Data centers can look like remote technology infrastructure, but their effects are local: they draw on electricity and water, and their construction can shape energy decisions for years. Paris Marx’s book Hyperscale: The Ambition and Excess of Big Tech’s Data Empires traces how communities outside the United States encountered those pressures early—and what their responses may suggest as US opposition grows.
Earlier warnings came from abroad
Marx says moratoriums in the Netherlands and Singapore date back to 2019. In 2022, attention turned to Ireland, where data centers were placing growing demands on the electricity grid. The issue was not simply whether companies wanted to build more facilities; it was whether the power system could accommodate them.
A separate case in The Dalles, Oregon, put water use and public information at the center of the debate. According to Marx, Google sought to keep its water-use figures secret through the city council. After the community obtained the numbers, it learned that Google was using 29 percent of the community’s water.
Then ChatGPT came out at the end of 2022, and the prospect of more data centers to support generative AI became harder to ignore. Places already experiencing construction and resource demands offered an early view of what could reach other communities.
Ireland shows how infrastructure pressure can shift
Dublin’s moratorium was passed in 2021 after the grid operator said there was not enough capacity to connect more data centers. Marx describes how the industry responded by looking beyond Dublin, exploring use of the gas network and generators, and eventually building dedicated gas power plants for data centers.
That sequence matters because a limit on one kind of infrastructure can redirect development toward another. A data center that cannot connect to the public electricity grid may still prompt new energy infrastructure elsewhere. In Ireland, operators also received emissions licenses from the Irish Environmental Protection Agency to burn fossil fuels. The agency warned that this would make climate targets harder to achieve.
The ownership of energy infrastructure also became a concern. Ireland’s grid is public, and only the public utility can build electricity lines. Marx says data center companies lobbied for legislation that would allow private companies to build cables, raising questions about equity in the system.
These examples offer a way to assess data center proposals: look beyond the building itself. Consider where its electricity will come from, what happens when grid capacity is limited, and who controls the infrastructure built to serve it.
An AI downturn could leave lasting assets
Marx believes the industry is in an AI bubble and questions whether all the planned facilities and power plants will be built before demand is reassessed. He points to obstacles in obtaining turbines, chips and other components. The timeline for construction may therefore collide with uncertainty about how much AI infrastructure will be needed.
There is precedent for a technology build-out leaving behind facilities after companies fail. Marx recalls that during the dotcom boom and bust, data center companies went under, while Equinix survived and bought many data centers in preparation for demand to return.
He argues that the current situation could be different because facilities designed for AI depend on chips that may need to be replaced within a few years. If demand later returns, the buildings could remain while the equipment that made them valuable no longer serves the same purpose.
Power plants pose a related challenge. Even if planned data centers are never built, gas plants could still exist and seek other customers to justify their operation. The infrastructure decision can therefore outlast the business case that prompted it.
Community needs can shape a different model
Marx points to a community-led project in New Zealand as an alternative: creating a more accurate speech-recognition model for the Māori language and using local cloud providers to host the data. The example connects AI infrastructure to a defined community purpose and local hosting, rather than assuming that ever-larger facilities are valuable by default.
He argues that governments should set firmer rules for data center development alongside rules for AI products. He also calls for a more basic assessment: how much computation is needed, what resources are justified, and what people receive in return.
Those questions bring the debate back to communities facing construction decisions. Ireland’s experience shows how electricity constraints can push operators toward gas and private infrastructure; The Dalles shows why access to resource-use information matters. New Zealand’s example suggests that communities can also ask what infrastructure should enable, and who it should serve.