AI and the Water Wasting Machine

An argument that’s frequently made against the use of AI (specifically, popular chatbots based on LLMs, like ChatGPT, Gemini, Copilot, etc.) is that AI is harmful to the environment. The hardware required to create, train, and run chatbots consumes a significant amount of energy, which in turn requires the use of natural resources, specifically water. Many articles have been written recently that focus on AI’s water usage, with some saying that AI is “accelerating the loss of our scarcest natural resource,” “draining water from areas that need it most,” and that by 2030 it will “match the annual drinking water needs of the United States.”

While the fact that AI needs some amount of water to operate isn’t disputed, there is much more debate around how much it uses. Some AI-defenders, for instance, have argued that the amount of water AI uses is either negligible or at least comparatively negligible given the amount of water that other things use and that we are seemingly okay with. For instance, Sam Altman postulated that a single ChatGPT query uses “roughly one fifteenth of a teaspoon” of water per query, while others have written that the AI industry as a whole in the US uses about as much water as its golf courses.

Indeed, “you’ve been thinking about AI’s water use wrong” content is a plentiful resource in 2026. For example, an article from Wired, fittingly titled “You’re Thinking About AI and Water All Wrong,” notes that many early estimates of AI water use are likely off by a not-insignificant amount. And a popular Substack article, less-subtly titled “The AI water issue is fake,” lists the following items alongside the approximate amount of water they take to produce when compared to a prompt for a chatbot:

Leather Shoes – 4,000,000 prompts’ worth of water
Smartphone – 6,400,000 prompts
Jeans – 5,400,000 prompts
T-shirt – 1,300,000 prompts
A single piece of paper – 2550 prompts
A 400 page book – 1,000,000 prompts

These kinds of comparisons are intended not only to challenge the narrative that AI uses a lot of water, but also to defuse the moral argument against AI use on the basis of water consumption. As the Substack author notes: “If you want to send 2500 ChatGPT prompts and feel bad about it, you can simply not buy a single additional piece of paper. If you want to save a lifetime supply’s worth of chatbot prompts, just don’t buy a single additional pair of jeans.” In other words, we risk being moral hypocrites if we criticize AI for its water usage while still reading books and wearing comfortable clothes.

I am not going to question the exact numbers concerning AI water use. Instead, I want to consider the merits of the “you’ve been thinking about AI’s water use wrong” arguments, specifically whether we really are moral hypocrites for criticizing AI on the basis of its water use. If we are okay with playing golf and using phones and doing all sorts of other things that use water, must we refrain from criticizing AI on the same basis? Are the people who think that we shouldn’t use AI because of its water use really thinking about it “all wrong,” or chasing a “fake” issue?

I think that the answer is no: we can still criticize AI on the basis of its water consumption. To illustrate, consider the following thought experiment:

The Water Waster: A company announces a new product: the Water Waster 3000. Here’s how it works: it uses [one/one hundred/one thousand] gallons of potable water every day to spin a wheel. The wheel is not connected to anything, and its only tangible benefit is that some people enjoy watching the wheel spin.

We can write the thought experiment so that the Water Waster 3000 uses different amounts of water. However, no matter which number we chose, we would likely reach the same conclusion: that operating the Water Waster 3000 is a waste of water. Someone who really enjoyed watching wheels spin might argue that, in comparison to other forms of entertainment, a single spin of the wheel uses only a fraction of the amount of water required by other industries. But this argument doesn’t hold much water (so to speak): it doesn’t matter that other things use much more water; what matters is that the Water Waster 3000 wastes it.

It is, of course, unfair to say that the Water Waster 3000 is the exact same thing as an AI chatbot. However, one might argue that AI chatbots and much of the AI industry are akin to the Water Waster 3000 in the sense that the AI industry does not, by and large, produce anything of significant enough value, or at least not enough value to warrant its environmental impact. As argued in the aforementioned Wired article: “People who don’t think twice about eating a burger or buying a new T-shirt are angry about LLMs and water because they are rejecting the entire premise that AI is worth the price of its water use.”

