I am tired of hearing that AI is “going to use up all the water,” that it consumes excessive resources, and that we should stop its development because of it. Many times, instead of arguing, I simply listen, because I genuinely want to know whether the real concern comes from caring for the environment, or whether I am just talking to someone who saw an Instagram post and decided that was enough to be against AI.
If you are one of those people who does not “support” artificial intelligence because “we are going to run out of water” or because “it uses too much energy,” let me tell you that the first thing you should stop doing is watching any type of content on streaming platforms such as Netflix, YouTube, HBO Max, etc.
Water is used for cooling systems in computers, servers, and data centers. And if we make an estimate for comparison, an average message to ChatGPT uses approximately ~0.82 mL of water, while one hour of video streaming can use approximately ~114 mL, considering the indirect water associated with electricity consumption. You would have to send around 138 messages to ChatGPT to use approximately the same amount of water you use watching one hour of Netflix or YouTube.
This is based on estimates published by Sam Altman, CEO of OpenAI, in June 2025, combined with estimates of water consumption associated with electricity generation. Today, there are even more efficient models that use less, and yes, there are also models and tasks that can use more —for example, video generation, images, long tasks, or agents—, but this comparison is based on an average text query, which is the most common use.
If we compare it with Gemini, Google's AI, based on Google's public data, the math gets even more interesting. One message to Gemini uses approximately ~0.61 mL of water, including the estimate of indirect water from electricity. In other words, it would take around 185 messages to use approximately the same amount of water as one hour of video streaming.
| Activity | Direct water | Electricity-related water | Total estimated water |
|---|---|---|---|
| 1 Gemini text prompt | 0.26 mL | ~0.35 mL | ~0.61 mL |
| 1 ChatGPT query | ~0.32 mL | ~0.50 mL | ~0.82 mL |
| 1 hour of streaming video | Unclear / small server share | ~114 mL | ~114 mL |
Energy and CO₂
To compare things more directly, let us review the numbers for energy consumption and CO₂ emissions, since, for some reason, many media outlets have focused on water consumption as if energy use and emissions were not important.
The International Energy Agency estimates that one hour of video streaming uses around 77 Wh and emits around 36 gCO₂. This means that, when comparing energy use, one hour of streaming is equivalent to approximately 226 ChatGPT messages or around 321 Gemini messages. On top of this, Google has published its estimate of emissions per Gemini prompt, so we can calculate that one hour of streaming is equivalent to approximately 1,200 Gemini messages in CO₂ emissions.
Worried about the environment? Stop eating meat
Now, let us be direct. If you are that worried about water or the environment, you would be better off starting by reviewing your meat consumption. Our World in Data reports that producing 1 kg of beef can emit around 60 kgCO₂e. So, for every 100 g of beef you eat, approximately 6 kgCO₂e were emitted. That is equivalent to around 200,000 messages with Gemini.
What do you think now?
Data Centers
At the end of the day, the main problem is not that someone asks ChatGPT a question. The real problem is the accelerated construction of data centers: these enormous infrastructure farms that keep servers, data, and services connected to the internet.
As you can imagine, the companies building the most data centers are Google, Meta, Amazon, Microsoft, and other tech giants. And this is where there is a serious conversation to be had. There are reports of communities complaining about water use, energy, land, noise, and possible environmental impacts around new data centers. That does deserve attention, regulation, and transparency. But we also need to tell the complete story: data centers do not exist only for AI. They are what make it possible for you to watch Netflix, shop on Amazon, use Facebook, WhatsApp, Google, iCloud, Spotify, Uber, digital banking, online video games, and practically everything we do on the internet.
Amazon, for example, does not build data centers only for AI; much of that infrastructure exists for AWS, e-commerce, streaming, and other digital services. Google, meanwhile, is investing heavily in improving the efficiency of its data centers, reducing emissions, and finding cleaner energy sources. One clear example is its commitment to Enhanced Geothermal Energy: a technology that aims to generate clean, constant electricity using the Earth's heat.
I do believe that data centers are, in reality, the main environmental challenge in this conversation. But I also believe we will see more regulation, more public pressure, and more innovation so that they do not harm nearby communities or the environment.
Just remember: data centers are not 100% for AI. They are the infrastructure that makes it possible for you to watch Netflix, shop on Amazon, open Facebook, send WhatsApp messages, store photos in the cloud, search on Google, and yes, also use artificial intelligence.
The discussion should not be “AI is bad because it uses water.” The discussion should be: how do we build more efficient digital infrastructure, with clean energy, better cooling systems, more transparency, and less local impact?
Because if we are going to criticize the environmental impact of AI, let us do it with data. Not with Instagram screenshots.
Calculations and Fact-Checking
- The Gemini numbers come from Google: 0.24 Wh, 0.03 gCO₂e, and 0.26 mL of water per median text prompt.
- The ChatGPT numbers come from Sam Altman: 0.34 Wh and 0.000085 gallons of water, equivalent to around 0.32 mL. Importantly, this is not a complete external audit of OpenAI; it is a public statement from Altman.
- The IEA estimates that one hour of streaming consumes 0.077 kWh, or 77 Wh, and that its central emissions estimate is 36 gCO₂ per hour.
- The indirect water calculation uses the factor from Argonne National Laboratory: 0.39 gallons of water consumed per kWh of electricity in the United States. That is where the ~114 mL per hour of streaming comes from.
- Google has a real enhanced geothermal energy project with Fervo in Nevada, already operating and supplying carbon-free energy to the grid that serves its data centers.