Sam Altman has found himself at the centre of another debate over AI’s environmental impact, this time involving something far less technical than data centres or energy grids: an almond.The OpenAI CEO recently said that 38,000 ChatGPT queries use roughly as much water as producing a single almond, a comparison that quickly spread online and left plenty of people asking a different question – what about the carbon emissions?The discussion picked up after a post from @IndianTechGuide on X shared Altman’s claim, bringing fresh attention to the amount of water associated with everyday AI use.Altman made the comparison while talking about criticism of ChatGPT’s water consumption during an appearance on Alex Heath’s Sources podcast. He pushed back against what he described as exaggerated claims circulating online about how much water a single AI query uses.“You know, I saw this thing going around about the water usage of ChatGPT. And it was like, every time you run a simple ChatGPT query, it’s like you run your shower for like six hours and the water never comes back and just it’s done,” Altman said.He then brought almonds into the conversation.Altman said that 38,000 ChatGPT queries use about the same amount of water as producing one California-grown almond, although he acknowledged that he didn’t have the exact calculations with him at the time.“There’s a question of like where this came from because the people scarfing down 12 almonds at a time don’t feel like they’re doing something horrible from a water perspective, for the most part,” he said.The comparison was catchy. And that’s probably part of the reason it travelled so quickly online.But it also opened up a much bigger conversation about how AI’s environmental footprint should actually be measured.
The numbers don’t line up quite so neatly
Almonds are already known for requiring significant amounts of water to grow, particularly in California, where drought and agricultural water use have long been closely watched.The Almond Board of California estimates that producing one almond requires about one gallon of water when the full growing process, including the development of the almond tree, is considered.Altman has previously estimated that a single ChatGPT query uses about 0.3 millilitres of water, or roughly one-fifteenth of a teaspoon.Those figures don’t produce the same result as the 38,000-query comparison.At 0.3 millilitres per query, one gallon would work out to around 12,600 queries – considerably fewer than 38,000.That doesn’t necessarily mean one figure is simply right and the other is wrong. Water-use estimates can change depending on what’s being counted, how the calculation is made and whether the figure refers only to direct data-centre water use or includes other parts of the system.Still, the gap was enough to get people talking.
Altman pushes back on data-centre water claims
Altman also addressed concerns about the amount of water used to cool large data centres.He argued that the picture often presented online doesn’t reflect how modern facilities operate, saying evaporative cooling was more common in the past.“It is true that data centers at one point used evaporative cooling, but they have not done that in a long time. Like if you look at a modern, very large data center, it uses, like, the equivalent amount of water as an office building in terms of people, you know, running the sinks and the toilets and whatever.”Altman described some of the online discussion around data-centre water use as a “robust meme” and said the claims don’t “(hold) up to any scrutiny.”But for many people following the debate, the bigger issue isn’t just how much water is used for cooling.It’s everything around it.
And then came the CO2 question
One X user challenged the focus on water and pointed toward the emissions created by the energy needed to run AI infrastructure.“What about the CO2 emissions though ? I’m not sure they’re building more gas turbines and consuming more coal to produce almonds. And we all know these models don’t just consume energy and resources for inference but also during the training process,”That’s where the almond comparison gets complicated.A person asking ChatGPT a question doesn’t see the servers, electricity infrastructure or cooling systems working behind the screen. The request feels almost weightless. Type something, press send, get an answer.But at the scale of billions of queries, even small amounts of energy and water can add up.One commenter tried to illustrate that scale, writing:“12L of water is required to grow and almond. Chatgpt alone get 2.5 Billion prompts a day. So 65 million almonds a day. And 800 million liters a day. That is around 200-400 million peoples water, This is only from ChatGPT BTW,”Others took a more philosophical view of the issue.“We have to fill our stomachs with grains and fruits, not with data or AI. Water is for nature, it is for living life, not for inanimate things,” one user wrote.Another pointed out that looking only at water used directly by a data centre may leave out water consumed elsewhere in the energy chain.“That’s cooling water at the datacentre. the power station making the electricity uses its own water, and that’s often the bigger number.”And then there were, inevitably, the jokes.“Don’t eat almonds to sharpen your brain, use that water for ChatGPT. Make ChatGPT your brain 😂😂” one commenter wrote.
The scale of AI changes the conversation
There’s a reason the almond analogy is easy to understand. Comparing tens of thousands of AI queries with one familiar food item gives people a simple way to think about something that is otherwise difficult to visualise.But the comparison also has an obvious limitation.People don’t send one ChatGPT query and stop. AI services handle enormous volumes of requests every day, while large models also require significant computing resources during development and training.That’s why some users argued that looking at the environmental cost of a single prompt misses the bigger picture.“The almond comparison works great until you multiply it by the billions of queries ChatGPT actually runs a day,” one person wrote.And that may be the more useful question behind the online debate. Not whether one ChatGPT query is comparable to one almond, but what happens when millions or billions of small actions are repeated every day.AI’s environmental footprint isn’t one number. It involves electricity, water, hardware, data centres and the infrastructure needed to keep all of it running.So while Altman’s almond comparison has made for an unusually memorable internet debate, the question people are now asking is considerably bigger: how much environmental cost comes with the convenience of having AI available every time we open a chat box?