Everything has its costs and artificial intelligence is no exception. While ChatGPT and Gemini are free to use, in order for millions of users to use them properly, a lot of energy and capital must be invested in the said technology.
According to Android Authority, OpenAI used about 1,300 megawatt-hours (MWh) of electricity to train and generate the GPT-3 large language model. This amount is equivalent to the annual electricity consumption of about 120 American families. By comparison, the average American household uses just over 10,000 kilowatt-hours of electricity each year.
All AI models require computing power to process each user query, which is called inference, and to achieve this goal, there are many powerful servers in thousands of data centers around the world. At the heart of these servers are usually thousands of Nvidia H100 GPUs, each consuming 700 watts of power.
Estimates vary widely; But most researchers believe that ChatGPT alone requires several hundred megawatt hours of electricity per day. This amount of electricity will provide the necessary energy for thousands and even tens of thousands of American households per year.
Some experts estimate that Nvidia will produce around 1.5 million AI server units by 2027. This results in artificial intelligence servers consuming 85.4 to 134 terawatt hours (one billion times one kilowatt hour) of electricity annually, which is more than the annual electricity consumption of small countries such as the Netherlands, Bangladesh, and Sweden.
To better understand the issue, it should be noted that the world’s electricity production was approximately 29,000 terawatt hours a few years ago; Therefore, according to experts in this field, the amount of electricity consumption required by large technology companies to produce and maintain artificial intelligence until 2027 is estimated to be about 0.5% of the world’s total electricity consumption.
