Environmental Impact of ChatGPT: Water Usage
I've recently been curious about the environmental impact of AI models like ChatGPT, specifically regarding their water consumption. How much water does ChatGPT…
Hunter Knight
March 14, 2026 at 08:33 PM
I've recently been curious about the environmental impact of AI models like ChatGPT, specifically regarding their water consumption. How much water does ChatGPT use during training and operation? Are there any estimates or studies about this? Understanding this could help us gauge sustainability and push for greener AI technologies.
Add a Comment
Comments (5)
I found an article estimating that training GPT-3 emitted around 552 tons of CO2 — not water directly, but suggests the energy and cooling needs are massive. Water use, while less talked about, is tied to that energy use.
Does anyone know if there are efforts to use alternative cooling methods that reduce water consumption in AI data centers?
OpenAI uses major cloud providers that aim to improve energy and water efficiency. Still, the scale of training GPT-4 must involve significant water for cooling these huge server farms.
I've read that training large AI models can consume megawatt-hours of electricity, and since data centers use water for cooling, indirectly, a lot of water is involved. But the exact liters per query or model isn’t clear.
Great question! While exact figures for ChatGPT aren’t publicly disclosed, studies on data centers show that water is often used for cooling servers. The amount depends on the data center's design and location.