The Data Center Boom Has a Biodiversity Problem

Somewhere not far from a river, a plain grey building the size of several football fields hums day and night. It has no windows and almost no people inside. What it has is thousands of servers, and what those servers need, in enormous quantities, is land, electricity, and water. Multiply that building by the thousands now being constructed to feed the world’s appetite for cloud computing and AI, and you have one of the fastest-growing and least-discussed pressures on the natural world. This is an attempt to look at that pressure honestly, from every side, by people who help build the very systems that cause it.

Why the data center boom is accelerating

For two decades, data center growth tracked the steady rise of the internet. The current surge is different in kind. Training and serving large AI models consumes orders of magnitude more compute than the web applications that came before, and every major technology company is racing to build the capacity to run them. The result is a construction wave: larger campuses, built faster, in more places, than at any point in the history of computing.

The crucial point is where that capacity lands. Facilities are sited for cheap land, cheap power, and cool climates or abundant water, which means they cluster in a handful of corridors. The environmental cost is not spread thinly across the globe. It is concentrated in specific landscapes, which is exactly what makes it a biodiversity problem rather than a diffuse carbon statistic.

The land footprint we do not see

A data center is not just a box of computers. It is a land transformation. Campuses increasingly cover hundreds of acres, and they are rarely built on wasteland. They rise on farmland, wetlands, and forest edges, often in the same green corridors wildlife uses to move, feed, and breed. When habitat is cleared or carved into pieces, the species that depended on it do not simply relocate. Many of them decline.

Fragmentation is the subtler harm. A single large site does not just remove the acres it occupies. It splits what remains, cutting a population in two, blocking a migration route, and isolating groups that then lose genetic diversity over time. Ecologists have measured this effect for decades around roads and housing. A sprawl of windowless warehouses does the same thing, only we are less used to counting it.

Thirsty machines and the water problem

Keeping those servers cool takes staggering amounts of water. A single large facility can consume millions of litres a day, often drawn from the same aquifers, rivers, and reservoirs that local ecosystems and communities rely on. In water-stressed regions, that drawdown lowers water tables, shrinks wetlands, and warms the water that is returned, all of which hits aquatic life first and hardest.

The timing makes it worse. Cooling demand peaks in hot, dry weather, which is precisely when rivers run low and ecosystems are already under strain. A facility that is a minor draw in a wet spring can become a serious competitor for water in a drought summer, pulling against the needs of fish, amphibians, and the people living alongside them.

A server farm does not look like deforestation or a drained wetland. But trace its land, water, and power back to the source, and that is often exactly what it is.

The power problem behind the power problem

The electricity has to come from somewhere. New data centers are driving a surge in demand that, in many places, is met by keeping fossil plants alive longer or building new generation. That means more emissions, more mining, and more transmission lines cutting through habitat. The climate impact loops straight back to biodiversity, because the species already struggling with a warming world are asked to absorb one more pressure.

Even clean power is not free of footprint. New solar and wind capacity needs land and transmission corridors of its own, and if it is simply consumed by data centers it does not displace a single tonne of fossil generation elsewhere on the grid. Genuinely additional renewable capacity helps. Capacity that would have been built anyway, relabelled to make a facility look green, does not.

The quieter local harms

Closer in, the effects are physical and often overlooked. These sites radiate heat, run bright through the night, and produce constant low-frequency noise from cooling systems and backup generators. For nocturnal animals, migrating birds, and insects, artificial light and sound disrupt navigation, breeding, and feeding in ways we are only beginning to measure.

Impacts worth tracking

• Habitat loss and fragmentation as campuses expand into farmland, forest, and wetland, breaking the corridors species need to survive.
• Water stress on rivers, aquifers, and wetlands from cooling demand, with fish and amphibians among the first to suffer.
• Local warming and pollution from heat discharge, light, and noise, which disrupt nocturnal and migratory wildlife.
• Compounding climate pressure from fossil-backed power, accelerating the biodiversity loss already underway.
• Cumulative regional damage as dozens of facilities cluster in the same low-cost land and power corridors, concentrating the harm in one place.

What building responsibly looks like

None of this is an argument to stop building. It is an argument to build with the full cost in view, and the choices that reduce the harm are known and available.

• Site on degraded or industrial land rather than habitat, and avoid intact wetland, forest, and wildlife corridors.
• Use closed-loop and air cooling where water is scarce, and reuse waste heat for nearby buildings instead of dumping it.
• Power facilities with genuinely new renewable capacity that would not otherwise exist, not relabelled existing supply.
• Measure and publish water, land, and energy impact per facility, so the cost can be scrutinised rather than hidden.
• Treat biodiversity as a siting and design constraint from day one, not a restoration project tacked on at the end.

Frequently asked questions

How much water does a data center actually use?

It varies enormously with design and climate, but a single large water-cooled facility can draw on the order of millions of litres a day. The figure matters less than the context: the same water often feeds local rivers and aquifers, so the impact is heaviest in dry regions and dry seasons.

Do AI data centers really harm wildlife?

Indirectly but measurably. The direct harm is habitat loss and fragmentation where campuses are built. The indirect harm comes through water withdrawal, heat and light pollution, and the emissions from the power they draw, all of which press on species already under strain.

Is AI worse for the environment than regular computing?

Per task, training and running large AI models is far more compute-intensive than traditional web services, so the energy and water footprint per useful output is higher. The scale of the current build-out is what turns that per-task cost into a landscape-level pressure.

Can data centers ever be environmentally sustainable?

They can be far less harmful. Siting on degraded land, closed-loop cooling, waste-heat reuse, and genuinely additional renewable power all shrink the footprint substantially. Sustainable in an absolute sense is harder, but the gap between a thoughtless facility and a careful one is enormous.

Key takeaways

• The AI-driven data center boom concentrates its land, water, and power demand in a few vulnerable corridors.
• The clearest harms are habitat loss, fragmentation, and water stress on rivers and aquifers.
• Fossil-backed power links the boom straight back to climate pressure and the biodiversity loss it drives.
• Heat, light, and noise from facilities disrupt nocturnal and migratory wildlife at the local scale.
• Responsible siting, closed-loop cooling, and genuinely new renewable power can cut the damage dramatically.

The cloud feels weightless because we never see it. But it sits on real ground, drinks real water, and casts a real shadow over the living systems around it. We build AI, so we are part of this, and pretending otherwise would be dishonest. If the next decade of AI is going to be built on these buildings, the least we owe the ecosystems paying the bill is to notice, measure, and build better, which is the standard we try to hold ourselves to at Vesper Labs.