1 SEP 2026 — A study of one Virginia data-centre site found that eight full-time gas turbines could affect more than 2.5 million people, causing an estimated 3.4 to 6.5 additional premature deaths a year and US$53m to US$99m in annual health damages. The turbine shortage is being solved as a manufacturing problem. The cost of solving it lands on whoever lives near where the turbines run.
The figures
The Virginia estimate covers one facility with eight turbines in continuous operation, in the corridor known as Data Center Alley, and reaches across multiple counties. The pollutants at issue are smog-forming compounds and hazardous air pollutants including formaldehyde, with asthma, respiratory disease and certain cancers among the associated risks.
Supply is the reason turbines are being installed rather than grid connections awaited. GE Vernova reports being mostly sold out through 2030. SpaceX has bought roughly 830 acres at Bastrop, Texas, between March and June for a foundry casting turbine blades, which Elon Musk says could accelerate turbines coming online by up to 18 months.
In Memphis, where the Colossus facilities have operated since 2024, the NAACP has repeatedly said the site lacks required permits and pollution controls. The International Energy Agency projects global data-centre electricity demand to roughly double by 2030.
Read the estimate for what it is
Numbers of this kind come from health-impact assessment, which models emissions, disperses them across a population and applies concentration-response functions drawn from epidemiology. The output is a statistical expectation across a large group rather than a count of identifiable people.
The method is standard, underpinning the cost-benefit analysis behind most air-quality regulation. The uncertainty is also real. A range of 3.4 to 6.5 is nearly a factor of two, and the assumptions about stack height, meteorology and baseline health status each move it.
This does not make the estimate dismissible. It means the result is a range, best used as an order of magnitude — the health cost of running one such site is tens of millions of dollars a year, not tens of thousands and not billions.
Why turbines rather than the grid
A data centre needs power on the schedule of its construction, and an interconnection queue does not run on that schedule. In several US markets the wait for a large new grid connection is measured in years, and a facility built to serve a training run cannot wait for it.
On-site gas generation solves that by bypassing the queue entirely. It is faster, it is controllable, and in most jurisdictions it is permitted under air rules rather than through the interconnection process, which is a different and often quicker approval.
So the pollution is not incidental to the build-out. It is the mechanism by which the build-out happens on the timeline the industry wants, and any account of data-centre growth that leaves out on-site generation is describing a different, slower industry than the one operating.
Gas turbines are the clean option in their own category
A modern gas turbine is far from the worst way to make this power, but the comparison usually being made in the coverage is the wrong one.
Against the grid it displaces, on-site gas may be roughly comparable or worse depending on the local generation mix, and it is certainly worse than the same power arriving from a mix containing nuclear or renewables. Against diesel generation, which is the other technology available at this speed and scale, it emits far less particulate matter and far less sulphur.
It still emits pollutants. Gas turbines produce nitrogen oxides, which drive ozone formation, and formaldehyde among the hazardous air pollutants, regardless of their efficiency. Selective catalytic reduction cuts the first substantially, at capital cost and a small efficiency penalty, which is why the presence or absence of controls is the fact worth asking about rather than the fuel.
The permit question is the one to follow
The allegations in Memphis are about permits and controls rather than about emissions in the abstract, and that distinction is where the actual regulatory fight sits.
Air permitting generally turns on whether a source is major or minor, which depends on its potential to emit. A site running turbines described as temporary, or as backup, or as operating below a threshold, can attract a lighter permit than the same equipment declared as continuous primary generation. Whether the equipment then runs that way is a compliance question that is checked, if at all, afterwards.
This is how the category generally works, not a finding about any specific site. It explains why these disputes are about paperwork rather than chemistry, on which both sides usually agree.
Why this arrives here next
Johor, Batam and central Thailand are absorbing data-centre investment on the same timeline pressure, and the grid constraints are at least as tight. The turbine option is available in this region too, and the regulatory frameworks that would govern it are generally less developed than Virginia's rather than more.
The region does have different siting economics. Land is available further from dense population, and a facility placed away from a city has a smaller modelled health impact for the same emissions — which is a real mitigation and also a way of moving a cost onto people with less capacity to contest it.
We reported on the Bastrop foundry and why blade casting is the bottleneck in turbine supply. That piece asked whether the turbines can be built. This is the other half of the question. The two halves are usually reported by different people to different audiences.
View author profile → · Editorial policy
