, a researcher at VU Amsterdam's Institute for Environmental Studies who has spent the last decade studying the sustainability of emerging technologies, noted that the pattern mirrors cryptocurrency mining, where economic incentives directly opposed environmental goals. In AI, the "bigger is better" dynamic creates similar misalignment, as companies scale models to improve performance, regardless of energy cost.

"It incentivizes the use of more resources through a 'bigger is better' dynamic," explained Alex de Vries-Gao, researcher at VU Amsterdam's Institute for Environmental Studies.

Alex de Vries-Gao, Researcher at VU Amsterdam's Institute for Environmental Studies

The International Energy Agency projects that data centers will account for around 3% of global electricity demand by 2030, with AI responsible for approximately half of that share. At that point, data center carbon emissions would be comparable to roughly half the global aviation industry. While this remains a small fraction of total global emissions, the concentration of environmental and health impacts in specific communities makes the problem urgent and politically contentious.

As AI companies race to scale their models and deploy new data centers, the real environmental reckoning will not be felt globally but in the neighborhoods surrounding these facilities, where air quality degrades, water becomes scarcer, and health costs mount. Policymakers and communities are beginning to push back, recognizing that the true cost of AI expansion is being paid by those living closest to the infrastructure that powers it.