On the outskirts of a mid-sized American city, a windowless warehouse the size of several football fields hums around the clock. Inside, tens of thousands of servers process financial transactions, stream video, train machine learning models, and store the digital flotsam of modern life. Outside, neighbors sometimes notice something harder to quantify: a change in the air.
The connection between data centers and air quality is not straightforward, but it is real, and it is growing more consequential as the infrastructure of the internet expands at a pace that would have seemed implausible a decade ago. The rise of generative AI tools after 2022 accelerated demand for compute power so sharply that utility companies, environmental regulators, and community organizers found themselves scrambling to understand what was happening to their grids, and what was coming out of the smokestacks being built, or extended, to power these facilities.
The Scale of the Problem Starts With Energy
Any honest accounting of data centers and air quality has to begin with electricity. Data centers do not, for the most part, burn fuel on-site. They consume electricity, and it is the generation of that electricity, much of it still derived from fossil fuels in many parts of the world, that produces the bulk of their air quality impact.
The International Energy Agency estimated that data centers globally consumed roughly 240 to 340 terawatt-hours of electricity per year as of 2022, representing somewhere around one to 1.3 percent of global electricity demand. That figure, already large, was widely expected to grow substantially through the mid-2020s as AI workloads intensified. By 2024, analysts at several research organizations were projecting that AI-related computing alone could drive data center electricity consumption to multiples of its earlier baseline within a few years, though precise figures remained contested given the rapid evolution of hardware efficiency.
When a data center draws power from a grid that relies on natural gas or coal plants, those plants burn more fuel to meet demand. The combustion releases nitrogen oxides, sulfur dioxide, fine particulate matter, and carbon dioxide. These are not abstract chemical formulas. Nitrogen oxides contribute to the formation of ground-level ozone, which is associated with respiratory illness. Fine particulate matter, particles smaller than 2.5 micrometers in diameter, can penetrate deep into lung tissue and has been linked in research literature to cardiovascular disease, premature death, and developmental problems in children.
The critical variable is what is called the “emissions intensity” of the regional grid, meaning how much pollution is produced per unit of electricity generated. A data center in a region powered largely by hydroelectric or nuclear energy produces far less pollution than one powered by coal, even if they are identical in size and efficiency. This geography of emissions is uneven and often overlooked in corporate sustainability reports.
Local Impacts and the Communities Closest to the Fence Line
The air quality story is not only about regional grid emissions. It also involves what happens on-site and in the immediate vicinity of these facilities.
Most large data centers maintain diesel backup generators, sometimes hundreds of them, to ensure uptime during grid outages. These generators are among the dirtier pieces of equipment in modern industrial settings. When they run during testing, which is required periodically to ensure reliability, or during actual outages, they emit diesel exhaust containing fine particulate matter, nitrogen oxides, and volatile organic compounds.
In Northern Virginia, which by the early 2020s had become home to one of the highest concentrations of data center capacity anywhere in the world, community and environmental groups raised concerns about generator testing and its contribution to local air quality. The region had longstanding challenges meeting federal air quality standards for ozone, and some local advocates argued that the cumulative emissions from generator testing across dozens of facilities deserved more rigorous regulatory scrutiny. This became a matter of public discussion involving local planning boards and the Virginia state legislature, though the outcomes of various proposed measures evolved over subsequent years.
The diesel generator issue illustrates a broader regulatory gap. In the United States, the Environmental Protection Agency regulates stationary diesel engines, but permitting and enforcement often occur at the state and local level, and the pace of data center construction has in some jurisdictions outrun the capacity of regulators to keep up. Facility operators have generally argued that generators run infrequently and that their aggregate emissions are modest compared to major industrial sources. Environmental advocates counter that in areas with many facilities clustered together, cumulative impacts matter and are not always captured in individual permit reviews.
Water, Cooling, and Secondary Air Quality Effects
Data centers also consume enormous quantities of water for cooling, and this creates a secondary connection to air quality that is often overlooked. Cooling towers evaporate water to dissipate heat, and in doing so they can release aerosols into the surrounding air. Under certain conditions, poorly maintained cooling systems have been linked to outbreaks of Legionnaires’ disease, a severe form of pneumonia caused by Legionella bacteria.
Several outbreaks in urban areas over the past decade were traced to cooling towers in commercial or industrial buildings, though data centers were not uniquely implicated. The point is that the sheer concentration of cooling infrastructure associated with large data center campuses creates maintenance and monitoring obligations that have public health dimensions beyond simple energy accounting.
There is also the question of what happens when data centers displace other land uses or reshape local energy markets. In some regions, the rapid growth of data center demand has strained grid infrastructure, leading utilities to delay the retirement of older, less efficient power plants, or to bring mothballed fossil fuel generation back online. Research by grid analysts and environmental organizations in the mid-2020s examined whether data center growth in certain markets was effectively extending the operating life of plants that would otherwise have closed, with corresponding implications for air quality in communities near those plants.
