The footprint of artificial intelligence is brutally material with its data centres requiring heavy mining to make their microchips, huge electricity demand, escalating water use with rising costs that are pushed onto consumers. Erald Kolasi warns of a forthcoming backlash in an interview with The Mint.
Artificial intelligence is sold as weightless. It appears on our screens as a stream of prompts, answers, images and predictions, seemingly detached from the physical world. But Erald Kolasi wants to drag the discussion back to earth. AI, he argues, is not just software. It is an infrastructure system of data centres, chips, steel, cement, cooling towers, electricity grids, water supplies and political favours. And that system is now expanding fast enough to reshape local economies, public utilities and democratic conflict.
“When we talk about AI,” he says, “I’m talking about deep learning neural networks… ChatGPT, Gemini, Llama, Grok… image generators and video generators like Sora.” What matters, he insists, is that these systems are physically grounded. “Your prompt goes to the data centre. There’s a lot of computer chips in there. They do a bunch of very fast calculations, and then they give you an output.” That is why AI needs energy. “You need to produce a lot of electricity to keep these data centres going.”
To produce the steel for just one hyperscale data centre takes about 100 million gallons of water.
That alone would be significant enough. But the real story, Kolasi argues, is bigger than electricity demand narrowly understood. The computers in data centres run hot. Historically cooling was air-based. “Now, increasingly, they’re liquid powered,” he says. “So they’re consuming a lot of water.” In the US, he notes, the scale-up has been staggering: “In 2024, they consumed 4% of national electricity. In 2000, it was 0.1%.” His conclusion is blunt: “It’s just crazy the growth that they’ve had.”
Kolasi’s argument becomes sharper when he moves upstream. The data centre is only the visible end-point of a much larger material chain. To build one hyperscale site, he says, takes vast quantities of steel and cement. “To produce the steel for just one hyperscale data centre takes about 100 million gallons of water.” The chips inside those facilities depend on one of the most complex manufacturing systems ever created. TSMC produces many of the specialist chips. ASML makes the lithography machines used to produce them. “For an extreme ultraviolet lithography machine,” Kolasi notes, “it takes ASML to transport one of those machines to TSMC in Taiwan, three cargo planes, 40 freight containers, and 20 trucks on average.” AI may look like an ethereal digital service. In reality it is a globe-spanning industrial build-out.
He is particularly interested in what happens when this growth collides with everyday infrastructure. For roughly two decades, US electricity consumption was comparatively flat. Then AI arrived as a fresh demand shock. “There’s been a mad rush to generate more electricity in the United States here,” he says. Utility companies are scrambling to build new capacity, and consumers are picking up the tab. Kolasi is careful not to pretend there is only one cause of rising bills. In California, for example, wildfire damage has also pushed costs higher. But his broader judgment is clear: “The rise of data centres in the United States is… probably the major factor over the past five years, driving electricity rates higher.”
That matters politically because the costs are no longer abstract. “It’s leading to sharply higher electricity bills,” Kolasi says. “It’s leading to a lot of public anger.” The same applies to water. Data centres, he argues, are relying heavily on public systems while returning little to local watersheds. “They’re consuming a lot of that water from public water utilities,” he says, and “they don’t even return much of it… because then through evaporative cooling, they throw it out.” The result is obvious enough: “Now water rates are also rising.”
“There’s been a mad rush to generate more electricity in the United States here,” he says. Utility companies are scrambling to build new capacity, and consumers are picking up the tab.
Kolasi describes growing local resistance across the US, particularly in water-stressed regions. In Arizona, he says, the issue is obvious: “It’s very dry, so there’s huge concerns around water supply.” In places such as Tucson and elsewhere in the state, planned data centres have triggered fierce opposition. “A lot of planned data centres have been canceled,” he says. Local meetings “can get pretty heated.” Becker, Minnesota, he notes, turned down an Amazon data centre.
Part of the anger comes from the secrecy. “These AI companies, when they come in, they’re very shady and cagey,” he says. “They come in with non-disclosure agreements.” Residents may see “this huge facility going up” without being told whether it is owned by Amazon, Google or Microsoft. “Once those things are disclosed,” Kolasi says, “that leads to a lot of local public opposition.”
He gives the example of Musk’s xAI operation in Memphis to show how far the scramble for power has gone. According to Kolasi, the company ran gas turbines for the Colossus supercomputer without proper permits for a prolonged period, arguing they were outside normal Clean Air Act rules. “It has gotten that bad,” he says, that firms are “scrambling to put… modular energy generation units on site for these data centres if they can’t always rely on the utility grid.” Whatever the legal details, his wider point is unmistakable: the AI boom is already pushing against environmental rules and public tolerance.
At the federal level, he sees little meaningful restraint. “The federal government is totally behind big tech right now,” he says, “totally behind Sam Altman and Dario Amodei and Elon Musk and whatever they’re doing.” Republicans are openly enthusiastic because AI is driving growth. But Democrats, he argues, are implicated too. “The Democrats were in power when ChatGPT came into the public consciousness,” he says, and “Biden didn’t really do much to stop them.” Real resistance, for now, sits lower down: states, cities, local boards, communities. That is where officials are considering tougher taxes, better measurement of consumption and ways to contain the hit to household bills.
This industry is heading for a major fight in American political economy.
Kolasi thinks this tension is only going to intensify. “This industry is heading for a major fight in American political economy,” he says, “because this level of growth is clearly unsustainable.” He compares the current boom to the railroads in the 19th century: explosive expansion followed by inevitable bust. AI, he says, is following the same pattern. “It’s making life more unaffordable. It’s destroying local ecology in many places. And that is absolutely a political issue.”
He expects a crash. “Every time there’s been a major technological wave in modern history, there has been some kind of crash,” he says. The dot-com bubble is one example; railways another. The trigger this time, he argues, is debt. AI firms are borrowing heavily. Private equity is borrowing heavily. Banks are exposed. “Eventually it’s unsustainable, this debt is going to go toxic.” But he is equally clear that a crash will not make the industry disappear. Growth rates may fall. Ownership may change. Some projects may collapse. Yet “the world of AI… is here to stay.”
That may be the most important point of all. Kolasi is not saying AI vanishes. He is saying the fantasy vanishes: the fantasy that intelligence can be scaled without friction, that digital systems somehow escape the material world, that the costs can be hidden in distant landscapes and utility bills without triggering a reaction. The politics of AI, in his account, are no longer about the future alone. They are already showing up in drought zones, in planning rows, in power prices, in dirty backup turbines and in the question ordinary people are beginning to ask: who benefits, and who pays?
