OpenAI details Habitat, the storage platform behind ChatGPT's 22M requests per second
OpenAI published a technical post describing how Habitat grew from a Python library into a globally distributed storage platform serving 1 billion ChatGPT users.
OpenAI has published a technical write-up about the storage layer behind ChatGPT. At its centre is a system called Habitat, which started life as a Python library and was rebuilt into a globally distributed storage platform.
According to figures shared by the company, the platform now serves 1 billion ChatGPT users and handles 22 million requests per second. The post walks through how the team reached that scale.
Why it matters
Storage rarely gets the attention model quality does, but it is among the first components to become a bottleneck as a chat product's user base grows. OpenAI's path from a library to a distributed platform reads as a practical case study for teams on a similar growth curve.
The article is the first part of a multi-post series. For the architectural decisions, the problems encountered and the fixes applied, read the full text on OpenAI's official blog.
