The tech industry’s accelerating buildout of infrastructure to power artificial intelligence is rapidly turning an industry once lauded as “clean” and environmentally friendly into an air polluting, ecosystem destroying, water guzzling behemoth. Now, there’s an intensifying rift on the left about how to approach what was, until recently, a steadfast Democratic ally.
Progressives are now at a fork in the road with two very different options: a political reckoning with Silicon Valley or a rapprochement paid for with environmental havoc.
Some pundits and industry figures have counterintuitively argued that the proliferation of data centers to power AI is a good thing for the environment. The massive energy demand for training artificial intelligence will, in this telling, necessarily prompt a massive investment in clean energy and transmission infrastructure to meet that demand, thereby catalyzing a world-altering transition toward renewable energy. This argument, already suspect years ago, is entirely untenable now.
Following US President Donald Trump and company’s evisceration of the clean energy investments from the Inflation Reduction Act (IRA), the narrow path of AI buildout being aligned with a green transition is now completely walled off. The choice now is whether the United States continues to aid and abet Silicon Valley’s environmental rampage or to fight it.
At present, there is simply no way to have the scale of AI buildout that the United States is seeing without terrible environmental downsides.
Even prior to Republicans torpedoing the IRA, AI electricity demand was growing faster than both renewable energy production and overall grid capacity. Without strong additionality regulations to require that new data centers be powered by the construction of new renewable energy generation, the AI boom will continue to increase consumption of fossil fuels.
Much of the increased energy demand was already being met by natural gas before the Republican spending package. It’s only going to get worse now. Without the clean energy tax credits, the advantages of incumbency that fossil fuels enjoy mean that the AI energy boom will further hook us on unsustainable resource consumption.
The firms building out AI infrastructure know this and often point to major investments in clean energy to protest characterizations of data centers as environmentally disastrous. But there are two major problems there. First, those investments may be in totally different locations than the actual data centers, meaning the centers are still consuming dirty energy. Second, and more importantly, at our present juncture in the climate crisis, we need to be actively decreasing our use of fossil fuels, not just containing increases in dirty energy production. (It’s worth noting that AI is also being used to enable more fossil fuel extraction.)
And the environmental destruction doesn’t stop there. The Trump White House recently moved to exempt data centers from environmental review under the National Environmental Policy Act, or NEPA, paving the way for tech companies to despoil local environments without a second thought, and limiting opportunities for the public to gain information about data centers’ environmental impacts.
Perhaps nothing captures the excesses of AI quite so clearly as its water usage. Despite some pundits glibly claiming that there’s actually tons of water to go around, data centers threaten to worsen already dire droughts. We’re already beginning to see this in arid places like Chile and the American Southwest.
The Colorado River’s mismanagement is the stuff of public policy legend at this point. Aquifers across the Western US are being depleted. People were not mulling the idea of partially rerouting the Mississippi River for giggles. There is, unequivocally, a water crisis unfolding. And those data centers are very, very thirsty. A single data center can use millions of gallons a day.
There are already more than 90 data centers in the Phoenix area alone. That’s hundreds of millions of gallons of water a day. Protesting that “there’s plenty of water” is not just detached from the drought-stricken reality, it’s dangerous.
Data centers are being built in arid places intentionally; the low humidity reduces the risk of corrosion for the processor stacks warehoused there. Fresh water supplies, when depleted, are not easily renewed. Devoting more of it to cooling GPUs means less for drinking, irrigation, fighting wildfires, bathing, and other essential uses.
And there isn’t a way to bring water to the arid environments to mitigate that, either. Some people point to desalination, but that isn’t tenable for multiple reasons. To start, most of these data centers tend to be inland, as the sea air has similar corrosive effects as humidity. That, in turn, means that even accepting desalination as a cure for water scarcity, data centers would require transporting massive quantities of that purified water over significant distances, which would require complex energy-and resource-consuming engineering projects unlikely to proceed within the hurry up and go of our AI bubblish moment. (Desalination also has its own serious environmental harms.)
At present, there is simply no way to have the scale of AI buildout that the United States is seeing without terrible environmental downsides. The only choice left is whether to get out of Silicon Valley’s way or whether to slow the industry’s pace.
