The global popularity of creating “Ghibli-style photos” using ChatGPT, a generative AI model, appears to be putting severe strain on OpenAI’s infrastructure. OpenAI has expressed concerns that if the overload continues, the launch of new services could be delayed.
Sam Altman, CEO of OpenAI, posted on X on the 1st, stating, “Focusing computing resources on ChatGPT image generation may delay the release of new features. Some functionalities could be paused, and services may slow down.” Earlier, on March 28, he had warned about infrastructure overload, saying, “People are loving ChatGPT, but our GPUs are melting.”
The surge in OpenAI’s infrastructure load stems from the release of ChatGPT-4o’s image generation model on March 25. Shortly after its launch, users worldwide began transforming their photos into Ghibli-style images. Beyond Ghibli, people are also converting their pictures into styles resembling Pixar, Disney, The Simpsons, Detective Conan, and Crayon Shin-chan.
The issue lies in the explosive growth of new users as the feature went viral globally. Altman reported on March 31 that “sign-ups increased by 1 million in just the last hour.” Usage has also skyrocketed in South Korea. According to data from the platform company IGAWorks, ChatGPT’s daily active users (DAU) hit a record high of 1,252,925 on March 27.
OpenAI plans to implement demand-control measures to address the overload while striving to maintain its new product development pace as much as possible. The company has introduced speed limits on certain features and restricted the number of daily image generations for free-tier users. Altman said, “We’re working as fast as possible to develop new products,” adding, “If anyone has 100,000 GPUs lying around, please reach out ASAP.”
