Despite all the talk of “the cloud,” generative artificial intelligence relies on physical infrastructure: land, transmission lines, cooling systems and water. t does not float somewhere above the material world. It occupies land, draws water, consumes electricity and reshapes local environments.
Behind every AI system are data centers, which are massive physical facilities whose environmental footprint is now substantive enough to provoke political conflict, public resistance and regulatory backlash.
The expansion of this infrastructure is not just a technological development but a governance question — and current decision-making structures leave the communities that bear its costs with too little influence over how it is built. Universities are not bystanders in this buildout. They are major purchasers that help normalize demand while treating infrastructure as an external problem.
The scale of these data centers remains easy to overlook because public discussions tend to stay abstract, even though the inputs are measurable in units like megawatts, gallons and ratepayers’ bills.
In recent work, Vrije Universiteit Amsterdam researcher Alex de Vries-Gao estimates AI systems could require up to 23 gigawatts — on par with the electricity consumption of entire countries — while annual water use worldwide may reach hundreds of billions of liters, much of which is disclosed inconsistently or not at all.
In the United States, one widely cited estimate puts data centers’ indirect water footprint — water used at power plants to generate their electricity — at about 211 billion gallons in 2023.
And in parts of the country, the energy load from data centers is being used to justify new gas infrastructure, a buildout that can lock in carbon-intensive capacity for decades.
As a result, electricity grids are pushed closer to capacity, fresh water is diverted away from agriculture and residential use and carbon- intensive infrastructure is locked in for decades. What is often presented as an invisible digital upgrade is, in actuality, a long-term reshaping of energy systems and local resource allocation.
Across the country, proposed data center developments are triggering organized resistance, from rural counties considering moratoriums on new construction to suburban communities mobilizing against projects that threaten water availability and raise utility rates.
The conflict involves more than just impacts. Once a data center is built, transmission upgrades, substations and cooling systems harden into the landscape, contributing to what researchers call “carbon lock-in.”
There is limited empirical evidence today’s most resource-intensive generative AI systems are producing measurable emission reductions at scale, in part because independent environmental data is sparse and inconsistent.
• Alexander Voorhees is an Opinion Columnist who explores how national politics and institutions shape campus life and democratic legitimacy in his column “On Public Life.” He can be reached at [email protected].
