Air quality monitoring
But recent developments in Memphis, Tennessee, suggest a more complex reality.
The rise of artificial intelligence (AI) has dramatically increased demand for high-powered data centres.
With that, the emissions footprint of the digital economy is shifting from abstract electricity consumption to a very real, very local air quality concern.
In mid-2024, a data centre deployed dozens of gas-fired turbines at a Memphis facility to power its artificial intelligence offering.
Satellite and infrared imagery later revealed that most of the turbines were in fact active, contradicting company claims.
Environmental groups estimate the facility could emit up to 130 tons of nitrogen oxides (NOₓ) annually, placing it among the top emitters in Shelby County.
Data centres do not typically combust fuel during normal operation.
But their dependence on backup diesel or natural gas generators creates a direct emissions risk, especially in areas where grid capacity is constrained or unreliable.
These generators emit NOₓ, fine particulate matter (PM₂.₅), carbon monoxide (CO), and volatile organic compounds (VOCs).
Even routine, weekly or monthly, releases pollution into nearby communities.
In peak summer conditions, some centres participate in demand response programs, switching to backup power to ease grid strain.
While this reduces outages, it often coincides with already poor air quality, exacerbating public health risks.
In Loudoun County, Virginia, home to the world's largest concentration of data centres, over 4,000 diesel generators are permitted, with a collective backup capacity of 11–12 GW.
If even a fraction operate during a grid event, the emissions could rival those of a large fossil-fuelled power station.
The facility sits in South Memphis, a historically Black neighbourhood with elevated asthma rates and a long legacy of industrial pollution.
The site's lack of transparency and failure to secure permits have provoked community outrage and raised awareness of a wider trend: digital infrastructure is often sited in vulnerable communities, where land is cheap and permitting scrutiny is low.
This isn’t isolated.
In Silicon Valley and Northern Virginia, residents are raising alarms over generator emissions, refrigerant leaks, and waste heat from sprawling data centre campuses.
With minimal state or federal disclosure requirements, many communities are turning to low-cost sensors, like PurpleAir, to monitor what they’re breathing.
Several types of emissions and by-products around data centres warrant closer attention from air quality professionals.
Backup generators, for example, are rarely equipped with continuous emissions monitors, yet they can release substantial quantities of NOₓ and particulate matter during outages or even during scheduled testing.
Keeping logs of generator runtime and fuel use is a practical starting point, and more advanced facilities should consider deploying fixed-point or portable exhaust monitoring systems.
Ambient air quality near data centre clusters is another key concern.
Low-cost sensors capable of detecting NO₂, ozone, and PM₂.₅ can be deployed around campuses to identify pollution events and track spatial and temporal trends.
These networks are especially useful in establishing exposure gradients and identifying localised pollution hotspots.
Cooling infrastructure also presents a monitoring challenge.
Data centres rely heavily on hydrofluorocarbon-based refrigeration systems, which may leak under high thermal loads.
Tools such as infrared detectors, handheld gas sensors, and digital refrigerant tracking platforms are well suited to identifying and quantifying leak events.
Although these emissions may not show up in traditional criteria pollutant monitoring, they contribute significantly to greenhouse gas inventories and can pose indoor air quality risks.
In addition, waste heat discharge from cooling systems can raise local ambient temperatures and exacerbate urban heat island effects.
Monitoring this thermal load requires infrared thermography or spatially distributed temperature sensors that can map microclimate changes around large facilities.
Finally, indirect emissions must be considered.
A data centre’s electricity consumption may result in substantial NOₓ, SO₂, and CO₂ emissions elsewhere on the grid, especially if the local mix is fossil fuel-heavy.
To gain a full picture of impact, professionals should link real-time electricity use with grid emissions data, enabling power source attribution and offsite footprint analysis.
Despite their substantial energy use and emissions potential, data centres are typically classified as minor pollution sources under the Clean Air Act.
This status often applies as long as their backup generators operate fewer than 200 hours per year, allowing them to bypass extensive monitoring and reporting obligations.
In Memphis, their data centre took this a step further by categorising its gas turbines as temporary, avoiding even these minimal requirements.
Only after public advocacy and satellite thermal evidence revealed the scale of operations did the company begin submitting permit applications.
Similar strategies have been observed in other regions, such as California and Virginia, where banks of diesel generators have been installed under the guise of emergency use, often with minimal environmental review.
While some jurisdictions are beginning to push for greater transparency - for instance, Virginia now publishes a public registry of air permits for data centres - oversight remains patchy.
In many cases, generator operation logs are stored onsite and inaccessible.
Emissions data is typically modelled rather than measured, and the public is asked to accept claims of negligible pollution without access to supporting evidence.
Fortunately, the air quality monitoring technologies required to address these issues are already in widespread use across industrial sectors.
Community and perimeter monitoring can be achieved through low-cost sensor networks.
Large combustion sources, such as backup generators, can be equipped with continuous emissions monitors to track NOₓ, CO, and particulate levels in real time.
Cooling systems can be monitored using refrigerant tracking platforms and leak detection equipment.
Satellite and drone-based thermal imagery can help pinpoint unreported generator use and verify operational status.
And integrated data platforms can combine grid mix data, generator runtime logs, and known emissions factors to create reliable estimates of direct and indirect emissions.
To ensure data centres do not become stealth contributors to local air pollution, the air quality monitoring community can champion:
IET 36.3 May
Product directory
.jpg)
.jpg)
