OpenAI scales storage system Habitat to 1 billion ChatGPT users

OpenAI says it has rebuilt the storage system behind ChatGPT, turning an internal tool called Habitat from a simple Python library into a globally distributed platform that now serves more than 1 billion users. The company describes the shift in a new post on its website, the first in a planned series about how it scaled storage to meet ChatGPT’s growth.

Storage, in this context, means the infrastructure that saves and retrieves data, such as conversation history and account information, every time someone uses ChatGPT. OpenAI says Habitat currently handles 22 million requests per second, a scale it attributes to years of re-engineering rather than a single redesign.

According to OpenAI, Habitat did not start out built for this kind of load. The company says it began as a Python library, a reusable piece of code written in the Python programming language, used internally to manage data access. As ChatGPT’s user base grew, OpenAI says that early setup could no longer keep pace, pushing engineers to turn Habitat into a distributed system, meaning one that spreads data and requests across many machines instead of relying on a single server.

Why the rebuild mattered

OpenAI frames the Habitat project as a response to demand it says arrived faster than its original infrastructure could handle. The company states that ChatGPT now has more than 1 billion users, though it does not define whether that figure refers to weekly, monthly, or all-time signups in the source material provided. OpenAI does not specify in this post what triggered the initial move away from the Python library, nor does it give a timeline for when each stage of Habitat’s evolution took place.

The company also does not detail, in the material published so far, what technical architecture replaced the original library, what database or storage engines Habitat relies on, or how the system distributes data across regions. OpenAI says this is the first part of a series, suggesting those technical specifics may follow in later posts.

What is confirmed is the scale OpenAI is now operating at: 22 million requests per second across a storage system supporting a ChatGPT user base OpenAI puts above 1 billion. The company presents this as evidence of infrastructure work that happened largely out of public view while ChatGPT’s usage climbed.

OpenAI has not said when subsequent parts of the series will publish or what they will cover. Readers looking for detail on Habitat’s current architecture, the engineering decisions behind specific scaling milestones, or comparisons to how the system performed at earlier points in ChatGPT’s growth will need to wait for those follow-up posts, since none of that appears in the material released so far.