Since its launch at the end of 2022, ChatGPT fascinates as much as it worries due to its impressive results. But running OpenAI’s chatbot is a significant cost. This could reach $700,000 a day, according to Dylan Patel, chief analyst at SemiAnalysis, a consulting firm specializing in electronic components, interviewed by The Information media.
This cost is mainly attributable to the technological infrastructure used to operate this generative AI. ChatGPT requires enormous computing power on “expensive servers” to respond to user requests, whether to write a cover letter, plan a vacation, compose a song or write a poem.
Several million dollars a day
However, the running cost of ChatGPT could be even higher in reality, since the estimate was made on GPT-3. But the latest version unveiled in March, GPT-4, would be even more expensive to operate. Additionally, training the language models to improve the chatbot likely costs tens of millions of dollars, according to Dylan Patel.
Since the operational expenses are greater than the costs of training these language models, the bill quickly rises when generative AI is deployed at scale. This is the case of ChatGPT, which has nearly 200 million unique visitors per month, according to Similarweb data compiled for The echoes. Therefore, OpenAI’s chatbot would cost several million dollars a day to operate optimally.
Faced with the explosion of uses around generative AI, the challenge for players in the sector is therefore to limit the surge in their operating costs in order to stay in balance. In this sense, Microsoft is currently working on a chip dedicated to AI to reduce costs, with a dedicated team of more than 300 employees, according to The Information. Called Athena, it would be in development since 2019 and could be the subject of internal use by Microsoft and OpenAI from next year, according to two sources familiar with the matter cited by the same media. As a reminder, Microsoft has invested several billion dollars in the company behind ChatGPT.