Each request at Chatgpt consumes 32ml of water
Openai boss Sam Altman recently published information on the Chatgpt energy requirements in a blog entry. According to it, each chattgpt request consumes about as much electricity as an oven requires use per second.
Altman gives numbers, but no details
In the official blog Sam Altman yesterday gave his company Openaai specific information on the water and electricity consumption of chatt use. However, Altman did not name any sources. Altman insisted that the “costs of intelligence” would soon approach the electricity price. “People are often curious how much energy a chattk query uses; the average query consumes about 0.34 watt hours, which corresponds approximately to the consumption of an oven in a second or the consumption of a highly efficient light bulb in a few minutes,” said Altman.
Openai recently spoke of 1 billion inquiries per day
Altman gave the water consumption with 0.000085 US gallons per average chatt request, which corresponds to the equivalent of around 32.2 milliliters. So far, he left open what exactly Altman sees as a “average” request. It is also currently unclear how much inquiries chatted daily. Most recently, Openaai spoke of one billion inquiries a day in December 2024. If you use this value, almost 322,000 liters of water are required every day to process users’ inquiries at Chatgpt.
As far as the electricity requirement is concerned, 340,000 kilowatt hours would be required every day. It should be borne in mind that the number of inquiries from Chatgpt may have increased significantly within the six months since the publication of the last official user numbers of Openaai. AI providers have been massively criticized for the enormous energy requirements of their services for some time.
The Washington Post Last year, together with researchers, had calculated that the creation of an email with 100 words would need “a little more than one bottle of water” using Chatgpt with GPT-4. In addition, water consumption depends on the location of the respective data center, it was said at the time.