The Renewable Energy Offset: Real Progress With Real Limits
Technology companies have been among the most aggressive purchasers of renewable energy in the corporate sector. Microsoft, Google, Amazon, and Meta have all made large commitments to match their electricity consumption with renewable energy procurement, and they have signed power purchase agreements for wind and solar projects representing tens of gigawatts of capacity.
These commitments have materially accelerated renewable energy development in some markets, and the companies involved are not wrong to point to this. But several important caveats apply.
First, renewable energy procurement is not the same as renewable energy consumption in real time. A company can purchase a renewable energy certificate corresponding to a unit of electricity generated by a wind farm in one part of the country while the data center itself draws power from a coal-heavy grid in another region. The net accounting may balance on paper, but the local air quality near the data center is shaped by what that local grid actually burns, not by what a wind farm produces somewhere else.
Second, the timing of renewable generation and data center demand often do not match. Solar panels generate electricity during the day; data centers run around the clock. Bridging that gap requires either battery storage (still expensive and constrained in scale) or continued reliance on dispatchable fossil generation during overnight hours and cloudy periods.
Some companies, notably Google, have publicly discussed ambitions to match energy consumption with clean generation on an hourly basis rather than an annual average, which is a meaningfully more rigorous standard. Whether and how quickly that approach becomes widespread in the industry remained an open question as of 2024 and 2025.
Third, the recent acceleration in AI-driven demand has created a race between efficiency gains and raw consumption growth. Individual chips and servers have become more efficient, but the scale of deployment has grown so fast that aggregate energy demand has continued rising. This dynamic, sometimes called the rebound effect, is a recurring pattern in the history of energy-intensive technologies.
What the Research Actually Shows
Academic and policy researchers have produced a growing body of work on data center emissions, though much of it acknowledges significant uncertainty given how rapidly the sector is evolving.
Studies examining the lifecycle emissions of cloud computing have generally found that cloud infrastructure can be more efficient per unit of computation than on-premises corporate data centers, because hyperscale facilities achieve higher utilization rates and can more easily integrate renewable energy. This suggests that the continued migration of computing workloads to large cloud providers could reduce emissions relative to a world of many smaller, less efficient facilities, all else being equal.
However, the same research consistently flags that overall demand growth can overwhelm efficiency gains. The question is not only whether data centers are getting cleaner per unit of output, but whether total output is growing faster than the grid is decarbonizing.
Research on the health impacts of data center-adjacent communities is less developed. Some peer-reviewed work has examined the contribution of data center generator testing to local air pollution, particularly in heavily impacted regions like Northern Virginia. The findings have been mixed in terms of magnitude, but the studies have generally supported the concern that cumulative emissions in high-density data center markets warrant monitoring, particularly for communities that already face elevated exposure to air pollutants.
Environmental justice researchers have noted that the communities most exposed to data center-adjacent pollution, including diesel exhaust from generators and emissions from nearby power plants dispatched to meet data center load, are often lower-income communities and communities of color, who also tend to have less political leverage in local land use and permitting decisions. This argument has become more prominent in public discourse as data center construction has accelerated.
Looking Forward: Pressure, Policy, and the Possibility of Change
The air quality consequences of the data center boom are not inevitable, but addressing them requires action on several fronts simultaneously.
Grid decarbonization is the single most powerful lever. If the electricity feeding data centers comes from sources that do not combust fuel, the primary pollution pathway is eliminated. The pace of grid decarbonization in major data center markets, including the United States mid-Atlantic region, the European Union, Southeast Asia, and parts of the American Southwest, varies considerably, and the addition of large new loads from data centers creates pressure that can work in either direction: it can accelerate renewable development if companies invest accordingly, or it can stress grids in ways that slow the retirement of fossil fuel plants.
Regulatory frameworks are also evolving. Some jurisdictions have moved to require more rigorous cumulative impact assessments for large data center projects, and there have been policy discussions in multiple countries about whether the sector should face emissions reporting requirements analogous to those applied to heavy industry. The state of Virginia, the European Union, and several other authorities were engaged in policy development around data center environmental oversight as of 2024 and 2025, though the specifics of what was adopted and implemented in each case continued to develop.
On the technology side, advances in liquid cooling, which can dramatically reduce both water consumption and the need for large air-handling systems, have gained traction. More efficient chip architectures reduce the energy required per computation. Some researchers have explored the possibility of locating data centers in colder climates where free-air cooling is possible for much of the year, reducing both energy use and the associated emissions from power generation.
What is clear is that the infrastructure question can no longer be treated as a purely technical or economic matter. When a new data center is proposed for a community, the conversation increasingly includes questions about what it will mean for local air quality, who will bear the health costs of the energy it requires, and whether the regulatory frameworks in place are adequate to the scale of what is being built. Those are not questions that will be resolved by the technology sector alone.
For readers who spend significant time near server infrastructure or in environments with diesel backup systems, monitoring local air quality can be informative. A portable air quality monitor can help track particulate levels in your home or neighborhood, providing concrete data on what you and your family are actually breathing.
As an Amazon Associate, The Rough Idea earns from qualifying purchases.