Paris Marx warns US communities of data-center costs in new WIRED interview

Limits on electricity connections do not necessarily stop data-center development: they can push operators toward their own gas-fired power. In an October 25 WIRED interview, author Paris Marx pointed to Ireland as a warning for US communities weighing AI infrastructure, arguing that its demands on electricity, water and emissions deserve scrutiny alongside the technology’s promised benefits.

Marx, a longtime technology-industry critic and host of the Tech Won’t Save Us podcast, spoke with WIRED’s Molly Taft about his new book, Hyperscale: The Ambition and Excess of Big Tech’s Data Empires. His examples trace conflicts that predate ChatGPT, rather than treating today’s US opposition as a sudden reaction to generative AI.

When a grid limit becomes a gas plant

The Irish example begins with a physical constraint. Marx said Dublin imposed a moratorium in 2021 after the grid operator concluded it lacked capacity to connect more data centers. The restriction concerned access to electricity through the grid. Operators responded by seeking sites elsewhere in Ireland and other ways to supply their facilities.

Those alternatives included tapping the gas network for generators and eventually building dedicated gas power plants, he said. That changes the infrastructure behind a data center without removing its need for energy. Marx also said operators received fossil-fuel emissions licenses from Ireland’s Environmental Protection Agency, which warned that the additional burning would make climate targets harder to meet.

The dispute extended to who could build electricity infrastructure. Marx described an Irish public grid where only the public utility could build power lines. Data-center companies lobbied for legislation allowing private companies to build electricity cables, he said, raising concerns about fairness within the system. His warning reaches beyond whether an individual project wins a connection.

Water offers another lesson about local oversight. Marx accused Google of seeking to keep its water-use figures secret in The Dalles, Oregon, with the city council helping shield them. He said the community obtained the figures after the better part of a year and learned that Google used 29 percent of its water. He cited the episode as an early reason to investigate data centers.

The buildings and chips have different futures

Marx believes the AI industry is in a bubble. Asked what a collapse could leave behind, he questioned whether planned facilities would even be completed before demand was reassessed. Shortages of turbines, chips and other equipment complicate that timing, he said. His scenario is a warning about uncertain demand, not a prediction of a specific collapse date.

He contrasted today’s plans with the dotcom bust, when surviving companies such as Equinix bought data centers from failed operators and held capacity for returning demand. For AI facilities, he raised a complication: under the assumption that chips need replacing within a few years, the equipment that gives a building much of its value might no longer be useful when customers return.

Gas plants could have a different afterlife. Marx expects plants built for data centers to seek other customers if their intended demand disappears, rather than simply shut down. That is his concern about the buildout’s lasting consequences: the market for AI computing could change while the fossil-fuel infrastructure remains.

How many resources are actually justifiable to put into this project?

Paris Marx, speaking to WIRED

A community purpose, not just more capacity

The interview also considers an alternative in Marx’s book: a community-led Māori speech-recognition project in New Zealand, using local cloud providers to host its data. Taft highlighted it as a different approach to AI, tying computing to a specific community purpose rather than treating infrastructure growth as the goal.

Marx wants governments to pair stronger data-center rules with regulation of AI products themselves. He argues that promised benefits have not justified the financial, social and environmental resources committed. That is his judgment, not a cost-benefit finding presented in the interview. His proposed test starts with what communities get from the computation, not merely whether developers can supply enough power.