ChatGPT consumes enough power in one year to charge over three million electric cars

The electricity consumption of artificial intelligence systems has skyrocketed. A new report revealed that the yearly amount of electricity necessary for ChatGPT, one of the largest so-called language models, could charge 95% of electric vehicles in the United States or power Finland or Belgium for one whole day.

Since its launch nearly two years ago, OpenAI’s ChatGPT has quickly captured the public’s attention, amassing an estimated 100 million monthly users within just two months. However, while its capabilities, from drafting essays to solving complex math problems and writing code, are impressive, they come with a hefty energy cost, BestBrokers wrote in a new report.

Each ChatGPT query consumes an estimated 2.9 Wh of electricity, nearly ten times more than a standard Google search.

Submitting a query to an AI model like ChatGPT is known as inference. Energy consumption depends on many variables such as query length, number of users, and how efficiently the model runs.

Each ChatGPT query costs 0.04 cents

The firm’s team roughly calculated the total annual electricity usage for generating responses alone. ChatGPT consumes 226.8 GWh each year to process 78 billion prompts. It is enough to fully charge about 3.13 million electric vehicles, each with an average battery capacity of 72.4 kWh, according to the presentation. That would be 95% of electric cars on US roads at the end of 2023.

In other terms, so much electricity, for handling requests, could also power 21,602 homes in the country for a full year or 0.02% of the total. The cost of the annual consumption amounts to USD 29.7 million or just under 0.04 cents per query.

Each ChatGPT query consumes an estimated 2.9 Wh of electricity, nearly ten times more than a standard Google search

The energy ChatGPT uses in a year to answer questions could also charge 47.9 million iPhones 15 every day for an entire year, each with a battery capacity of 13 Wh.

According to the calculations, the said large language model uses 621.4 MWh of electricity every day to handle over 200 million user queries. To put it in perspective, it consumes more than 21 thousand times the daily energy of an average US household.

ChatGPT’s yearly energy consumption for handling prompts could power all of Finland or Belgium for a day, or keep Ireland running for over two days, the authors said.

It required 62.3 GWh to train GPT-3

Training large language models or LLMs is a highly energy-intensive process as well. The AI model learns by analyzing large amounts of data and examples. It can take anywhere from a few minutes to several months. Throughout the training, the CPUs and GPUs, the electronic chips designed to process large datasets, run nonstop, consuming significant amounts of energy.

For instance, training OpenAI’s GPT-3, with its 175 billion parameters, took about 34 days and used approximately 1.3 GWh of electricity.

But as models evolve and become more complex, their energy demand increases as well. Training the GPT-4 model, with over a trillion parameters, consumed some 62.3 GWh of electricity over 100 days.