If we approach the argument this way, then we can also avoid the charge of hypocrisy. After all, people need t-shirts and shoes: those things require water to produce, but they’re worth it. Sure, producing clothing has a negative impact on the environment, but since the trade-off is acceptable, it’s not a waste of water, just a use of it.

But does this argument free us from the charge of hypocrisy? After all, we do all sorts of things that arguably do waste water, but still do not subject them to as much scrutiny. For instance, while people certainly need shoes and t-shirts and such, we definitely don’t need as many as people tend to acquire (for example, the environmental damages of so-called “fast-fashion” – mass-produced, cheap and disposable clothing – are well-documented). That AI-enthusiasts often compare the water use of the AI industry to golf courses is thus particularly apt, not only because of the alleged similarities in the amount of water they use, but also because golf doesn’t serve any practical need. Golf is frivolous (arguably), so if we’re happy as a society to accept using a lot of water so that some people can more easily hit a small ball into a small hole that’s really far away, then arguably we should be okay using a similar amount of water for chatbots and other potentially much more useful things.

Of course, the claim that people are hypocritical when it comes to criticizing AI for its environmental impact does not negate the fact that AI use still has an impact on the environment. Using AI can then still be bad because of its water usage, just as it is (arguably) bad that we use a lot of water on golf courses. After all, the claim of moral hypocrisy is not a claim about the rightness or wrongness of an act, but instead about whether one has the grounds to criticize others because of their own acts.

So, perhaps AI is bad because it uses water, but then so is everything else because everything uses water. Isn’t it unfair, then, to single out AI?

Well, not really. We are, after all, in the unenviable position of living in a system where hardly any of our acts as consumers or users of technology are environmentally neutral. If this is enough to undermine our moral authority to criticize, then we could not criticize any acts that have a negative environmental impact, unless those acts are significantly disproportionately more extreme compared to what we do now. But it’s not clear that this is a fair standard to hold people to. Remember our Water Waster 3000: we seem perfectly within our rights to criticize the existence of such a machine, even though we use other, more useful machines that also use water.

The Wired analysis is then perhaps not entirely fair to AI critics: it is not necessarily that people who choose to eat burgers or buy new t-shirts don’t give any thought to the environmental impact of their actions – they might – nor that they think that burgers and t-shirts are worth the environmental costs – they might think they ultimately aren’t. But you can still criticize someone for leaving their tap running even if you happened to have bought a new pair of jeans recently. If AI is a waste of water, then we are not hypocrites for criticizing it.

Of course, the AI-enthusiast would likely reject the idea that AI use really is akin to leaving a tap running or something like the Water Waster 3000, or indeed is a waste of water at all. While it is hard to defend the value of every individual use of an AI chatbot, we might still think that, overall, AI should be conceived of as a use of water that produces something of value, rather than a waste. Maybe what we need to calculate is something like “value-per-milliliter of water used,” where values above a certain level would qualify as being “worth it” while those below are “not worth it.” Where the critic and the enthusiast might disagree, then, is whether AI falls above or below that line.

The ongoing disputes around the amount of water AI uses make this calculation practically impossible. But the charge of hypocrisy assumes that we have an answer to this question, namely that AI is, in fact, worth the water it uses. If we reject this premise, or even just call it into question, then we do not lose our moral ground to hypocrisy.

So, where does this leave us? We’ve seen that it is not enough to defuse the environmental argument against AI use to simply say that AI uses as much or less water than other things that we tend to think are acceptable, either at the level of queries, companies, or the industry as a whole. Nor does the fact that we use water for some things mean that we automatically undermine our ability to criticize AI use on the same basis: we do not lose our moral standing to criticize AI for its environmental impact if we reject the idea that AI is worth its water use. While there will undoubtedly be many more arguments about how much water AI uses, it is still criticizable on that basis.